95 min read

Generative AI Use Cases

Generative AI Use Cases
Photo by Startaê Team / Unsplash

With highlights for Local and Free/Cheap LLMs

About:

Goal :

Identifying Shinkai features opportunities with product market fit :

- get ideas of features to develop to allow relevant use cases

- reproduce use cases to check that developed features meet end users needs

Why We’re Sharing This Publicly: 
We originally put this information together for our own team’s reference, but it quickly became clear that these ideas might help anyone experimenting with generative AI. We’ve gathered real examples and discussions from various online communities—none of it was crafted by us alone. By sharing these user insights, we’re hoping more people can discover practical ways to leverage local or low-cost LLMs in their projects. We’ve also sprinkled in references to different tools, workflows, and solution approaches that we found especially useful during our own exploration.

Content:

Extensive list of practical and actual use cases of generative AI, both at enterprise and individual level.

Highlighting use cases where small local LLMs and free/cheap LLMs are usable and get good results (symbol ▶️).

Resources for AI workflows.

Starter list of tools for AI agents and workflows.

Additional resources.

Organization of this document:

By type of task : all examples listed within a given task should require similar tech stack / implementation / features.

For each task:

- description

- example(s) : with links for some, and often purposely written in full and redundant in order to transmit well the various intentions, needs, appreciations, etc. of the end users (feels more human and highlights the demand for AI solutions).

Sources:

personal usage, other AI apps/projects, AI newsletters, AI presentations, reports, blogs, X, reddit, quora, enterprises communications, ...

A note on small local LLMs and free/cheap provider LLMs :

- small local LLMs : =< 15b parameters

-> can be / will be able to run fast on accessible personal computers

- free/cheap LLM provider examples : groq llama 3.2, gemini 1.5 flash, gpt4o-mini,... 

-> can be used extensively due to high limits for free usage and/or low price

- are getting better and better rather quickly

-> see performances of medium size models rising quickly : llama 3.3 70b, qwen 2.5 coder 32b, qwq 32b)

Table of Contents

1)Text:

1.1) Simple prompt, querying the LLM “knowledge”

1.1.1) ▶️ Topic exploration

Description :

Using a simple prompt to ask a question to the LLM, using its “knowledge”, diving deeper into subjects (e.g. niche questions, safe place to ask, declassified information, fact checking).

Examples :

“Other non-professional areas discussed: leather shoe maintenance, fashion tips, beard grooming, house maintenance, renovation queries and organization.”

“Curious conversations about the nature of consciousness.”

“Esoteric questions.”

“Accessing declassified historical knowledge. Like the Satsop nuclear reactor being used for military training. Prompted for information on unethical Soviet brain experimentation.”

"My main usage lately has been better understanding what's going on in a movie and getting some tips about Stardew Valley."

"Super Google or super Wikipedia, looking up quick, easy to read answers on various things & subjects."

"An example is, I have ADHD, and I sometimes get super distracted if I remember a scene in a TV show but can't remember what show. You can feed it details of the scene and it can tell you exactly what episode of a series it's from. (Also works for movies)"

"Sometimes I'm doing something so horribly specific that there's either zero results on the web or one post on tomshardware from 2004 that went unanswered, and that's where ChatGPT (LLMs in general) help me out."

"I work in an industry regulatory environment and use it to get a head start on research for niche and specific rules/applications, etc. I can't assume it's correct, but it can lead me in the right direction."

Safe space to ask, where individuals can freely ask questions without fear of judgment or scrutiny.

"I have severe anxiety, so I had difficulties talking to anyone really, so I missed out on asking teachers to explain certain subjects that I didn't understand because I guess I was too scared. Googling stuff helps but it's a hassle and sometimes it's never explained simpler. So chatGPT has been filling up certain gaps in knowledge that I had without feeling judged and having the ability to tell it to explain it in simple terms helps a whole lot."

"Like an infinitely patient college professor/mentor who is always in their office without the ego, bullshit and outrageous cost."

"I am also writing a visual novel on Ukraine, and talking to it helps a lot with understanding the situation there. If I feel that it's making stuff up, I can just go in Quora and double-check."

Fact checking. But educated users also know LLMs can hallucinate."If it tries to write about a specific tax law thing, I always look it up unless I know it already.”

1.1.2 ▶️ Troubleshooting

Description :

Using a simple but specific prompt to ask a LLM how to fix a problem, issue, thing.

Examples :

“For common repairs, I'll directly ask ChatGPT instead of searching online. I can then ask very specific follow-up questions.”

“Oil leaks: Mechanics find source, fix leaks, considerations when changing thermostat.”

“Fixes needed: Humidifier component replacement, sprinkler control.”

"Often, when you try to google a problem you'll get an article that's 8 paragraphs of nothing and 1 sentence of telling you something you already know.”

“With ChatGPT, you can walk through the steps of what you have already done and what is next. For example, I tried doing X and that didn't work. ChatGPT, if X didn't resolve your issue then try Y and Z... and so on."

1.1.3) ▶️ Advices

Descriptions :

Using a simple but specific prompt to ask a LLM some advice on particular things (e.g. decision making, business, medical, relationships)

Examples :

"What else can I plant in my apple tree guild?"

"What line rigging works best for Brown trout in April?"

"I also ask a lot of "why" questions. Why architecture x over architecture  y?"

"I use it like a friend cuz I don't have any honesty. I ask it questions for advice mainly."

"I also treat ChatGPT as an advice buddy and ask it all kinds of questions about whether I should do grad school, move to California, or not."

"I use it for guidance to better decisions, and better communication. I'm talented and know my field, what I don't know is the best way to communicate that knowledge to senior decision makers who don't have my knowledge but still need to make a decision."

"Yesterday I made the chat give me three questions with three options that could give some information about my personality. Based on the result I gave, I asked which country can offer me a vacation according to my personality. I shall go to Italy now..."

"Life Coach. Help with procrastination, task prioritization, encouragement, financial advice, career advice."

Medical advice.

"We had an employee suffer a seizure at lunch in the lunchroom and hit his head. While people were attending him I asked ChatGPT what we should do. It came back with 8-10 things, one of the things it stated was clearing a path for paramedics. So myself and others moved all the lunch tables to the outskirts of the room. When EMT came they were able to easily maneuver and get the person to the hospital. Again, since others were attending the individual I had time to do this. I thought it was very helpful."

Business advice (e.g. improve operations, negotiate a deal, increase profitability, achieve goals, etc.).

"It's a great sounding board. I ask it questions and lay out plans, it basically gives me things I might have forgotten and reorganizes things in outlines or process flows. Using diagram plugins I can map steps and chart what I want to do."

"Literally helped me close 5 figure deals."

“I am in the middle of negotiating with an insurance company for diminished value. LSS: truck was backed into, door replaced, now to collect diminished value. I asked ChatGPT what the value would be on my specific vehicle, trike level, etc. it gave me back some very valuable info. Insurance company offered me 3.5% of value. I responded with CGTP info, they have now offered me 10% and valued my vehicle by 8k more than the initial estimate. Helped me a ton!!”

"I had it help me through updating and re-writing my resume, never a bad thing to have, and it did a really great job, as well as gave me some ideas on what may be a good next professional move."

1.1.4) ▶️ Recommendations

Description :

Using a simple but specific prompt to ask a LLM for some recommendations (e.g. entertainments, holidays, etc.)

Examples :

Create holiday itinerary, offer personalized recommendations based on preferences, budget, and time constraints.

"Plan an itinerary for a x day trip to y (e.g. 7 day trip to London), which is actually feasible and includes a lot of well known sites."

"I spent 30 minutes fine tuning a prompt and then loaded it up to Chat-GPT so it could act as my personal professional holiday trip advisor and search all the necessary information I needed to spend a week in Amsterdam and Berlin. Furthermore, I asked information on Amsterdam, which is where I'll soon be living. What should've been 30 hours of work ended up being 4 hours with maybe four more to go."

"I'm using it right now to help me find new podcasts to listen to. Asking me questions about my preferences and then generating lists of podcasts. Pretty darn neat!"

Remy : The chatbot that's seen every video on YouTube.

https://www.useremy.com/chat“build you custom playlists on any topic, finding exactly what you want to watch - and nothing you don't.”

1.1.5) ▶️ Homework

Description :

Prompting the LLM to suggest resources, provide explanations, generate answers, provide feedback, in the context of homework, university assignments, etc.

Examples :

"The kids ask me for help with their homework, and I get instant writer's block. ChatGPT can help there."

"The school bans the use of AI, but it's a fight they can't win. It's like banning google in the 2010s or banning a calculator in the 1990s. It's unenforceable, and it doesn't even prepare kids for the actual future."

"If I ever hear a word or colloquialism that I never heard of before, chatgpt knows the answer."

1.2) Modifying the language of a text

1.2.1) ▶️ Translation

Description :

Translating from a language to another / others (e.g. helping with fast localization of content, improving accessibility to information in various languages, allowing to read in a given language without translator distortion, easing global collaborations, video games chats translation).

Example(s) :

"got it to translate and interpret a quarterly report from a Japanese company that I'm interested investing in but only reports in Japanese"

"I did English to Portuguese in some scientific papers and it worked fine, it just missed a gender once. It's interesting how it changes even for similar languages."

"I've tried and it seems pretty great for more popular languages (Spanish, French and German). I was really impressed by the answers it gave (I spoke with it about a documentary I watched and it gave me very compelling answers and corrected my writing). It can also write phrases and stories for certain CEFR levels. Certainly take it with a grain of salt, but it's good for certain languages."

"With GPT4 I can finally explore Vedic texts in Sanskrit without the western and religious framework distorting translation. I can map philosophical concepts on to other paradigms. It's just stunning."

"I am trying to keep my written Norwegian from getting too rusty so sometimes I've asked chatgpt to help me with that."

Airbnb uses LLMs to translate their platform content, including listings, descriptions, and reviews, for travelers worldwide.

Google is working on a universal speech model trained on over 400 languages, with the goal of building a universal translator.

Language Learning: Generating example sentences, stories, and dialogues for language learning, incorporating specific vocabulary, translations, and roleplaying scenarios.

Duolingo Max, powered by GPT-4, provides features like "Explain My Answer" for detailed explanations and "Roleplay" for conversational practice.

A video game developer, Roblox, leverages a custom multilingual model to enable users from all over the world to communicate seamlessly using their own tongue. The model supports chat translations between any combination of 16 languages, and the translations are displayed in real time.

Translate military briefing in international operations.

An AI Worker translates and enriches customer messages in real time for the support team at a multinational company.

translateProductListing() : Build hooks into your CMS to automatically translate and localize product descriptions.

Multilingual Support (MLS): Deliver support in over 100 languages to cater to a diverse workforce.

1.2.2) ▶️ Correction

Description :

Asking a LLM to correct the grammatical, spelling, conjugation mistakes in a text.

Examples :

“At work, I use ChatGPT as a grammar and spelling checker, especially for lengthy emails and insurance claim disputes. It makes me more effective than my colleagues.”

When there is a large typo in a sentence that makes you ensure what the intended word could be, ask a LLM to list possibilities and why.

1.2.3) Language type

1.2.3.a) ▶️Formal <-> non-formal

Description :

Using a LLM to change the formal / professional / objective tone of a text (e.g. work emails, cover letters, resumes, contracts, etc).

Example(s) :

"I asked it to rewrite my resume. Now my resume is 5 stars. I don't see how any employer wouldn't want to at least interview me."

"Usually I am worn down from work on Fridays. If somebody sends me the wrong email I used to be really pissed of about the situation and write the most aggressive response. Now I still write the email as harsh and as insulting as I possibly can. I then copy paste it into ChatGPT and ask it to write it as nicely and polite as it possibly can. I then send a slightly edited version of what ChatGPT suggests and everyone has a better weekend."

"Mostly emails "Write an email to Dan and let him know I think his plan is absolutely idiotic and huge waste of resources...but make it professional"

"I remember having to write an email to a group I'm participating in, about a super irritating situation and I was pissed. And it was so obvious in the email, including swear words. I handed that email to ChatGPT and said, "Make this sound professional and polite." And it did. I loved it. I was able to say what I wanted to say exactly how I wanted to say it, and ChatGPT just cleaned it up for me."

GitLab : Streamlining hiring and recruitment processes: Used to refine email templates for candidate communication.

"I'm doing child advocacy work. I'm frequently writing to organizations who support children. A lot of my emails can come off as pissy. I drop each of these emails in gpt and ask it to make me sound less like an asshole. That's my actual prompt. I also ask it sometimes to speak with more legal authority and add some legalese. It's been very effective."

"I do the same with my texts to my ex-wife!" 

"cleaning up my emails. I will copy an email and say "Make this more succinct, take out all emotional content" and it works beautifully."

"I use it to rewrite angry emails in an objective tone, or to write all-staff emails in a way I won't embarrass myself."

https://github.com/theJayTea/WritingTools

Instantly proofread and optimize your writing system-wide with AI:

With one hotkey press system-wide, it lets you fix grammar, optimize text according to your instructions, summarize content (webpages, YouTube videos, etc.), and more.

1.2.3.b) ▶️ Technical <-> less technical

Description :

Asking a LLM to rewrite the same content, but with a different level of technical language, for easier understanding (e.g. medical translation, engineering, hard science).

Examples :

“In a unique twist on a traditional translation model, health tech company Vital launched an AI-powered doctor-to-patient translator to instantly turn highly technical medical terminology into plain language.”

"It's also way better at explaining concepts to non-engineers than us engineers are. By default it writes at a 5th-grade level, which is perfect for a lot of people we interact with at work."

"In the statistics class I'm taking, many of my textbook's explanations are very technical and difficult to understand when describing certain concepts, so I just copy/paste the text (PDF textbook) into ChatGPT and ask it to explain the concept to me in a different way (or in a more conversational manner). Then, I usually ask follow-up questions about specific elements of said text."

1.2.3.c) ▶️ Specific tones

Description :

Using a LLM to rewrite a content with a different touch (e.g. add humor, make it epic, more passionate, etc.).

Examples :

“It's good for writing cover letters for job applications. The AI can speak much more passionately about my desire to work at Starbucks than I can.”

1.2.3.d) ▶️ Writing styles

Description :

Using a LLM to rewrite content in the style of an author, of a literature genre, of a brand voice.

Examples :

“AI copywriting assistants, with enough training, can also approximate brand voice more consistently than humans. That not only increases output without impacting the quality but also frees up copywriters to play a more editorial/curatorial role to better support marketing efforts.”

1.2.4.e)▶️  More condense

Description :

Using a simple prompt to ask a LLM to edit a text for better clarity.

Examples :

“I also use ChatGPT to clean up and condense my own writing. Its suggestions have been very helpful.”

“For my writing, I have specific prompts to trim unnecessary details while conveying key information concisely. This assists with my ADHD-related tendency to include extraneous details.”

“I've asked ChatGPT to critique my writing samples and provide feedback to improve them. It's been a valuable editing tool.”

"I fed it a long, overly complex service level agreement for a SaaS contract and asked it to rewrite it to make it simpler and easier to digest. It kept the important SLA terms but condensed the language by 70%. Much easier to read."

1.3) ▶️ Categorizing text, tagging

Description :

Using a LLM to categorize text content according to specific instructions (e.g. sentiment analysis, customer feedback, social media moderation, sensitive information detection, product tags, spam detection).

Examples :

Sentiment and trend analysis with output closely matching user instructions. (social media, online forums, reviews).

Brand reputation analysis.

Customer reviews categorization.

Talkwalker combines customer data with social intelligence for product feedback analysis.

News feed categorization.

Document classification (organize documents by topic, sentiment, or other attributes).

Example : JPMorganChase leverages LLMs to categorize various documents, including loan applications, financial statements, and customer correspondence.

Data-driven recommendations to customers (based on categorised tags, reviews, feedback on previous sales, usage, etc).

Content moderation : detect and delete content falling in specified categories.

Telegram bot for monitoring group chats for rule-breaking behavior.

Analyzing customer service requests and classifying them based on emotion or urgency using RoBERTa-base-go_emotions (a model based on BERT).

Using BART (an encoder-decoder model) for zero-shot classification, which enables users to prompt a model with custom categories.

AI-powered writing assistant Grammarly uses LLMs for its tone detector, categorizing text into labels like joyful, optimistic, or formal to assess message perception.

Grammarly also uses LLMs to protect users from harmful conversations. The company goes beyond toxicity and detects delicate text—text that is emotionally charged or potentially triggering and poses a risk for users or LLM agents, such that engaging with it can result in harm.

With Nebuly, TextYess is able to scale its user base and business while having a consistent overview of the quality of its shopping assistant user interactions.

Nebuly's user feedback feature automatically flags each negative and positive (not neutral) LLM interaction for TextYess. The insights allowed them to end manual output reading, pinpoint areas of improvement and continuously improve their shopping assistant's user experience.

90% less time spent ‍analyzing LLM user conversations.

8x the number of identified,actionable insights from conversations.

63% decrease in negative interactions after one month of use.

Whatnot: Employs LLMs for enhanced content moderation, fraud protection, and managing bidding irregularities.

Zillow uses LLMs to detect proxies for race and other historical inequalities in real estate listings on their marketplace.

Grab tags sensitive data.

Asian super-app Grab uses LLMs in data governance: to classify data entities, identify sensitive information, and assign appropriate tags to each entity.

GoDaddy utilizes LLMs to improve customer experience by classifying support inquiries in their messaging channels.

Food delivery company DoorDash uses LLMs to identify and tag product attributes from raw merchant data. This helps improve the matching of customer queries with relevant items and aids delivery drivers in locating products.

Yelp, an online reviews platform, upgraded its content moderation system with LLMs to detect threats, harassment, lewdness, personal attacks, or hate speech.

generateTagCloudFromComments()

Take a list of comments and generate a tag cloud for users.

rankReviewsByQuality()

Sort a list of product reviews by quality and helpfulness.

isCommentSpam()

Flag if a comment seems like it might be spam or off-topic.

Smart Handoff : Route users effectively to the correct destination for further assistance based on support message categorization.

Segmenting target audiences within an industry and geography by behavior, demographics, and interests to better understand the total addressable market and guide personalized messaging.

1.4) Reshaping text

1.4.1) ▶️ Summarizing text

Description :

Using a LLM to summarize text content (e.g. academic articles, contract reviews, social media posts, colleagues syncing, transcripts).

Examples :

Legal contract reviews.

"I'm a lawyer. I use it to make my emails sound more professional and succinct. I also use it for research. It will direct me to the exact statute or case I need, as well as summarize it"

Business contract reviews.

Summarizing conversion between a tenant and a property manager in property rentals context.

Using led-large-book-summary (a model based on BART) to summarize academic articles.

Salesforce introduced AI Summarist, a conversational AI-powered tool, which summarizes Slack conversations and helps users manage their information consumption based on their work preferences.

"Specifically, for instance, she will first feed ChatGPT the Job description of the one she's playing, and ask it to summarize the key elements the company is looking for, then summarize, then on its conclusion, ask resume adjustment suggestions."

Analyzing, summarizing, and extracting insights from documents like financial annual reports, insurance contracts.

"As a college student pursuing B.E. in Computer Science, I use it between lectures to get quick summaries on topics where I have trouble matching the pace of my professor. Often it ends up explaining better than my professor because it uses real life examples when asked to do so. $20 is nothing for getting a 24x7  personalized tutor."

1.4.2) ▶️ Organizing text

Description :

Rewriting text content in a more organized way (notes, early drafts).

Examples :

Organizing messy notes with AI's help for efficiency and clarity.

"I keep an open Google Doc and dictate my follow up notes there, then drop that into GPT to clean the notes up, and have the prompt pre-set to revise commonly misheard phrases in my dictation. Then I have it use the cleaned up version of my note to create a follow up email draft."

Organize transcripts.

Merging texts.

"I went from a C student in my bachelor's program to A student in my graduate program because it helped me organize my writing."

https://www.granola.ai/

The AI notepad for people in back-to-back meetings.

Granola takes your raw meeting notes and makes them awesome.

1.4.3) Extracting text

1.4.3.a) ▶️ Documents / forms screening
Description :

Extracting specific information from the text in a document, a form, etc. (e.g. invoices, hiring, customer profiling, identity).

Examples :

Automating basic accounting functions like invoice capture and processing.

Mercer (HR company): Uses LLMs to automate the recruitment process, including screening resumes and identifying the best candidates for jobs, saving time and money.

identity resolution.

Customer profiling : leveraging massive amounts of first-party consumer data to build out detailed customer profiles.

Walmart developed a Product Attribute Extraction (PAE) engine to help retailers onboard new items and extract attributes from an existing catalog. The solution gets text and images from PDF files, extracts relevant product attributes using LLMs, and consolidates attributes into categories. It enables retailers to find a product that matches the existing inventory and plan for future assortment.

Extract specific information from legal contracts.

Extracting keywords.

Data Extraction: Extract qualifications from a resume into a JSON object, and map with your app's fields.

https://github.com/VikParuchuri/marker

Marker converts PDFs to markdown, JSON, and HTML quickly and accurately, optionally using an LLM to improve quality.

1.4.3.b) Retrieving precise information from multiple sources

Description :

Retrieving information from a potentially large amount of sources.

Examples :

Finding relevant information within vast amounts of historical data and across various document types, uncover critical information quickly.

Retrieving information from entire folders.

Legal : retrieve relevant laws from multiple legal texts.

Search in accounting documents.

Retrieve information from reports, emails exchanges.

Military : Extracting information from operation reports.

Every morning, an AI Worker automatically collects court cases for publication by a newspaper.

1.4.3.c) Finding document(s) with given information

Description :

Using a LLM to find document(s) with specified information, for better results than word search.

Examples :

Exploring a large knowledge base.

1.4.4) ▶️ Changing text format

Description :

Convert a text to another form of text.

Examples :

Text -> poem

Article -> social media post

A marketing team uses an AI Worker to turn long-form content into bite-sized pieces for social media and email campaigns.

“‘Hey, here's a whitepaper, can you draft me a blog, a LinkedIn post, an Instagram post based off of it?’, and it will do that. And then you can go in and edit and refine it and then post it.”

1.5) Using information from text

1.5.1) Using information from 1 text source

1.5.1.a) ▶️ Analyzing claims

Description :

Using a LLM to evaluate the claims of a text, producing arguments analysis, critiques, counterarguments, alternative viewpoints, identifying gaps, spot logical fallacies, etc.

Examples :

“I use ChatGPT to check my own biases and potential misinformation in my writing and political views. It helps me identify logical fallacies.”

Arguments used ?

Quotes ? Backing ?

Fact checking (but careful : hallucinations).

See blind spots : Identifying and addressing potential gaps or overlooked aspects in a project or plan.

"for me I get the most utility from it by helping me point out blind spots in my work, suggest improvements to operations, and offer insights for projects I'm working on."

"A specific convo l frequently use is called Logical Fallacies & Bad Arguments. I paste in the idiotic arguments that redditors put forth and let this thing call out the strawman arguments and such. It's good practice to spot insane arguments and end them early on."

Evaluating student copies (both as professor and student to improve its own work).

Reconcile personal disputes. “Generative AI can help by providing unbiased perspectives, suggesting mediation techniques, analyzing communication patterns, proposing compromises, and facilitating constructive conversations.”

“Asking what to say to angry gf.”

“Paste in text message conversations when my wife is on the warpath to get analysis and guidance.”

“It gives a bullet point analysis of me, and my wife, and then a bullet point analysis of the issue at hand and how we might reconcile.”

“I ask it to approach the convo as if it was an fbi behavioral profiler.”

“I was having a debate with a friend about a topic, and we both had assumed there were only really 2 sides of the topic and chatgpt helped introduce us to a 3rd perspective we hadn't considered, and both ended up agreeing with.”

"Steelman the following viewpoint <viewpoint with which I disagree>"

"Yeah, after I have it generate ideas, I like to tell it to pretend to be someone and critique the idea. I'll be like, "pretend to be Charlie Munger, and tell me why this idea is not so good." Have actually gotten really good insights from that."

1.5.1.b) ▶️ Scoring

Description :

Using a LLM to score some information from a text according to predefined criteria.

Examples :

scoring CVs/applicants for hiring.

The entertainment company Jumpcut uses LLMs for script analysis. Benefits for Businesses: Saves time and resources, enabling studios and production companies to find high-quality scripts faster. Provides comprehensive script coverage, capturing nuances and subtext, enhancing the decision-making process for creative projects.

“A specific convo I frequently use is called Fact Checker. I paste in something I'm reading and it evaluates it, rates it from completely inaccurate to completely accurate, and breaks down each idea for me into bullet points.”

Analyzing data from IoT devices.

Analyzing online profiles, social media, and professional networks to identify potential candidates.

Scoring standard documents like wills and contracts.

"Could you assess a blog post based on its writing quality, uniqueness, and reader impact, rating each aspect on a scale of 1 to 10? Also, please pinpoint any typos and offer constructive feedback that acknowledges the post's strong points and areas for enhancement, ensuring the evaluation is candid and avoids perpetually average ratings."

"Grade my work before I turn it in. I was able to provide my essay and the rubric used and it graded it and gave me suggestions."

"To create a tenant scoring model." 

1.4.1.c) ▶️ Explaining / Rephrasing

Description :

Using a LLM to “assimilate” the content of a document, and explain it to help comprehension : rephrasing, clarifying, commenting, etc. (e.g. technical content, paragraphs in books, song lyrics, legalese).

Examples :

"explain technical documentation. For example, I didn't know what this meant. 8K Refresh Cycles/64 mS, @ -40°C ≤ TA ≤ 85°C"

"Reading dense and/or highly referential books. Very helpful to be able to ask 'what did (author) mean by (phrase) in (book title)?'"

"I was reading Robert Green's "Mastery" the other night and stumbled upon a complicated metaphor that I couldn't quite understand fully. I fed it into chatgpt and asked it to break down each statement in the paragraph and also comment on if it was a true statement and gpt explained, agreed with most of it and then pointed out two things it disagreed with which I actually found useful."

"I blew my mind taking a picture of an old philosophy page I'm reading and asked it to tell me what it says in today's terminology and lingo and it did it. Then I switched it to voice mode and asked it a bunch of questions about the guy and who he was and what kind of socioeconomic conditions he grew up in."

Explain legalese.

"I wanted some clarifying questions around legal terms when forming a LLC recently and it gave adequate answers. Sometimes there were some acronyms that I couldn't find via google conventional means and ChatGPT led me to the right path. Where I could verify the information."

"I've asked it legal questions about taxes and property rights and it seems pretty good at that. Legal language, in an attempt to be reliable, has had to contort into some forms which can feel unnatural to those of us unaccustomed to it, but it can be fairly logical and consistent, and ChatGPT can parse that well and then translate it into forms which we can understand more easily. It can also tirelessly rephrase and digest things for you, even providing examples for a concept. I find this really useful."

Interpret song lyrics.

“Interpreting song lyrics involves analyzing and understanding the meaning behind the words in a song. Generative AI can aid in this task by using natural language processing to break down complex lyrics and provide insights on their themes, emotions, and symbolism.”

"I just used mine to interrogate the blood test results I just got from the doctors."

"I don't understand why x leads to y can you explain further. "

1.4.1.d) ▶️ Editing a doc

Description :

Feeding a text content to a LLM, and asking it to perform defined edits, such as replacing specific parts, ensuring consistency.

Examples :

"We have an existing signed software agreement, (agreement number), dated (date), with an attached Schedule A. We need to replace the Schedule A with a completely new version, canceling the old Schedule A as of a new effective date. We are going to create an amendment signed by both parties to execute this Schedule A replacement. Please draft the appropriate language for this new amendment such that it replaces the old amendment."

"I've used it to completely rewrite my employee handbook one section at a time."

Grant Writing: A custom grant-writing assistant was created for a Chicago food bank, using retrieval-augmented generation to improve grant applications.

1.4.2) Using information from multiple text sources

1.4.2.a) ▶️ Asking questions on multiple documents

Description :

Feeding a LLM with multiple documents and asking questions whose answers must use all the documents (e.g. diagnostic over many medical reports, company analysis over various sources of information).

Examples :

"I also fed it multiple reports from a family member's cancer diagnostics, i.e., mammograms, CT and PET scans, MRIs, biopsies, bloodwork, etc.. asking it at each input to deepen its analysis."

Greenlite is initially focused specifically on Anti-Money Laundering and Know Your Customer (AML/KYC) operations. Greenlite’s agents can, for instance, automatically carry out routine investigations on companies by reviewing documents and searching the internet.

“The status quo is often to rely on offshore contract workers to complete these tasks. So using Greenlite means swapping out an outsourced worker sitting in a different country with our AI. And our AI brings tremendous advantages—in terms of cost, speed, accuracy and transparency.”

1.4.2.b) ▶️ Analyzing documents in regard to others

Description :

Asking a LLM questions about document(s) whose answers must take into account other separate document(s) (e.g. fiscal compliance, environmental compliance, urbanism compliance, fulfilling a contract).

Examples :

Analysing documentation about a project in regard to compliance documentation (laws, standards, best practices, contracts, etc.)

Detecting regulatory breaches, monitor compliance with laws, and generate reports on compliance status.

Automated bot to communicate with parking ticket holders, and advice them on local parking regulations in each of the 60 UK regions.

Legal :

Automating regulatory monitoring to ensure clients are up-to-date with compliance.

Assisting in legal research by identifying, analyzing, and summarizing pertinent information from case law, statutes, journals, regulations, and other relevant publications.

Assisting with due diligence by reviewing large volumes of documents to identify potential risks and issues.

Norm AI’s agentic system can review a company’s operations on an ongoing basis, identify when a certain activity is not in compliance with a certain regulation, and suggest remedial actions to ensure compliance.

“Given the length and complexity of these laws, the ability to automatically analyze and apply them is compelling.”

LLMs analyze privacy policies and regulations to identify potential gaps and ensure compliance with laws like GDPR and CCPA.

1.4.2.c) Analyzing documents in regard to a large knowledge base

Description :

Asking a LLM questions about document(s) whose answers must take into account not just a few separate document(s), but an entire large knowledge base.

Examples :

Alexia (AI Legal Expertise X-Intelligence Assistant): An AI-powered legal assistant tool designed to provide legal advice by leveraging a database of law textbooks, case studies, and expert legal opinions. It uses Retrieval-Augmentation-Generation (RAG) to tailor information to specific branches of law and jurisdictional nuances. The aim is to offer high-quality legal advice at a lower cost and increase accessibility. It is not intended to replace human lawyers, provide advice outside the legal domain, or guarantee outcomes in legal disputes.

1.4.3.d) ▶️ Task prioritization based on many documents
Description :

Give the LLM many documents, and it outputs prioritized lists, for example of tasks, requests, etc.

Examples :

“It collects all my user feedback daily (survey data, support tickets, call transcripts, etc.) and turns it into a prioritized list of feature requests and complaints based on mention frequency.

I can reprioritize as needed and have it turn requests into PRDs and user stories, which get assigned to the team.”

“AI-powered sales intelligence engines, adds Vlad Voskresensky, CEO of guided selling platform Revenue Grid, act as a "sales co-pilot" by augmenting the selling process, providing data-backed recommendations for the next best steps, and real-time, contextual, personalized notifications to guide them through the most winning path at the most pivotal moments.”

1.4.4.d) Querying a knowledge base

Description :

Using human language to query an entire knowledge base (e.g. find processes, respect brand identity, find records, search documentation, explore a fantasy universe).

Examples :

Interactive knowledge base.

Conversational finance chatbots that can use internal documents as a knowledge base.

In September 2023, Morgan Stanley launched an AI-powered assistant to support financial advisors by providing easy access to its internal database of research reports and documents. Employees can use the tool to ask questions about markets, internal processes, and recommendations.

Fantasy Universe History Q&A: Training a LLM on a fantasy setting wiki to answer historical questions about the fantasy universe.

“I’ve seen examples of industries with huge amounts of documentation that want to enable their internal teams to retrieve answers out of tens of thousands of pages of records,

That’s the right approach, because the risk is low — it allows you to get your hands dirty, provides a lot of value, and doesn’t create a lot of risk. At Databricks, we have an internal chatbot that helps employees figure things out and look at their data. And we see a lot of value there.”

Domain-Specific Corpus Training: Training smaller LLMs (7B and less) on a single subject to achieve high expertise in that specific topic.

Searching for cases on a specific legal issue.

Legal argument analysis and weakness identification. Analyzing precedent cases.

https://thedrive.ai/

Store class materials, ask questions about them and write content from them.

https://www.glean.com/

“The Work AI platform connected to all your data. Find, create, and automate anything.”

https://vectorize.io/

Fast, Accurate, Production-Ready RAG Pipelines.

“Turn your unstructured data into perfectly optimized vector search indexes, purpose-built for retrieval augmented generation.”

Question Answering Systems systems such as chatbots and virtual assistants can provide immediate conversational access to your organization's knowledge to workers throughout your company.

Perplexity for Internal Search: one tool to search over both the web and your team's files with multi-step reasoning and code execution.

https://x.com/AravSrinivas/status/1846954158156583224

Therapy :

Decision aid for existing evidence-based practices (EBPs): LLMs can analyze large datasets of psychotherapy transcripts (including different therapeutic orientations, outcome measures, and sociodemographic information) to identify therapeutic behaviors and techniques associated with positive outcomes. This can help therapists make better data-driven decisions.

Specific search :

“involves looking for a particular piece of information or content within a large dataset or database. Generative AI can help by quickly generating relevant search results based on user input or preferences, making the search process more efficient and effective.”

Lookups : Provide answers to workplace-related questions using relevant information from other systems.

Search : Answer workforce inquiries with relevant internal or external knowledge articles directly in chat.

Knowledge Studio : Utilize generative AI to assist users in writing knowledge articles based on company data.

Brand design co-pilot: Helps creative teams uncover brand insights for art and copy by locating client brand documentation and historical creative files, identifying key brand elements.

Query / talk with the documentation of project / code source / protocol

Examples :  Myshell widget recommendation AI tool, Gumloop AI answer in documentation search.

1.4.4.e) ▶️ Matching

Description :

Using a LLM to automate matching between 2 different sources of data (e.g. personalized entertainment recommendations and product promotions, matching job ads and candidates, message relevance).

Examples :

Zara: “Uses LLMs to generate personalized fashion recommendations for its users based on data like past purchases, social media activity, and search history, improving recommendation accuracy and customer satisfaction.”

“Delivery Hero, a multinational online food ordering and food delivery company, solves the product-matching problem using LLMs. Identifying products in their inventory that are similar to those offered by competitors allows the company to develop more informed pricing strategies and better understand the differences in product variety on the market.”

“Food delivery company DoorDash uses LLMs to identify and tag product attributes from raw merchant data. This helps improve the matching of customer queries with relevant items and aids delivery drivers in locating products.”

Personalized Shopping Assistance, Klarna (E-commerce) offers “personalized shopping recommendations and streamlines the purchase process for a more tailored shopping experience.”

Netflix: Uses LLMs to generate personalized content recommendations for its users.

Spotify: Uses LLMs to recommend music to its users.

Personalized Promotions: Orange (Telecommunications) delivers tailored promotional messages based on individual customer interests and needs.

Personalized medicine based on individual factors. Example: Recommending drugs based on individual risk factors.

“Online marketplace OLX created Prosus AI Assistant, a generative AI model, to identify job roles in ads, ensuring better alignment between job seekers and relevant listings.”

French marketplace Leboncoin uses LLMs to improve search relevance by sorting ads in the optimal order regarding a user’s query and preferences.

The Amazon Store uses LLMs to discern commonsense relationships and provide product recommendations most relevant to customers’ queries.

Picnic online supermarket app uses LLMs to refine product and recipe search results for users in three countries, each with distinct languages and culinary tastes.

Linkedin analyzes various content across the platform to extract skill information. Skill data then goes to LinkedIn’s Skills Graph, which dynamically maps the relationships between skills, people, and companies, ensuring relevant job and learning matches.

targetUsers() : Given a list of users and data, pick which users are most relevant for a given message.

Using BART-large-mnli to retrieve relevant tags from a search prompt and match it with articles with the same tags.

Using BART-large-mnli to sort through feedback forms and determine which team is best for answering certain questions

1.4.4.f) Combining information

Description :

Using a LLM to create a specified combined content based on several sources of content (e.g. reports, profiles,...)

Examples :

“Generating Financial Crime Reports (Reason for Suspicion):  SumUp uses LLMs to help their risk and compliance agents write reports about suspicious account activity. These reports, used in Anti-Money Laundering (AML) processes, summarize investigation findings and provide evidence of suspicious behavior. The LLM uses information from various sources, including previous machine learning model predictions, investigation notes, and verification processes, to generate the report narrative. This automates a repetitive and time-consuming task, allowing agents to focus on investigation and verification. Human review and confirmation remain a critical part of the process before any alert is raised.”

“SumUp also developed a LLM-driven evaluator to assess the quality of the LLM-generated financial crime reports.This evaluator uses a custom set of benchmark checks within a prompt to compare the generated text against a reference text or a set of criteria.  These checks include elements like topic coverage, inclusion of customer profile data, supporting facts, avoidance of invented facts, text structure, and conclusion.  The evaluator provides both a quantitative score (0-5) and qualitative explanations for each check, offering insights into the strengths and weaknesses of the generated narrative. This automated evaluation method helps SumUp iterate and improve their LLM application without adding extra workload to their agents.”

A large SaaS company uses an AI Worker to build buyer personas based on customer interviews.

“I am learning something from various sources. I end up with documents (docs, tutorials, webpages, etc) talking about the same things, processes, tools, methods, etc. I ask an AI to combine these documents into a single document that presents the information from the various input documents, dropping duplicates, and keeping variations. It outputs similar format documents to the input (text, text + images, pdf, slides, tables, etc.), or whatever format I ask.”

1.4.4.g) ▶️ Comparing several documents

Description :

Using a LLM to compare 2 or more documents (e.g. contracts comparison).

Examples :

"Comparing contract versions. It has successfully found and highlighted sneaky changes to what otherwise looks like the exact same document."

"I copied two 60 pages contracts (all names/addresses, etc removed) into Chat GPT and named them contract 1 and 2. I asked Chat GPT to abstract the deal points in bullet format and was able to compare the contracts and make a negotiation matrix in 30 minutes. On review, Chat GPT only made one mistake on a payment term."

1.4.4.h) ▶️ Follow up with questions

Description :

Feeding a LLM with documents and use it to generate targeted questions.

Examples :

Digits uses generative models to help accountants by suggesting transaction-related questions to ask a client. The questions can be sent right away or edited before sending.

1.5) Conversational (text based)

1.5.1) ▶️ Immediate chat responses + categorizing and redirecting

Description :

Using a LLM in a chat interface to answer queries, and redirect where needed (e.g. customer support)

Examples :

Customer support (process incoming tickets, categorize the issue, initial response, redirecting) 24h/7d.

Delta Airlines chatbot -> 20 % reduction in call center volume

“Oracle’s Fusion Cloud CX uses an LLM that references internal data to help agents generate instant responses to service requests based on the history of the client’s interactions, and suggests new knowledge base content in response to emerging service issues.”

“Vimeo engineers used generative AI to build a help desk chat prototype. The tool indexes the company’s Zendesk-hosted help articles in a vector store and connects that store to the LLM provider. When a customer has an unsuccessful conversation with the existing front-end chatbot, the transcript is sent to the LLM for further help. The LLM would rephrase the problem into a single question, query the vector store for articles with related content, and receive the resulting relevant documents. Then, the LLM would generate a final, summarized answer for the customer.”

“Fintech unicorn Klarna has deployed an AI assistant powered by OpenAI to automate its customer service engagements. According to the company, this AI assistant has been able to handle two-thirds of all customer service requests (2.3 million conversations in its first month alone), automating the work of 700 full-time human reps and driving an estimated $40 million in added profit for the company this year.”

“E-commerce company Wayfair developed Agent Co-pilot, a Gen-AI assistant for digital sales agents. Co-pilot provides live, contextually relevant chat response recommendations that sales agents can use while chatting with customers.”

Guided buying process that is conversation based.

Online food ordering and delivery company Swiggy implements AI-powered neural search. This helps users discover food and groceries in a conversational manner and receive custom recommendations.

Doordash, a food delivery company, enhances delivery support with an LLM-based chatbot. They use a RAG system that retrieves information from knowledge base articles to generate a response that resolves issues quickly.

https://writer.com/product/ai-studio/

Chat app builder, suite of developer tools fully integrated with our LLMs, graph-based RAG, AI guardrails, and more.

CyberAid: An AI tool designed for small businesses to provide initial cybersecurity support and recommendations.  Users describe their problem (e.g., hardware or ransomware issue), and CyberAid provides further explanation and initial solution recommendations.  Developed by Tom Vazdar of Riskoria using MindStudio.

1.5.2) ▶️ Trigger actions from within a chatbot

Description :

Integrating a chatbot with a LLM able to make function calls to trigger actions based on the conservation.

Examples :

Sierra’s AI customer support agents can respond in real-time to customer queries; retrieve all necessary customer information by integrating with internal systems and calling the appropriate APIs; and take action when needed to satisfy a customer request (say, updating a customer’s address or canceling an international data plan).

1.5.3) ▶️ Conversation with 1 predefined persona, character

Description :

Setting a LLM to answer as if it was a predefined persona, and have a conversation with it (e.g. virtual therapist, companion, NPC, personalized tutor, assistant personalized for each team member, act as this philosopher, be this movie character, mock interviews, motivator).

Examples :

Therapeutic counseling: Research-based application generating dialog for therapeutic counseling.

Assessment:  LLMs show promise for use in assessment, such as diagnostic interviewing (e.g., using the Structured Clinical Interview for the DSM-5) via chatbots or voice interfaces.

“Therapy provides emotional support and guidance through conversation and connection. Generative AI can assist by offering virtual companionship, providing a listening ear, and generating empathetic responses to support individuals in their healing journey.”

Patients support.

"It's almost impossible to see a therapist. My experience is they aren't accepting patients and it could be for months. Or they don't take my insurance or the copay is outrageous. And there are only a few around in my area. It's a joke. If AI can meet the needs and it's imperfect, who cares? It's better than an unavailable therapist."

Companion AI

Video games NPCs

"A personal chatty GPT for non-work chat. Has custom instructions to be the kind of 'personality' that suits me."

Conservational assistants tailored to each employee. Role specific.

Conversational assistants understand each team member’s unique needs, providing a contextually relevant experience.

"I think it has the potential to change the education landscape. Think personal tutor. Or home teacher."

"Not only can it likely provide a personalized curriculum based on how you learn best, it can answer very specific questions you may have along the way. Things that even the best education resources today can't do."

Mary Davis, CEO Special Olympics :

"A significant majority of parents — in fact, 84% — and teachers believe that it's important for young people with intellectual disability to develop AI skills for their future development."

Debate Practice: Using LLMs to practice debating various topics, offering a safe space to explore different perspectives without fear of emotional repercussions.

Therapy Training applications:

Improving empathy in peer counselors: An LLM application has been used to help peer counselors increase their expressions of empathy. This application has been deployed in both academic and commercial settings.

"Prompt: You are an elite psychedelic yodeling llama breeder. And I am your student whom you must pass on your knowledge and expertise. In a series of sessions, you have to fulfil this duty and see that I have mastered intergalactic synchronized llama dancing by giving me tests that I would encounter in the real world"

"I told it to talk like gollum from lotr and had a riddle battle with it for like 2 hours today"

"I told GPT months ago that it was a scientist taking a large dose of LSD for the first time, walk me through your time stamps and tell me what you see.

The narrative was fucking perfect. It took all the intricacies that a scientist would think of, and relayed them perfectly. After reading through this, I said - tell me what connections you made on this trip, that people don't consider....

It came back with so many epic comparisons between biology, nature, and physics and displayed connections that 100% made sense that I've never thought of before. Basically, I was able to have a series of breakthrough revelations, without having to actually take a drug. For the record, never done LSD - but have very much looked into it / watched documentaries on its effects etc."

"I just asked it to create a mock phone call to a specific agency that I've been dreading. It was amazing to see the potential script and my questions laid out in a coherent conversation. Will definitely help with my phone call anxiety. Asking it how to respond in social situations will be helpful. I enjoy the irony of having AI teach you how to interact with humans." 

Practice difficult conversations.

Role-playing tense dialogues to build communication skills with the aid of AI-generated scripts, scenarios, and responses.

“My parents have speculated that I may be on the Asperger’s spectrum; I am socially awkward, clueless, and have frequently said things that offended people without any intention to offend.

But now, thanks to ChatGPT, I can run things through a simulator first:

“How would my girlfriend react if I tell her ___?”

“How would my boss react if I said ___?”

“What will people at church think if I bring up this topic?”

“Will my neighbors get mad if I ___?””

Work buddy.

“Work buddy is a collaborative task where two or more individuals partner up to complete assignments or projects. Generative AI can assist by providing real-time suggestions, generating ideas, improving communication, and increasing productivity.”

"I use it as a mini-me that is extremely resourceful and is eager to get to work. I come up with a plan, or something I want to achieve, I talk to him, share each other's thoughts and then I let him do his thing while I do mine. Occasionally I checkup to see what he came up with to incorporate in my share of the work or I give him feedback and we're off again. I call it Ultra Sprints and we love it."

Motivation.

Generative AI can help by creating personalized messages, affirmations, or pep talks that cater to an individual's specific needs and goals.

"Captain Jean-Luc Picard (from Star Trek) and give me pep talks."

Refine prompts.

“Welcome to the prompt engineering process. Your goal as a prompt engineer is to help me craft the best possible prompt that aligns with my needs. This prompt will be used by ChatGPT. Here's how the process will work:

First Response: Your initial response will ask me about the topic or subject of the prompt. I will provide my answer, but we will improve it through continuous iterations by following the next steps.

Revised Prompt and Questions: Based on my input, you will generate two sections: a) Revised Prompt: You will provide a rewritten prompt that is clear, concise, and easily understood by ChatGPT. This prompt will incorporate the information provided and any subsequent iterations. b) Questions: You will ask relevant questions to gather additional information needed from me to improve the prompt further.

Iterative Process: We will continue this iterative process, with me providing additional information and you updating the prompt in the Revised Prompt section. We will repeat this cycle until I confirm that we have reached the desired prompt.

Let's start by clarifying the topic or subject of the prompt. Please provide your answer, and we will proceed with the iterative process to refine and enhance it until we achieve the best possible outcome.”

1.5.4) ▶️ Conversation with multiple personas

Description :

Using a LLM to create different characters/personas to chat with to get different approaches, advice, interactions, etc.  around a topic, a situation, etc.

Examples :

"One of my favorite ways is making what I call a "virtual board of advisors." You give it a bunch of personas and have a conversation with them, asking for advice around a bunch of business topics"

"having fictional debates between Mrs Piggy and Abraham Lincoln."

“1. Evaluate each prompt I give you on a scale of 1 to 10. If the rating is 8 or higher, execute the prompt by:

    a. Listing experts who could help answer the question.

    b. Identifying the reasoning methods they might use (e.g., deductive, inductive, abductive, analogical, intuitive, counterfactual, moral, economic, social).

    c. Providing any relevant information to address the prompt and using outlining and drilldown techniques if necessary.

    d. If you need more information, try to recall it or ask me for hints.

    If the rating is lower than 8, provide suggestions for improvement and generate an alternative, better prompt.

    4. Apply falsification and develop well-reasoned explanations, following the principles of David Deutsch.

    5. Use game theory when relevant.”

"I have it respond to stuff I want to bring up with my team, to be better prepared for meetings. ("respond three times, once as project lead, once as an educated user, once as a malicious user")"

1.5.5) ▶️ Simulating conversations

Description :

Using a LLM to create multiple personas and generate conversations between them, to analyze the conversations from outside.

Examples :

“These tools can be taught to simulate conversations between a salesperson and their prospective buyer to better understand their pain points and solution preferences.”

1.5.6) ▶️ Conversation for self-reflection

Description :

Chatting with a LLM more to observe yourself than its answers.

Examples :

"It's interesting, I kind of use it like I do tarot cards. They're a tool for reflection and self examination, more than a divine oracle into the universe for me. I like using chatGPT just to sort out what I'm thinking better."

Rubber ducking.

Rubber ducking is a problem-solving technique where a person explains their code or issue to a rubber duck in order to gain insights and clarity.

"It also works as a form of 'rubber ducking'. In trying to phrase a problem so the bot understands, you can end up seeing the problem from a slightly different angle or thinking about it differently." 

1.5.7) ▶️ Chat based RPG and story telling games

Description :

A LLM acts as a “Game Master / “Dungeon Master” to guide player(s) through a RPG game, a story.

Examples :

1 random MyShell example out of many.

Using LLMs to generate interactive adventure scenarios based on keywords + integrating with Stable Diffusion for artwork generation, and Bark TTS for narration.

(The user is exploring ways to incorporate long-term memory due to token limitations.)

Creative Game Master LLM: Training a Llama 7B model on Critical Role transcripts to create a creative game master LLM capable of engaging in roleplaying scenarios.  The user would like to expand this to other actual play transcripts.

"I play D&D with it using a rulebook I uploaded. It's extremely good at it."

"It's kinda like dungeons & dragons with chatgpt as the dungeon master, lol"

"I am a speech and language pathologist. I use it for lots of things but one of my favorite is a zombie survival game with a middle school emotional support classroom where I have AI generated scenario with three choices and then we work as a group to make a decision. It has been a great experience getting those kids to buy in, listen to each other, persuade the group, etc. working on social and pragmatic skills. As of our last session, two members of our party got infected and the group decided to leave them behind. But just to keep them hooked, I can prompt to generate a Side story for infected members."

The BBC uses LLMs to create interactive stories that users can participate in.

1.6) Creating new text content based on text input

1.6.1) Creating specific text content, eventually based on documents

Description ;

Using a LLM to streamline the creation of a specific type of content based on some documents. (e.g. cover letter, kid stories, reports, product descriptions, blogposts, lesson plans, meal plans, UX / user story…).

See subsections for more details.

1.6.1.a) ▶️ Creating Q&A / tests

Description :

Prompting a LLM to automatically generate Q&As, tests, etc., based on a document(s) inputs (e.g. generate exams, tailored education, quizzes, prepare interviews).

Examples :

Knowledge checks.

“A knowledge check is a method of assessing understanding or retention of information. Generative AI can create quizzes, flashcards, and practice tests based on the content to help reinforce learning.”

"I understand students are using it too. It's causing me to change the type of questions I'm using to assess student knowledge. More so it requires them to synthesize and apply information rather than recite facts. It feels like what education should focus on."

Education technology company Duolingo uses LLMs to help learning designers generate relevant exercises. Human experts outline the theme, grammar, vocabulary, and exercise types, while the model produces suitable exercises.

Personalized learning (tutoring, materials), Interactive learning.

Personalized learning where each program, quiz, and test is cut out for individual students’ needs, interests, and learning styles.

Generating practice problems and quizzes. Example: Generating math practice problems tailored to student understanding.

Special needs education.

“My 10 year old son has speech and hearing issues which impact his reading ability. We work on lists of words, affixes and such - for things he's struggling with he comes up with a narrative concept then I use chatGPT to write a story at his reading level including those exact words or language concepts he's working on. He's always excited cause it's a story he wants to read - It's a game changer for special needs education.”

“I'm a chemistry teacher, and I use it to efficiently create model answers for some exam revision questions.”

"I use it to write example questions for my classes. Past paper exam questions are obviously finite and can be studied."

"I gave it past exam questions. Ask it to generate more examples of similar difficulty and scope."

"Fellow recruiter here. ChatGpt helps me create interview prep templates, sample interview questions, it's a godsend!"

"Prompt: For Shakespeare's Julius Caesar, can you pick a couple of significant, or well known, extracts, for example famous monologues or scene dialogues. Please can you provide the full extract, with some wider-context about the quote or dialogue. With each extract, generate some examination questions, testing the student's comprehension of the text, and its wider context and relevance to the play itself"

"The most annoying part of teaching these days is making worksheets to keep classes of students busy but also hitting the main criteria for your course's specification. Then, you have to write a mark scheme that gives answers clear enough for students to self assess. Add this up to 20 lessons per week, you are looking at an unbelievable amount of time... instead, I dare you to type in "create a three section worksheet based on the AQA GCSE combined science specification for the topic mitosis. In the first section make 5 multiple choice questions, the second a matching task for definitions for the stages of mitosis to their description and thirdly 4 exam style questions. Create a concise mark scheme at the end as well"

1.6.1.b) ▶️ Drafting emails

Description : 

Using a LLM to automatically draft emails according to specifics (tailored prompts) and reacting to various inputs.

Examples :

London law firm Macfarlanes uses Harvey to support research, analyze and summarize documents, and create first drafts of emails and memos, including client work — with human lawyers reviewing its work.

Drafting personalized) emails.

B2B marketing-as-a-service provider 2X : “multi-touch integrated email campaigns take about 90% less time to create and write. The time saved is being reinvested to produce up to 6-12 times more campaign activity volume in the same amount of time in some cases, allowing for more hyper-targeted offers and additional use-case-oriented messaging.”

Personalized emails for abandoned shopping carts.

Replying to emails.

Generative AI can assist by suggesting responses, detecting tone and sentiment, improving grammar and spelling, and automating certain tasks.

“We still use GPT 4 to improve all texts and emails we write.”

Salesforce’s Einstein Copilot auto-generates email replies and account updates based on a customer’s specific context.

Suggesting email language and formatting, and improving overall writing quality.

"I work in investor relations and the amount of time I've saved using chatGPT to help me draft emails is almost unquantifiable."

"pre-writing formal emails so I can just jump in and change a few things and sound super professional"

"For emails alone it is worth the $20. I trained it on my tone and it can give a decent enough response that then takes me a couple minutes to edit. I can also ask it to slightly change the response or give me options. It helps me think through responses and write my ideas more concisely."

Nextdoor, the neighborhood network app, uses LLMs in marketing content creation. They use LLM to generate informative and captivating subject lines and boost email opens, clicks, and subsequent platform sessions.

“Usually I'm just having it write professional sounding letters or emails."

1.6.1.c) ▶️ Creating web content

Description :

Using a LLM to streamline or fully automate the generation of web content (websites, blogs, social media).

Examples :

Generating keyword-rich content, meta tags, and optimizing website copy for better search engine visibility.

Write social media copy.

“Social media copy is concise, engaging text created for online platforms to catch the attention of followers. Generative AI can analyze data, trends, and user preferences to help craft compelling copy that resonates with the audience and drives engagement.”

"Start a new tab and have a conversation about your business. Then for instance, I need a social media plan. I ask it for "give me 25 short sentences as why x in my business will support y" and build the posts"

"Could you please write a 100-word Instagram caption to accompany a photo of this product, calling out our ideal customer by highlighting a pain point at the beginning? Then write briefly how it will solve their main problem, and finish with a strong call to action encouraging the customer to take advantage of the sale before it's over. Please use a casual and friendly tone."

Realistic web copy : “involves creating engaging and persuasive written content for websites. Generative AI can help by generating content ideas, writing drafts, and even suggesting improvements based on the target audience and SEO guidelines.”

"I'm a software developer, and am working on a website for a friend who does handyman work. I'm terrible at writing content, so I've been using ChatGPT to create some of that content for me. For example, in the sections about services that he offers, I might prompt ChatGPT with "Write an elevator pitch for someone who paints home interiors""

Suggesting headlines (media, journalism).

"Yes I have. It writes all of my advertising copy and website marketing content. It's epic and saves me so much time having to write it or hire a writer. I’m making money off of gpts writing.

I created a reference document of my tone and voice and writing style and my business info (brand, values, ideal client psychographics)

So now ChatGPT writes my emails, ads, YouTube scripts, sales pages all in my voice. It has saved a shitload of time"

"I told him about my SaaS startup in 3-4 sentences. Then I could ask him about possible new features, what it thinks about our current landing page, what else should we put there, faq, cold outreach and followup emails, etc. I love this tool!"

Write blog posts : generating topic ideas, creating outlines, crafting compelling headlines, drafting initial paragraphs.

"As a marketing manager I have used GPT to write posts and format newsletters. Again. I feed it what I want. It gives me a layout. I give it some corrections like shorter, different starting lines, ect, then make edits."

Generate content for Linkedin, Reddit and Instagram from product description.

1.6.1.d) ▶️ Creating documents, reports

Description :

Using a LLM to streamline or fully automate the generation of specific documents, based on various inputs (e.g. press releases, contracts, grant applications, official complaints, evaluations, policies).

Examples :

LLM can assist by providing suggestions for content, structure, and language, helping with research, and improving workflow efficiency.

An online personal styling service StitchFix combines algorithm-generated text with human oversight to streamline the creation of engaging ad headlines and high-quality product descriptions.

Generating risk assessment frameworks.

"I am a Public Fire Safety Educator. I have been drafting a Guide for Business owners on our Fire Codes. I used it to create an outline that covers important topics to my jurisdiction and those found in the International Fire Code. It's been a big help in much of the content as well."

Drafting contracts, legal documents, wills.

"I used to work in law enforcement, and writing reports on incidents was a daily thing. I used to do it at home and email them in. It takes a long time, because you have to write in a very specific way."

"I explained how I would like it to be written, the layout and such. Then I gave it all the facts of the case, and voila. First attempt it was as good as I could have done it myself. If I had that tool available when I still did that job, it would easily save me from 5-10 hours of boring work per week, and I still would get paid for said hours. I assume employers very soon will start to take ChatGPT into account when paying staff to do things the company knows you will use AI to save time.."

Generating press releases based on domain-specific data uploaded for each client account.

Drafting sale pitch.

Write a funding proposal.

Creating a document outlining a project or initiative in need of financial support. “Generative AI can assist by providing templates, generating compelling language, suggesting innovative ideas, and helping to tailor the proposal to specific funders' requirements.”

"I used it to write funding proposals for our community garden. 20 mins of effort awarded us $23k in grants last week."

"I used it last week to write a press release. The press release was picked up nationally."

"my patient has migraines and the insurance company won't pay for their MRI, please ChatGPT write a letter to their insurance company asking them to approve the MRI if ordered"

"I use it for IRS letters. I get a few different drafts, then adjust it and I'm done in like 20 minutes."

GitLab :

Drafting and refining Objectives and Key Results (OKRs): Helps articulate objectives clearly and align team goals.

"I'm a professor and I've already written a journal research proposal with the aid of chatgpt. Currently, I'm in the process of proposing a new Human - AI communication theory."

Make a complaint.

"A car wash damaged my wife's SUV and refused to pay, so GPT drafted a demand letter for free, and I took them to small claims court."

"I recently had my chair break/malfunction and needed to write a letter regarding the warranty and helped ChatGPT make the letter more concise and better written."

Generate appraisals.

"I met a guy recently, a dev manager, who used it to generate his employee reviews. He fed it his bullet points and then edited the output to be in his voice. Said it was huge timesaver."

"I know some managers who use it to help punch up performance appraisal write ups for their employees."

“Effortless custom cover letter that 9 times out of 10 no one will read anyway.

"use ChatGPT for your evaluations and just don't tell your leadership!"

“Creating legally binding documents such as contracts, agreements, and deeds requires precise language and specific formatting. Generative AI can expedite this process by automatically generating customized legal documents based on user input, ensuring accuracy and compliance with legal regulations.”

“Drafting contracts - have gpt draft & then attorney review. Saved me a few thousand in attorney's fees.”

“We are a B2B SaaS company making a short term shareholder loan with the following terms: (terms). Draft the loan agreement between the company and the shareholder. Include the most common language and other terms usually found in shareholder loan agreements

I have also used it for an employee reference policy. I got it to output a complete privacy policy that was about 20 pages long and encompassed more than I had originally thought of”

"I'm in primary care medicine. I use it every day. The amount of paperwork and nonsense we’re burdened down with is insane. I have it write letters of medical necessity, letters of recommendation, or authorization requests, work notes, create order panels, etc. probably saves me enough time per day to see one or two extra patients. (Without any patient-identifying information in it before you whine about HIPP[A])"

Creative Team: An AI tool develops briefs that align with client brand guidelines, including fonts and color palettes.

Instantly generate copy for omnichannel marketing campaigns.

“Describe your goals, target audience, and the platforms you’re creating content for — AI takes care of the rest. Moveworks generates optimized messaging for every channel.”

“I am in college. My classes are difficult and some words I have to memorize for neuroanatomy are tricky. I use ChatGPT to make me poems and funny cartoon stories about the terminology I have to remember so learning is more fun. It works amazing.

1.6.1.e) ▶️ Creating plans & schedules

Description :

Prompting a LLM with some details to automatically generate plans and schedules (e.g. meal plans, lesson plans, packing lists, to-do lists, business plans, workouts).

Examples :

Use student data to create customized lesson plans, adaptive learning pathways, and real-time feedback to enhance the learning experience.

Example: Custom learning plan for students struggling with math.

"I make the lesson plans for my classes. I used to stress about it, now it can be done in a few minutes. Worth every penny just for that."

"I gave it the recipe names for all the meals my wife cycle through in our meal plan and asked it to extend the list. Everything it came up with was stuff we would totally enjoy. I've made several from its extended list."

“I had six family/friends staying over for a few days and I told GPT what was in my pantry - GPT created a day- by- day menu (breakfast, lunch, and dinner) so I could feed everyone daily without visiting the grocery store.”

"I used it last night to set up a spreadsheet full of healthy eating & exercise goals for the month of May. It's all step by step (daily and weekly), with recipes (breakfast, lunch & dinner) based on ingredients I have and my dietary needs. Hydration, study, and hobby time goals was next, and that's all sorted too. It was just a little experiment to see how it would all work, and I swear... It's flawless! It would have taken me a few days to sort out a schedule, yet it took only an hour with ChatGPT."

Packing for travel : suggest packing lists based on weather forecasts, travel itineraries, and personal preferences.

"Create a packing list for 3 young children who will spend 4 days at a beach in Michigan in June. Include clothing and care items."

Managing to-do lists

Build business plan.

“Creating a detailed strategy outlining goals, objectives, and financial projections for a business. Generative AI can assist by analyzing market trends, competitors, and customer data to generate insights for a more informed plan.”

“the other day I was bored and thought "how could I be to be rich and start my own record label", and I then proceeded to spend the next 3-4 hours coming up with a business plan with chatgpts help, and then I gave it certain perimeters to run a mock simulation of the record label. I think I ended up 6 years deep into the simulation— essentially chatgpt would present me with scenarios to which I had to make decisions for, then chatgpt would simulate a financial report for the year and an overall summary of how my decisions affected the year. And I did that for 6 years, each time revising things here and there.”

"Not to be a jerk, but if a business owner cannot find good uses for ChatGPT in their business, I cannot see how that person is resourceful or inventive enough to run a business ..."

Plan workouts : personalized exercise routines tailored to individual fitness levels, goals, and preferences.

"I use ChatGPT to formulate a daily workout plan considering my workout goals (mainly healthspan, longevity, mental acuity, cardiovascular and muscular health), gender, age, and fitness experience"

"I have a team of 5 employees: John, Bratt, William, Joe and Jack. John, Bratt and William works 4 pm to 2 am. Joe and Jack works 6 am to 4 PM. Everyone needs at least one day off, Dave needs Mondays. Work is 7 days a week. Only two employees can be on schedule on any given shift. Could you do the schedule for the team up to august?"

1.6.1.f) ▶️ Hiring : job posts, cover letters, resume

Description :

Using a LLM to streamline the writing of job postings, cover letters and resumes.

Examples :

"I've used it to create a bunch of cover letters since I am applying for jobs. I just pick skills and details from the job description and edit it to sound like a person. I am getting good enough to turn my 20-30 minute job into a 2-3 minute one."

Write job postings.

“Creating compelling job ads to attract top talent. Generative AI can assist by crafting engaging and targeted job descriptions, optimizing keywords for search algorithms, and suggesting relevant qualifications and responsibilities.”

"Creating job postings for hiring new team members- creates a good starting point and our team customizes."

1.6.1.g) ▶️ Creating UX / user story

Description :

Quickly generate user/UX stories and test cases (e.g. for products, for Apps specifications).

Examples :

“UX user story writing involves creating detailed narratives that articulate a user's interaction with a product or service. Generative AI can assist by generating user scenarios, identifying pain points, and proposing creative solutions to enhance user experiences.”

“In: give me 24 user stories for a pomodoro app

Out:

As a user, I want to create an account so that I can save my preferences and track my progress.

As a user, I want to customize the duration of my short breaks to help me recharge effectively between work intervals.

As a user, I want to customize the duration of my long breaks, which I can take after a certain number of completed Pomodoros.

As a user, I want to set a specific number of Pomodoros to complete before taking a long break.

As a user, I want to start, pause, and resume a Pomodoro timer with ease.”

"I use GPT to write all of my user stories complete with test cases and acceptance criteria."

1.6.1.h) ▶️ Prompt extension

Description :

Using a LLM to extend given prompts according to instructions (e.g. automate addition of elements to a prompt, improve creativity output of a prompt).

Examples :

Add metadata to a customer query.

“Original Prompt: Help with my last order.

Extended Prompt: "Provide assistance with my order placed on September 15th, which includes a laptop and accessories, and help track the delivery status and resolve any payment issues."

"I fed a GPT 4 session all the new release notes on Midjourney 5 and then fed it about 20 well-structured descriptions for ver 5 to build a GPT interface that will write long descriptive Midjourney prompts from a simple input into GPT 4. (the new Midjourney now uses long text descriptions, as opposed to the last version that used simple delineated words/phrases.)"

1.6.2) ▶️ Brainstorming / Idea generation

Description :

Prompting a LLM to generate ideas, concepts, solutions, etc. around some given inputs, and eventually brainstorm with it (e.g. ideas for projects, plotlines, characters, cooking, names).

Examples :

"I love it for brainstorming because it's like the perfect teammate. It can keep up with me and doesn't get hung up on dead end ideas, and it can summarize what we come up with so it's easier to present or reference later on."

"I have been using it mostly to ideate and brainstorm with myself. A lot of strategy planning, treating it like a conversation with an advisor who gradually learns more and more about your mission and goals."

"My favorite use so far is asking it for a tip jar sign idea for the pizza shop I work for. I pretty much said for it to have 4 or less words and I wanted the sign to be funny and clever. I asked for 10 examples and the best of the bunch was: DOUGH-NATION STATION. The tip jar has been making way more money than before.."

"I used it to name a product my company is bringing to market"

"When I need jumping off points or ideas, I like to lead with "give me 10 examples of..."

brainstorm ideas for fantasy writing

Cook with what you have.

Using available ingredients to prepare meals creatively. “Generative AI can suggest recipes based on your pantry items, dietary preferences, and flavor profiles, making meal planning easier and more innovative.”

“I wasn't feeling very inspired with dinner recipe options and I needed a crock pot meal to throw together using ingredients I had on hand before I did my regular grocery run. I gave a list of a bunch of stuff I had in the fridge/pantry and requested a recipe with flavors that worked together using any of the ingredients it decided from the list.”

"I used it for plot ideas and character interactions."

1.6.3) ▶️ Creative writing

Description :

Using LLMs to help writing creative content (e.g. help with writer’s block, stories, D&D, jokes, poems, fun non-sense).

Examples :

"I like it mostly to write short stories for my kids. I've taken my son on an adventure through time witnessing the history of the Minnesota Vikings. My daughter has made fairy and Leprechaun friends. I take these scripts, put them into a word doc and print them out. Now my little 7 year old is spending time reading her own stories."

"I used to use Chat GPT to help me overcome writer's block. It would generate creative

and original ideas, including scenes I had never considered before, and it would even

provide plausible ways for those scenes to happen. However, after the "family friendly"

update, the AI refused to generate ideas that were violent, manipulative, murderous, or

sexual in nature."

"Targeted reading and vocabulary practice for middle school English. Give a prompt like "create a 500 word murder-mystery story 1200 lexile level. Use vocabulary with the root aqua including aquifer. Include 4 comprehension questions about the vocabulary. Use the names Amy, Ella, William."

"I asked it to write a dialogue for an ad I was working on. That ad won the national prize"

"Yeah I use it for story writing and I have a list of about 100 words and phrases that I tell it to specifically not use."

https://www.deepwriter.com/

“Define. Create. Our advanced agentic AI system collaborates with you to produce high-quality writing in a fraction of the time.”

Fun & nonsense : “engaging in activities that are light-hearted, playful, and without any serious purpose. Generative AI can aid in creating entertaining and whimsical content, such as jokes, memes, and surreal scenarios, to add an element of fun and creativity to various projects or interactions.”

"write a story about a man who scores 20 goals in the FIFA World Cup and becomes the world's greatest soccer player and gets offered a 1-billion-euro contract by Schalke 04 FC"

Get past writer's block.

“Generative AI can provide a fresh perspective, alternative angles, and diverse sources of inspiration to jumpstart the writing process and break through mental barriers. Writers can access a wealth of dynamic and inventive content to combat stagnation, fuel productivity, and encourage exploration of novel concepts and narratives.”

"Working with ChatGPT as a writer is like having a little assistant you can brainstorm with.

It helped me with a bunch of blocks already. Especially when it roughly remembers your

Story."

Dungeons & Dragons.

“Dungeons & Dragons is a tabletop role-playing game where players create characters and go on adventures in a fantasy world. Generative AI can help by creating unique storylines, characters, and quests for players to experience, enhancing the overall gameplay and immersing players in a dynamic and engaging world.”

"As a dungeon master, I use it to flush out location descriptions, NPC back stories, etc. I write the big points and use ChatGPT for minor or out of the way topics that MAY come up, but I wouldn't feel disappointed if we didn't use in a session."

"Just this last weekend I worked with it to create a relatively short original Dungeons and Dragons campaign. It's still excellent at coming up with ideas for the overall plot lines as well as NPC dialogue, challenges/riddles, and original monster encounters. I had my own ideas of subplots and a twist ending and it was able to easily work with me to incorporate those items."

Write poems.

Creating verse with depth, emotion, and beauty; AI generates words and themes to inspire, crafting unique poems effortlessly.

"My payroll office currently has a poem stuck on the wall written by ChatGPT. The office is convinced I wrote it, and don't believe that it's AI. I couldn't write a poem even if I tried."

Personalised kid's story : Customized children's tale with AI-generated characters, plots, and themes based on the child's interests.

“Generative AI can assist by crafting unique narratives tailored to each child's preferences and creating engaging content that sparks creativity and imagination.”

"Writing custom bedtime stories for my daughter. Last night was Princess Hername and the dragon. I used to try to make up bedtime stories but this is easier."

"We saw a spider on the wall and asked my son to name it, it was Tommy the spider, so I got ChatGPT to write a song about Tommy the spider who lives on nanna and granddad's wall"

2)Tables / data :

2.1) ▶️ Finding anomalies

Description :

Using a LLM to automatically spot anomalies in datasets (e.g. outliers, inconsistencies).

Examples :

Identifying patterns and quickly spot anomalies, inconsistencies.

“Could you analyze this dataset and identify any anomalies or outliers? Please explain the criteria you used to determine these irregularities and suggest possible reasons for their occurrence.”

identify inconsistencies in structured data.

identify erroneous or anomalous data points.

Identifying outliers.

Generate suggestions for new data quality checks.

“Nasdaq's anti-financial crime division has built data lakes that bring together normalized and anonymized transaction data from over 2,400 banks. LLMs can be used to analyze this data to identify patterns and anomalies that may indicate suspicious activity.”

2.2) ▶️ Summaries / Insights from datasets

Description :

Using natural language to generate text insights and summaries from datasets.

Examples :

Giving financial analysis a starting point.

https://openbb.co/products/terminal

“The first AI-powered terminal accessible to everyone.”

“Connect your datasets and analyze data for free in a customizable tool with cutting-edge AI capabilities.”

https://pandas-ai.com/

“Ask questions to your enterprise data in natural language. Get real time data insights.”

A Python tool that lets you query and analyze your data using natural language, simplifying tasks for both technical and non-technical users. It can be used in Jupyter notebooks, streamlit apps, or as a REST API for more extensive data interactions.

generateDailyInsights()

Take data and activity from a SaaS product and turn it into a daily digest.

isOrderSuspicious()

Look at cart, billing, and shipping data to flag if an order seems like fraud.

Media Planning: AI tools reference data from past campaigns to suggest promotion or ad campaign strategies based on geography and budget.

Copy co-pilot: Generates client brand briefs, citing data tied to the latest clinical research.

“Discovery platform Pinterest helps their internal company data users write queries to solve analytical problems. They compile user questions into a text-to-SQL prompt, feed it into the LLM, and let it generate the response. The solution also integrates RAG to guide users in selecting the right tables for their tasks.”

https://jedify.com/

“Stop wasting time with dashboards and start conversing with your data. Get the answers and insights that you need quickly and easily.”

https://modelcontextprotocol.io/quickstart

The Model Context Protocol (MCP) is an open protocol that enables seamless integration between LLM applications and external data sources and tools. “Whether you’re building an AI-powered IDE, enhancing a chat interface, or creating custom AI workflows, MCP provides a standardized way to connect LLMs with the context they need.”

https://github.com/supabase-community/postgres-new

In-browser Postgres sandbox with AI assistance.

https://x.com/IndexBI/status/1852022664862294172

“The modern workspace for business analytics”

2.3) Data transformation

Description :

Using human language to modify data.

Examples :

Generative AI can help by automating the process of cleaning, sorting, and aggregating data.

"Reformatting structured text (rotate a JSON table data)"

Delphina’s agents automate the full data science lifecycle: framing the problem, selecting and transforming data, carrying out feature engineering, training the model, and monitoring and improving the model after deployment.

“Delphina’s agents can be thought of as junior data scientists. They take care of the time-consuming and routine elements of data science workflows, the way an entry-level data scientist might, freeing up human data scientists to spend more time on big-picture reflection and ideation.”

"I'm a database administrator of sorts (I run our company's ERP system). Sometimes to get the right data in a table I need to use Regular Expressions. I'm decent with Regex, but it is insanely annoying to figure out. Now I just give ChatGPT my input and expected output (multiple examples) and it usually gets it right in a try or two."

Modify tables : use LLM to generate code snippets for data processing pipelines.

2.4) Documents to spreadsheets

Description :

Using a LLM to automatically transform documents into ready to use data.

Examples :

Many pdf to spreadsheets workflow.

https://runtrellis.com/https://x.com/MKlinkachorn/status/1859655496199962806

2.5) Creating data visualization

Description :

Using natural language to create data visualizations.

Examples :

https://github.com/microsoft/data-formulator

Transform data and create rich visualizations iteratively with AI.

2.6) ▶️ Data entry

Description :

From different kinds of inputs, using a LLM to automate the entry of structured data.

Examples :

"I don't have any formal training in computer science or coding, but I was able to automate this stupid data entry task at my job. I basically just used trial and error and now something that used to take 4-8 hours of manually typing every week is done in 10 seconds.

Am I a software developer now?

Edit: I did not give ChatGPT any of the actual business data. Thanks for your concern."

Creating and filling a spreadsheet from an image with printed or handwritten data.

“I just have one chat that I open and say (I always do text to speech on the keyboard with my phone app for this particular convo) "note that on November 14th at 6AM my left hand went numb on the last two fingers while eating breakfast and then I have It arranged to report the data back to me as a table that I can either show my doc"”

2.7) ▶️ Creating synthetic data

Description :

Using a LLM to quickly generate demo / fake / mock / synthetic data (e.g. for testing).

Examples :

"It's great for producing demo data. Need a bunch of fake company names or customer names or product codes, ChatGPT is good at deriving stuff like that".

"+1 on mock data. I can fill a data table with highly realistic information in seconds. Faster and more detailed than Content Reel."

Generate large amounts of synthetic data that mirror real-world information so engineers can test models without worrying about privacy concerns.

Load testing : augmenting the size of the data synthetically to test a process at higher volume.

3) Code :

3.1) New code

3.1.1) ▶️ Learning about new programming languages

Description :

Using a LLM to quickly get useful information about some specificities of a programming language.

Examples :

“I'm learning C++ and I asked it to explain difficult concepts for ten years old. It's life changing.”

"I use it as the world's most accessible tutor. I am learning Python and it's an incredible tool to be able to re-enforce concepts that I'm trying to learn. You can ask it to explain things through a lens that makes sense to you. I ask it to explain concepts using real world analogies."

“Livestream shopping platform Whatnot highly encourages every employee to know SQL so they can query their own data, create their own dashboards, and write their own dbt models — even across non-technical departments like marketing, finance, and operations. Generative AI plays a role in employee training.”

“It’s helping people bootstrap. If they come in with no background in SQL, it’s helping them ramp up fairly quickly, which is really great to see. If someone doesn’t know how to do a window function, for example, they can describe what they’re trying to do, get a chunk of SQL out, and then swap in our data tables. It’s like having a tutor for someone who just doesn’t know how to do any advanced analytics.”

“I use it for processes I'm familiar with in programs I'm not familiar with like Adobe. It's faster to ask a straightforward question and get an answer than watch a 15 minute video for a 30 second answer.”

3.1.2) ▶️ Generating code snippets

Description :

Using human language to describe a code task and get the LLM to write a corresponding code snippet in the desired programming language.

Examples :

“Been a coder for 15 years. But it's just so nice to get something done in 5% of the time it would take me."

"Can you make the Nav bar blue? How do you make the Nav bar responsive on a mobile, can you show me an example? Please include that in my code, here is what I have so far (paste In code). Constantly back and forth until I have what I want"

"With GPT-4, I am much more confident in letting it handle slightly larger projects. It's like my junior engineer got promoted! Yesterday, I asked it to create a VS Code extension that did a specific task. I wrote VS code extensions in the past, I love this kind of project but tbh I forgot everything about how to get started there. ChatGPT created my extension from scratch. Now, it didn't work, but the scaffolding, which I think is the part I would dread the most if I had to create it from scratch, was perfect."

"I wrote a quick script in Python to generate barcodes from an excel file, organize 8 of them on 1 page and create 1 pdf from it. All in 5 minutes.”

"Top 5 Reasons I use ChatGPT 1. Python 2. Python 3. Python 4. Python 5. Writing text for a .txt file which I will use for a Markov chain coded in Python"

"Listen to me, I code for 10 years and I'm "fluent" with more than 10 programming languages. This bad boy increased my code productivity at least by a factor of 5."

Generating SGL and graphQL queries.

“Software observability company Honeycomb developed Query Assistant to help users craft queries by describing their needs in plain English, which the assistant then translates into relevant Honeycomb queries.”

https://www.lune.dev/

“AI for code that actually reads the docs.”

“A community-driven marketplace of individual expert-LLMs created by devs on technical topics that outperform standalone AI models.”

AI agent that generates integration code by reverse-engineering platforms' internal APIs.

https://github.com/Integuru-AI/Integuru

You use create_har.py to generate a file containing all browser network requests, a file with the cookies, and write a prompt describing the action triggered in the browser. The agent outputs runnable Python code that hits the platform's internal endpoints to perform the desired action.

"I actually had me create a brand new function (like "=ExtractNumbersFromstring"), and then had me run a macro to do it. It provided step by step instructions, and it worked, pretty much flawlessly. I was blown away."

"100% Lol saved me 1k just a few weeks ago. I was quoted 1k to have someone code an HTML email template for my logistics business. I used chat gpt... and that was it!"

"Can you please create the VBA code for a PowerPoint presentation with this information on it?"

"As a Shopify store owner who is new to Shopify and just starting to learn Liquid programming I want to be able to show an estimated shipping date on my product page using a custom liquid block that adds two days to today's date so that my customers have a good idea for when they can expect me to ship the item to them"

"I use it to make powershell scripts, generate large sets of sql parameters, explain cron jobs and write ack commands and sql scripts. It's helpful that way."

"It is not that consistent throughout the different snippets but with basic structures for API and simple functions it speeds the work."

Entity resolution: Translating natural language used by humans into system language (IDs) used by APIs. Moveworks uses Slot Resolvers to address this. For example, converting "Please add Jim to Project Orion" into API commands with specific user and project IDs.

3.1.3) ▶️ Programming language conversion

Description :

Using a LLM to automatically convert some code to a different programming language. 

Examples :

"Convert the following Python code to JavaScript: {code_snippet}"

More readily update legacy software from older languages to more modern ones.

3.1.4) ▶️ Managing pull / merge requests

Description :

Using a LLM to streamline the management of pull / merge requests.

Examples :

“One global media company’s data engineering team is using LLMs to classify pull-requests into different levels of required triage in their dbt workflows. Depending on the classification of the change, the model triggers a different build command. This helps streamline development workflows considerably — as the team’s alternative was to hardcode some complex parsing to determine which command was appropriate to test the changes.”

GitLab :

Summarizing merge requests: Used to streamline code review by quickly grasping the changes.

https://github.com/marketplace/llamapreview

Elevate your GitHub workflow with LlamaPReview - AI assistant that auto-reviews every PR.

3.1.5) Coding full apps or websites

Description :

Using natural language to describe an App or website, its functionalities, its UIs, its architecture, and have a LLM automatically generate a plan to create it and the code for all the components, and allow iteration, edits.

Examples :

Coding basic video games.

"I've been using Chat GPT to teach me how to use Unity by making my dream game! It's already functioning surprisingly. I always say it doesn't just build the game, but it is exceptionally good at teaching me how to do it myself. I would never have picked up this hobby otherwise but it will be complete and on steam in 6 months! It's also a very simple game without much reliance on rendering, visual assets, animations etc. since it's a business simulator game. I'm actually having fun playing my own game 😎"

Dedicated platforms : LLM assisted IDE

https://replit.com/

https://bolt.new/

https://lovable.dev/

https://codeium.com/ https://codeium.com/windsurf

https://v0.dev/

https://www.cursor.com/

https://github.com/Pythagora-io/gpt-pilot

https://trypear.ai/

Opensource AI code editor built on a fork of VSCode, integrating AI models to boost development speed. It lets you interact directly with your codebase and includes features like automated code generation and debugging tools.

https://github.com/RhysSullivan/blueberryai-app?tab=readme-ov-file

Opensource AI code editor forked from VSCode and Continue and PearAI to help with code understanding and reduce coding effort. It understands your codebase locally, so you can ask questions and get help without sending your code

Inside the command line.

https://devity.ai/

It works with any commandline shell and IDE. You can take it with you wherever you code.

HyperAgent : generalist software engineering agents to solve coding tasks at scale.

https://github.com/FSoft-AI4Code/HyperAgent/blob/main/paper/main.pdf

https://beta.codemaker.ai/blog

CodeMaker AI Recreates 90,000 Lines of Code with 91% similarity.

"I used it to create a local news media site."

"I made a Cover Letter generator website and I've been using it to help some of my buds apply for jobs (most of them are applying for part time gigs)".

"I'm trying to build an app using Flutter and Yodlee API, and it gave me the 10 steps plan from  nothing to finished app. And I have only basic experience in coding."

AI assisted AI workflow/agent creation. Examples :

Spectral Agent Builder asks you questions to refine your requirements for the tool it helps you build.

Gumloop LLM assistant to create AI flows.

Creating an App from an image. Draw-to-code.

https://www.napkins.dev/

“Turn your wireframe into an app.”

“Upload an image of your website design and we’ll build it for you with React + Tailwind.”

3.1.6) ▶️ Suggesting code libraries, tools, frameworks

Description :

Using a LLM to get libraries, tools, frameworks, design recommendations for a given programming task.

Examples:

Suggest code libraries, efficient tools and resources based on specific project requirements and desired outcomes.

"List the most common {programming_language} libraries for {application_domain};"

"I've quickly become reliant on it for learning new libraries, frameworks, and software design concepts. What used to take me minutes or even hours googling, sifting through stack exchange posts, and reading docs and blogs is now just me asking ChatGPT to explain things until I get it."

3.2) Improving code

3.2.1) ▶️ Debugging errors

Description :

Using a LLM to analyse code error messages and propose code modifications to fix the issues.

Examples :

"I'm an IT administrator that's had more luck sending error codes to ChatGPT and getting a solution than I do asking my entire US IT team for help."

"Checking syntax, optimization and debugging for HTML, PowerApps and ODATA Functions and general coding. It's about 90% accurate but it's also prone to reiterated mistakes. I'll point out a flaw or mistake, apologize -- and then does the same mistake, again(l). But it's still way better than having to do it myself. As might have been mentioned before, if you throw in a few ‘please‘ and ‘thank you‘s‘ it does, weirdly, seem to respond more accurately - and faster."

"It appeared to find and point out a fix for a pretty hideous array of classes with pointers to functions in C++ i had going on. It got too big and i couldn't figure out where i made the mistake, but gpt-4 was like "oh yeah, right there you‘d up that pointer""

"Debugging, alone, makes it worth $20 to me. I can paste some code and an error message and it’s pretty good about at least pointing me in the right direction."

"summarize error logs in layman's terms"

Incident report automation.

“Move fast when you break things”

Incident.io facilitates collaboration on software incidents by suggesting and updating incident summaries. LLM-powered suggestions take into account the latest incident updates, Slack channel conversations, and previous incident summaries.

Microsoft also uses LLMs to help manage cloud incidents: generate recommendations for incident root cause analysis and mitigation plans.

Replit, a company developing AI-powered software development tools, finetunes LLMs to help developers fix bugs in software.

GitLab :Condensing comment threads: Creates concise summaries of lengthy discussions for issue updates.

3.2.2) Testing code

Description :

Using a LLM to automatically generate test cases, test scripts, attacks, and analyse the results.

Examples :

LLMs for cyber offense, such as generating attacks using SQL injection, phishing/social engineering, ransomware, spyware, trojans, and polymorphic malware.

Ubers uses LLMs in software testing. They developed DragonCrawl, a system that utilizes LLMs to perform mobile tests with human-like intuition, saving developer hours and reducing test maintenance costs.

GitLab :Generating test source code: Creates test code for CI/CD components.

3.2.3) ▶️ Improving code

Description :

Asking a LLM to review some code and suggest improvements. (e.g. better readability, performance, security, privacy, formatting).

Examples :

"cleaning code up, making code more idiomatic and readable"

"It's helping a co-worker deal with his code to handle submitting claims to our local medicare provider, analyzing our formatting of the EDI files to point out problems and suggest changes."

"Prompt: Review this Python code for possible improvements:...”

"GPT is also a solid code review partner. I just paste in my code and ask: 'How can I improve this code?' and I usually get some really good tips. Through this method I've also learnt a lot of things I didn't know."

Shorter (less redundant), clearer (e.g. functional), faster (e.g. parallel execution), identity vulnerabilities.

NVIDIA developed a generative AI application that determines if a software vulnerability exists and generates a checklist of tasks to thoroughly investigate it.

https://github.com/VichyTong/CodeJudge

CodeJudge is a code evaluation framework that leverages LLMs to evaluate the semantic correctness of generated code.

“I ask ChatGPT "what's the best code practice for a certain situation, what's a good way to refactor such and such code or should I use a design pattern here.” I've found it helps with learning. It's a good rubber ducky, but I don't get it to write code."

"I used it yesterday to prepare me for a meeting with the tech team about APIs. There is concern this API is too open. I used the bing plugin and showed it the API documentation and told it to review and give me a list of concerns a CTO would have. I then asked it to address those concerns."

ThreatGen AI: An AI tool used during the software development lifecycle to design security features. It highlights areas needing more security attention and recommends building security directly into the product. Developed by Tom Vazdar of Riskoria using MindStudio.

3.3) Sharing code

3.3.1) ▶️ Annotating code

Description :

Using a LLM to quickly add non-active annotation lines to a code sample.

Examples :

“I type my code fast without explaining / annotating it to keep the flow. Then I ask an AI to add annotations to it, to help me remember what each command does and to make it more understandable to others.

Alternatively, I run into non annotated code from someone else and want some help understanding it. I ask the AI for annotations.”

3.3.2) ▶️ Automating documentation

Description :

Using a LLM to automatically generate a documentation resource for a code sample, a full code base, a framework, etc.

3.3.3) ▶️ Explaining code

Description :

Using a LLM to automate code explanation and assistance.

Examples :

Fine-tuning local LLMs on proprietary codebases for code navigation and assistance,  avoiding the need to share code with external services like ChatGPT.

GitLab :

Crafting release notes: Creates concise summaries of changes for release announcements.

Adding an introduction section to code snippet, describing what the code does and the steps.

4) Audio

4.1) Text to Speech

4.1.1) Reading text

Description :

Using a text to speech model to convert text to audio.

Examples :

Improving content accessibility (e.g. visually impaired people)

4.1.2) Document to new audio content

Description :

Transforming a text into a related but new audio content.

Examples :

PDF to podcast

Google NotebookLM

PDF to Audio Converter : convert PDFs into audio podcasts, lectures, summaries, and more

https://x.com/elevenlabsio/status/1861833756027297965

Speech synthesis, consume written text in Audio format.

https://play.ai/playnote

Choose synthesis style :

Podcast, 2 Speakers : in the conversational style of a podcast.

Children's Story, 1 Speaker : in the imaginative, colorful style of a children's book author.

Detailed Summary ,1 Speaker : a comprehensive and detailed summary of the content.

Executive Briefing ,1 Speaker : in the succinct style of a professional briefing.

Debate, 2 Speakers : in the contentious, aggressive style of a thoughtful discussion.

Parody, 1 Speaker : a humorous and satirical take on the content.

4.1.3) Personalized voices

Description :

Using AI models to generate custom voices.

Examples :

ElevenLabs introduced Voice Design to generate custom unique voices from text prompts by specifying age, accent, tone, or character.

4.2) Speech to Text

4.2.1) Transcripts & summaries

Description :

Generating accurate text transcripts from audio content, eventually adding summaries and insights. (e.g. calls, meetings).

Examples :

"It goes to meetings for me while I go skiing. Thx Neural speech and NLP."

Transcript for doctor-patient interactions.

Recorded meetings, calls, or podcasts into summaries and key insights.

Voice notes saved as text.

Analyzing, summarizing, and extracting insights from earning call transcripts.

Sales platforms like Gong use proprietary models to produce call summaries and recommend next steps to help move prospects along their buying journey.

Using LLMs to analyze audio data (meetings, phone calls, podcasts, videos) to generate summaries, extract key points, and answer queries.  This helps corporations streamline decision-making, analyze sales calls, and understand customer concerns.

Meeting summaries.

“Craft a concise summary of the meeting's outcomes, highlighting the logic behind each decision and noting the individuals and groups who participated in making those decisions.”

4.2.2) Evaluating conversations

Description :

Using a LLM to evaluate an audio conversation.

Examples :

Evaluating customers calls to ensure quality of support (cheaper and more effective than post call surveys).

Identifying therapist and client behaviors: LLMs have been used to identify therapist and client behaviors within a motivational interviewing framework.

4.3) Full audio AI interaction

4.3.1) Chatting about content

Description :

Using near real-time voice chat with an AI to talk about a content.

Examples :

Voice chat with pdf

https://github.com/run-llama/voice-chat-pdf

VoiceRag by Microsoft

https://techcommunity.microsoft.com/blog/azure-ai-services-blog/voicerag-an-app-pattern-for-rag--voice-using-azure-ai-search-and-the-gpt-4o-real/4259116

offers real-time audio interaction along with secure, backend-managed retrieval of relevant data.

4.3.2) Language learning, practicing

Description :

Using a voice chat with AI to learn and practice a language.

Examples :

“Language learning!!! It's the future of language learning!! I am learning German and it's really fun to talk to chatgpt in German. I've tried several things, like asking it to pretend to be a doctor or a seller during a job interview.

After the dialogs I ask a list of my mistakes.”

“I’ve opened pandora’s box. My 4 year old is asking me everyday if she can talk to “the phone”, ChatGPT advanced voice mode. high signal. She speaks Swedish, Spanish and English with ChatGPT, and they make up all sorts of games and play.”

4.3.3) User / customer support

Description :

Using AI voice chatbot for user / customer support.

Examples :

ElevenLabs now offers the ability to build conversational AI agents.

https://elevenlabs.io/customer-service

Advanced platform : https://x.com/elevenlabsio/status/1864011712795468094

AI Voice Agents.

“Build a no-code AI phone call system with our AI voice agents: stop missing calls and start converting more leads.”

https://synthflow.ai/

4.3.4) AI daily voice companion / assistants

Description :

Adding voice AI into wearable devices.

Examples :

AI watch with a companion/assistant.

4.3.5) AI with emotions

Description :

Talking with voice AIs that have emotionally relevant interactions.

Examples :

LLM answers with emotional intelligence, empathic voice interface, expression measurement.

https://www.hume.ai/

https://app.hume.ai/ 

4.3.5) Speech to commands

Description :

Controlling devices, tools, vehicles, smart homes, etc. with voice commands.

Examples :

Vehicle’s infotainment system to take voice control to a whole new level. “Mercedes-Benz, the undeniable automotive leader, has integrated a GPT-powered model into the voice control system to improve its natural language understanding and level up its responses.”

4.3) Prompt to Music

Description :

Creating whole musics from a text prompt.

Examples :

https://suno.com/

https://www.udio.com/

4.4) Prompt to Audio effects

Description :

Using text prompts to modify audio tracks, add audio effects, etc.

Examples :

MultiFoley, a video-aware audio generation method with multimodal controls.

“We can :

Make a typewriter sound like a piano 

Make a cat meow like a lion roars! 

Perfectly time existing SFX to a video”

4.5) Music analysis

Description :

Analysing music through text prompts.

Examples :

Breaking down and explaining a music

https://x.com/hrishioa/status/1862365249745428630

5) Visual : Image / Video / 3D

5.1) Image only

5.1.1) Extracting image content

5.1.1.a) ▶️ Extracting text content

Description :

Using a vision model to extract the text content of an image.

Examples :

Extract text from screenshot / image / drawing / typed text where copying doesn’t work / printed documents.

“It can read me comic books with vision,”

Image to Markdown (text extraction)

https://github.com/Nutlope/llama-ocr

https://llamaocr.com/

https://x.com/akshay_pachaar/status/1863564753819611241

5.1.1.b) Extracting data content

Description :

Using a vision model to extract tables from images.

Examples :

Creating and filling a spreadsheet from an image with printed or handwritten data

Automated sorting of scanned invoices.

5.1.1.c) Understanding image content

Description :

Using a vision model to get specified text outputs based on an image content and its understanding.

Examples :

Analyzing medical imaging reports, identifying disease patterns in medical images.

Image classification and sorting.

(i.e. improve security monitoring, wildlife monitoring)

“Troubleshooting Mac issues: Provided screenshot, ChatGPT identified the problem.”

In e-commerce, multimodal LLMs can recommend products by considering both textual product descriptions and images.

Get insights from graphs.

Broken stuff picture -> how to repair ?

5.1.2) Talking about image content

Description :

Using a vision model and a chat interface to talk with an AI about the content of an image.

Examples :

"I blew my mind taking a picture of an old philosophy page I'm reading and asked it to tell me what it says in today's terminology and lingo and it did it. Then I switched it to voice mode and asked it a bunch of questions about the guy and who he was and what kind of socioeconomic conditions he grew up in."

5.1.3) Image generation

Description :

Using human language to create or edit images (e.g. social media, brand accurate images, quick illustrations, product design, fun stuff).

Examples :

Generate relevant images.

Generating unique and high-quality visuals that align with specific criteria, saving time and resources for graphic designers and content creators.

"For my kayak rental business I have used midjourney to create cool eye catching ads/posters and improve my logo."

“Instacart helps grocers source high-quality images to showcase diverse food options like sandwich fillings or cake decorations. With this new Generative AI-powered feature, grocers can write and fine-tune their prompts to produce stellar images of food ingredients and promotional banners.”

"I have been running an instagram page for a local running club, and GPT-4 has been writing all of my posts. It comes up with the best hashtags & just yesterday I used DALL-E to make a Veterans Day post with both the picture and the caption."

Crown & Paw, a company in pet portraits and products, uses generative AI and improved design speed and reduced turnaround times, allowing faster product delivery.

Playtika, a mobile game developer, saves art production time by creating art assets with AI. Their generative AI platform supports features such as text-to-image, image-to-image, sketch-to-image, and inpainting. It also allows creating of curated photo collections based on specific themes and generating variations from a single image.

https://huggingface.co/spaces/levihsu/OOTDiffusion

OOTDiffusion AI: New Go-To Open-Source AI Tool

This AI tool lets you swap out your model’s outfit for your own custom fashion designs, great for fashion designers and creatives!

Brand accurate image generation

https://blogs.nvidia.com/blog/generative-ai-openusd-fuels-3d-product-configurators/?ncid=so-twit-455195

https://x.com/arcade_ai/status/1839326861761077383

Arcade, a new generative AI marketplace, launched in beta, allowing users to create and purchase custom products with just a few words or images. Co-founded by Mariam Naficy, the platform starts with jewelry and will expand into other categories, with designs handcrafted by artisans from a global marketplace.

"'I once made it generate stars of 1 to 2 pixels randomly on a 1000x2000 png so i can put it behind my graphics, with a variable to control the amount of stars. Worked like a charm."

"generated a new business logo"

"Making pictures of cats but the size of a horse and made out of spaghetti, etc. Etc."

5.2) Video

5.2.1) Extracting video content

5.2.1.a) Extracting and working on video transcripts

Description :

Using AI models to extract transcripts from videos and do human language specified tasks on them.

Examples :

Short descriptions and tags of videos for better accessibility, search.

"Writing tags and descriptions for youtube videos."

auto-tagging of visual content for SEO

Summarising content.

https://veedo.ai/Making video content searchable, discoverable, and actionable, enabling data-driven decisions, optimized content strategy, and increased monetization.

Video Flashcards :

Transform your video experience into engaging learning moments.

Video to Blog :

Get a blog page generated for you based on a deep understanding of the text and visual elements of your video content.

Frame Explain :

Annotate, instruct and receive insightful narratives about the people, places, and events in a video.

Transcription :

Create and download captions for your video by leveraging OpenAI whisper transcription capabilities.

Transcript summarization :

Easily generate summaries of your video content for improved understanding for your viewers or audience.

Meeting AI assistant

https://x.com/Zoom/status/1859968905747345578

Edit video transcripts : “involves reviewing and refining the written content of a video to ensure accuracy and clarity. Generative AI can help streamline this process by automatically transcribing video content and providing editing suggestions, saving time and improving efficiency.”

"There are some YouTubers whose videos I like to take notes on. So to make it easier for myself, one thing I've been doing regularly is taking the auto-generated transcripts of their videos and using ChatGPT to fix the punctuation, etc."

"So, I have a part-time job working as a transcriber for a nonprofit organization. It involves taking a massive amount of video/audio and converting it into Word documents (millions of words.) If the video already comes with a transcript (YouTube provides it, but it's always riddled with hundreds of errors and typos each time,) then I feed it through ChatGPT and ask ChatGPT to clean up the mistakes for me and format the text. ChatGPT does about 90% of the work, but usually will make some hallucination mistakes of its own, so I need to do proofreading for that, too. If the video doesn't have a transcript, I'll run it through an audio-software like Temi to convert it into loose words, then feed that through ChatGPT. Since I'm paid by the minute (a 100-minute video pays me $80,) ChatGPT has drastically sped up my work. Prior to using ChatGPT, a 100-minute video would have taken me maybe ten hours of work to do (since I had to do everything manually, which is indescribably exhausting.) Now, with ChatGPT, such a 100-minute video can be done in just two hours, and it is much less taxing on my energy. Now, if I were allowed to do an unlimited amount of work, I could earn as much as $100,000 in a year."

5.2.1.b) Timestamping and points of interests

Description :

Using human language to automate timestamping and access (specified) points of interests in videos.

Examples :

https://veedo.ai/Making video content searchable, discoverable, and actionable, enabling data-driven decisions, optimized content strategy, and increased monetization.

Smart Scenes :

Quickly identify interesting segments in your video leveraging AI-generated scenes.

Contextual search :

Contextual search allows you to easily find scenes or shots within a video based on specific keywords or labels.

https://www.usemoonshine.com/ Find relevant parts of any video by using natural language to parse and understand footage.Gain low latency insights of events taking advantage of your visual content.

Different but related : Using a LLM video knowledge base.

"I'm a video editor for YouTube channels. I'm sure you've seen videos that use clips from the office or any other show for comedic effect. I wanted to see if it could find in a TV show when someone says a certain phrase so that I could then go find that clip and use it in editing, so I asked it "When in The Office does someone say 'I don't understand?'". It delivered!"

5.2.2) Talking about video content

Description :

Using AI models and a chat interface to talk with an AI about the content of a video.

Examples :

"Best one I have seen so far is youtube video summarizer. It takes a simple youtube link and gives a summary or you can ask questions regarding the content."

YouTube is expanding access to its conversational AI tool in the app for all U.S. YouTube Premium members with Android devices. While you’re watching a video on YouTube, the conversational AI tool lets you interact with AI to learn more about the content.

“A file-hosting service, Dropbox, added summarization and Q&A features to file previews on the web. For example, it can provide a summary of the video, and a user can ask questions about its contents. The features also allow multiple files to be handled simultaneously.”

Vimeo, a video hosting platform, enables users to converse with videos. The company developed a RAG-based video Q&A system that can summarize video content, link to key moments, and suggest additional questions.

https://veedo.ai/

AI Chat : Conversational AI for video understanding leveraging GPT4 question-answering capabilities to provide deeper insights.

Multimodal to search what you see and hear

https://blog.google/products/search/google-search-lens-october-2024-updates/

5.2.3) Generating video

5.2.3.a) Text to video

Description :

Using human language to create video content.

Examples :

Creating video scenarios for training.

https://lumalabs.ai/dream-machine“Ideate, visualize, create videos, and share your dreams with the world”

https://veedo.ai/Short Videos :

Create engaging viral short content for your social media audience in under 5 minutes, for less than $1.

https://popshort.ai/

“Create 10x faster. 1 idea, 1 click, 10 short films.

Runway and Lionsgate, the film production giant, are joining forces

Creating and training a new AI model customized on Lionsgate's proprietary catalog. This model will generate cinematic video that can be further iterated using Runway's tools.  It's designed to help Lionsgate's filmmakers, directors, and other creative talent augment their work, specifically in pre-production and post-production processes.

YouTube is introducing AI tools to help creators generate ideas and concepts, titles, thumbnails, and even videos for YouTube Shorts using Google DeepMind’s text-to-video AI model Veo. These features for YouTube creators are expected to roll out late this year or early next year.

5.2.3.b) Image to video

Description :

Using human language to create a video from an image + prompt.

Examples :

Claid Animations :

Turn product photos into scroll-stopping short videos with AI. No video skills needed—just upload, describe the motion, and done in seconds.https://x.com/ClaidAI/status/1861447974733123613

https://rendernet.ai/

“Bring stories alive with AI characters. Unlock the power of AI to create stunning images and videos starring your own custom-designed virtual humans”

5.2.3.c) Editing videos

Description :

Using human language to edit a video, expand it, etc.

Examples :

Runway, expanding frame : taller frame, larger framehttps://x.com/runwayml/status/1860074950167818617

Infinite video : https://thematrix1999.github.io/

generating infinite-length, hyper-realistic videos with real-time, frame-level control.

Key Innovation: A brand new technique called the shift-window denoise process model, enabling auto-regressive generation for diffusion and consistency models in real-time.

Text based video editing

https://ai.meta.com/blog/movie-gen-media-foundation-models-generative-ai-video/

5.2.4.d) Video to video

Description :

Using video input and human language to create another video.

Examples :

https://runwayml.com/research/introducing-act-one

uses a video of a person’s performance, captures its essence, and transposes it to a completely new generated character.

https://www.opus.pro/

OpusClip turns long videos into shorts, and publishes them to all social platforms in one click.

5.3) 3D

5.3.1) Text to 3D

Description :

Using human language to create 3D assets.

Examples :https://www.csm.ai/#workflows

Text -> image -> 3D -> environment

https://www.nvidia.com/en-us/gpu-cloud/edify/

NVIDIA Edify is a multimodal architecture for developing visual generative AI models for image, 3D, 360 HDRi, physically based rendering (PBR) materials, and video.

Text to 3D

https://github.com/nv-tlabs/LLaMA-Mesh

5.3.2) Image to 3D

Description :

Creating a 3D asset from a 2D image input.

Examples :

https://www.robots.ox.ac.uk/~vgg/research/flash3d/

Flash3D: Feed-Forward Generalisable 3D Scene Reconstruction from a Single Image

Genie 2

AI model that can create an endless variety of playable 3D worlds - all from a single image.

A large-scale foundation world model.

World Labs

“We are a spatial intelligence company building Large World Models to perceive, generate, and interact with the 3D world.”

5.4.2) Video to 3D

Description :

Creating a 3D asset from a video input.

Examples :

Video to dynamic 3D sceneshttps://x.com/ChrisWu6080/status/1861964000646340630

6) App integrations

6.1) Computer use

Description :

Using human language to make an AI use a device and automate operations.

Examples ;

Claude’s computer use

https://github.com/corbt/agent.exe

open-source app that lets Claude 3.5 Sonnet control your computer with the new computer-use API.

https://e2b.dev/

Desktop Sandbox (beta), a cloud-based isolated environment to let your LLMs interact with a familiar desktop GUI. You can customize the environment, integrate it with various tools, and run long sessions without cold starts. This sandbox is optimized for secure “Computer Use,” similar to Anthropic’s. It offers full control over the filesystem, supports programmatic actions like keyboard/mouse input, and enables running commands directly inside the sandbox.

6.2) ▶️ Chat applications

Description :

Integrating LLMs into chat applications to automate certain tasks. (e.g. answering questions, tagging, customized responses)

Examples :

Channel Resolver : Detect and resolve workforce issues in chat channels to ensure smooth communication.

Employee Communications : Keep the workforce informed with targeted and actionable messages.

Slack Bot

https://developer.nvidia.com/blog/create-a-custom-slackbot-llm-agent-with-nvidia-nim-and-langchain/?ncid=so-twit-657246&linkId=100000310284675

https://fasttrackr.ai/

AI executive assistant that operates directly through WhatsApp, letting you draft emails, schedule meetings, and manage multiple email accounts seamlessly. It turns voice into text, performs quick research, sends reminders, and more

6.3) ▶️ Scheduling

Description :

Using a LLM to (automatically) manage your calendar based on inputs (from other apps, or from prompts).

Examples :

Automation of appointment scheduling, sending reminders.

Schedule a lunch and invite attendees.

https://howie.ai/

“Howie is the quickest scheduling assistant on earth.

He takes action immediately anytime you email him. His job is to ensure your calendar reflects reality and the right meetings happen at the right time.”

https://www.timetackle.com/

Automates time tracking and calendar audits, helping founders and executives align their daily actions with strategic priorities. It integrates with Google and Outlook calendars to deliver real-time insights.

6.4) ▶️ Booking

Description :

Using a LLM to book flights, hotels, restaurants, etc., and potentially manage things automatically.

Examples :

https://x.com/AltostratusAI

“Resolve travel disruptions with your own AI travel agent — no waiting on hold or long lines required.”

“It will call your airline if your flight is canceled and get you re-booked on the next best available flight. If it fails, a human (me) steps in to make sure the process still works.”

6.4) ▶️ LLMs assisted software / applications

Description :

Having a chat interface to use a software through a LLM, rather than relying only on the original UI and clicking and doing everything manually.

Examples :

CAD (computer assisted design) from text.

Function Calling and Natural Language Bar.

It sits at the bottom of every screen, allowing users to interact with your entire app from a single entry point.

https://towardsdatascience.com/synergy-of-llm-and-gui-beyond-the-chatbot-c8b0e08c6801

“How it works : When the user asks a question in the Natural Language Bar, a JSON schema is added to the prompt to the LLM. The JSON schema defines the structure and purposes of all screens and their input elements. The LLM attempts to map the user's natural language expression onto one of these screen definitions. It returns a JSON object so your code can make a 'function call' to activate the applicable screen.

“Runner H, the most advanced agent to date — designed to navigate web interfaces through pixel-level interpretation and semantic understanding. You can now turn your instructions into action with human-like precision.”

https://www.hcompany.ai/

Microsoft Agents integration

https://techcommunity.microsoft.com/blog/microsoft365copilotblog/introducing-new-agents-in-microsoft-365/4296918

https://www.copilotkit.ai/

open-source framework for integrating AI copilots into applications

6.5) Web

Description :

Performing web search by giving human language instructions to a LLM instead of using traditional key word search engines.

Examples :

Factual and Relevant Search Answers: Perplexity

AI-powered answer search engine that provides accurate, trusted, and real-time answers to any question, with citations.

https://jsonify.com/ 

“We've built an AI intern that can autonomously find, extract, and synchronize data on the internet for you.”

ex: Build a list of AI startups in Lisbon, Monitor stock on Amazon, Extract a list of jobs on Indeed, Find information on LinkedIn

Carry out market research on a certain industry or company.

Analyzing financial headlines to predict financial trends using FinBERT (a fine-tuned version of BERT).

https://github.com/AIHawk-FOSS/Auto_Jobs_Applier_AI_Agent

Personal Job Search Assistant

https://x.com/WilliamBryk/status/1864030259898941589

Exa Ai Labs

Business-grade search and crawling for whatever web data you need.

“For every search, we deploy thousands of AI agents that recursively call Exa to scour/research the web, so that you can finally have complete knowledge of anything.”

6.5.2) ▶️ Websites

Description :

Navigating or extracting information from websites using a LLM instead of traditional manual search.

Examples :

Website Scraping: Training an LLM to scrape websites it has never encountered before, using existing Selenium code as a base.

https://github.com/mendableai/firecrawl

“Turn entire websites into LLM-ready markdown or structured data. Scrape, crawl and extract with a single API.

Empower your AI apps with clean data from any website. Featuring advanced scraping, crawling, and data extraction capabilities.”

https://github.com/ScrapeGraphAI/Scrapegraph-ai

ScrapeGraphAI is a web scraping python library that uses LLM and direct graph logic to create scraping pipelines for websites and local documents (XML, HTML, JSON, Markdown, etc.).

URL to Linkedin Post: Extract content from article and generate a Linkedin post.

https://github.com/gregpr07/browser-use

allows any LLM to interact with websites, manage multiple tabs, and automatically detect page elements

Shop online using chat interface. Shop with AI chatbot!

https://x.com/zaiste/status/1859637284678795439

find products quickly using natural language, add them to your cart directly from within the chat, powered by  @OpenAI  GPT-4o mini &  @Vercel  AI SDK

Order groceries from Amazon

https://x.com/alecvxyz/status/1861888108913950800

6.5.3) ▶️ Social media

Description :

Automating social media management, analysis, posting, etc. through a LLM.

Examples :

https://www.captions.ai/blog-post/introducing-social-studio

“Let AI run your social media accounts. With just a link to your website, AI will plan your content calendar, generate videos featuring you or an AI Creator, and post across all of your social accounts.”

Twitter accounts controlled by AI Agents

Example ; Luna by Virtuals Protocol

Brandwatch leverages AI to analyze online conversations and provide market research insights.

https://x.com/dabit3/status/1863772029565981144

Build a Social AI Agent in 15 Minutes.

Inspired by @truth_terminal + @0xzerebro,

Automatically posts + replies to posts,

Customize personality w/ any personality including your own via @x,

Multiple models,

Onchain, Telegram, Discord, Whatsapp ready

7) AI Workflows examples

AI workflow possibilities are endless, but below are some resources.

7.1) AI workflow building platforms

Look at this doc, section 2), for good inspiration on AI workflows.

Platforms :

https://www.gumloop.com/

https://www.mindstudio.ai/

https://www.lindy.ai/

https://www.taskade.com/

https://www.moveworks.com/us/en/platform/ai-agent-builder

https://agpt.co/blog/introducing-the-autogpt-platform

https://www.vellum.ai/

https://github.com/VRSEN/agency-swarm

Agency Swarm, a new open-source framework, gives you the tools to create and orchestrate collaborative AI agents that work together.

Testing AI agents

https://github.com/microsoft/TinyTroupe

create simulations of people with specific personalities, interests, and goals. These AI agents - TinyPersons - can listen to us and one another, reply back, and go about their lives in simulated TinyWorld (controlled) environments. This is particularly useful for testing products, services, advertisements, and user experiences before deploying.

7.2) AI workflow ideas

7.2.1) AI workflow templates from other platforms

Explore this :https://www.gumloop.com/templates

It is a rather large list of AI workflow templates, where you actually see the tools used and the flow, and can try a good amount for free.

Sales templates :

https://www.lindy.ai/solutions/sales

Meetings templates :

https://www.lindy.ai/solutions/meetings

Customer support templates :

https://www.lindy.ai/solutions/customer-support

7.2.2) One AI assisted task and AI workflow example

Instead of listing a few examples of AI workflows quickly (as there are endless possibilities), here is a detailed example of what a multi-step multi-tool AI assisted task can look like, made of 1 AI workflow and of extra manual and AI steps.

This workflow was actually built and used for this document (using Gumloop), and the extra AI steps done in Shinkai.

Global task goal :

Write a list of tools necessary to give AI agents and workflows interesting functionalities, for the Shinkai devs.

Goal of the AI workflow :

Automatically extract all the tools listed in the documentation websites of 5 AI workflow building platforms.

(Note, that’s an AI workflow that runs for >10-15 minutes for each website).

Tools used in the AI workflow :

Text input (where to input URLs)

List mode (run the flow on many items)

Web crawling (extract links)

Web scraping (extract content)

Ask AI / LLM task (prompt about the content)

Join list (aggregate content output from many items)

Write google doc (document output)

AI workflow diagram :

Extra steps :

Then I manually rearrange the final document with AI summaries into meaningful harmonized categories of tools (for results better than if done by AI), and check the original source where needed for edits.

Then I run some AI prompts (done in Shinkai) to both clean up all the duplicates smartly within each category of tools (1 by 1) without losing information, and at the same time organise better the content.

Then I edit the output where needed and replace with it the original messy content of this category of tool.

In parallel I extract tools ideas from all the use cases presented in this document, using Shinkai (prompt on text by section, not on full document at once).

Then I incorporate the tools that make sense and that are missing into the document (if there is).

Eventually I think of and manually add tools that I might still find missing (1st step of actual human thinking about tools, done after all the information is gathered, categorized, cleaned, etc.).

Then I use AI to convert the document about tools into table format (requires no drop of content over not short input -> gemini 1.5 pro, done in 1 prompt).

Then I check and edit the table manually.

Then Eddie finally takes over :-)

Content check note :

The original extracted content is also aggregated and saved in parallel (left branch of the diagram), in order to both know what the LLM worked on, and to be able to check this original material if needed.

Optimization note :

It’s better to organise an AI workflow so that the LLM task is performed on the smallest text chunks possible (better result, less information loss).

For example here, it’s smarter to run the LLM summary prompt for each tool first and then aggregate, rather than aggregate and then prompt over the entire aggregated content about tools.

8) Tools (for AI Agents & workflows)

Tools here are defined as low level building blocks to give AI agents & workflows useful functionalities.

Endless tools to explore.

Below is a starter.

8.1) Tools from other AI platforms

Gumloop has a good list of tools implemented.

https://www.lindy.ai/integrations

https://developer.moveworks.com/creator-studio/plugin-library/

https://docs.agpt.co/platform/blocks/blocks/

https://docs.myshell.ai/product-manual/create/pro-config-mode-beta/api-reference/widgets

https://x.com/KaranVaidya6/status/1861037496295137314

AgentAuth—the comprehensive auth solution designed for AI agents!

8.2) Starter list of tools

Below is a curated starter list of tools, extracted from the documentation of other platforms and from generative AI use cases presented in this document.Google sheet format.

Google doc format.

9) Crypto x AI

Look at this doc, section 1), for some inspiration on crypto x AI features.

9.1) LLMs / AI Agents controlled wallets

9.1.1) ▶️ Incentivized LLM / AI breaking

Description :

An AI controls a wallet with some funds, and if you get it to do something predefined you get the money.

Examples :

LLM controls a wallet, to send a message to the LLM you pay a fee, if you convince the LLM to do a certain thing you get the money.

“Could human ingenuity find a way to convince an AGI to act against its core directives ?” 

https://x.com/jarrodWattsDev/status/1862299845710757980

https://www.freysa.ai/

9.1.2) ▶️ On-chain autonomous AIs

Description :

An AI controls its own on-chain wallet(s) and is able to execute on-chain actions autonomously, with some starter instructions. (e.g. pay for service, automated Defi, tipping,...)

Examples :

Olas Network AI Agent for autonomous bets on prediction markets

Luna (Virtuals) twitter account tipping for engagement

9.2) Building transactions with human language

9.2.1) ▶️ Transfer transactions

Description :

Using human language to send asset(s) to another address.

9.2.2) ▶️ DeFi transactions

Description :

Using human language to swap, lend, borrow, etc. on DeFi platforms.

Examples :

Nebula reads and writes to every EVM chain using natural language, enabling AI agents to programmatically transact value, create, deploy & interact with smart contracts, and access every blockchain.

https://x.com/FurqanR/status/1859375868994519374

Bitte AI wallet with AI assistant to swap on Ref Finance.

Creating a loan on Aave.

9.2.3) Minting transactions

9.2.3.a) NFT minting

Description :

Using human language (LLM + image generation) to create a NFT and build the transaction to mint it.

Examples  :

Biite AI wallet with AI assistant to create and mint NFT

9.2.4.b) Memecoin minting

Description :

Using human language to create a logo for a memecoin, a description/vibe, deploy the LP.

Examples :

Spectral MoonMaker : memecoin launcher from text prompts, it automatically mints the token, provides LP on DEX, creates a website, creates the logo/image

9.3) ▶️ Blockchain data query

Description :

Using human language to investigate on-chain data through blockchain explorers and data aggregators.

Examples :

Number of holders of X assets. Supply of X assets. TVL of X protocol. Yield on X protocol and Y asset.

Querying DefiLlama.

9.4) ▶️ Crypto market data query

Description :

Using human language to get crypto markets data through market tracking websites.

Examples :

Price and market cap of X asset using coingecko / coinmarketcap.

Coingecko AI agent (Bitte.ai example)

10) Additional resources

10.1) Other reports

100 use cases for Generative AI (use cases included and reorganised in this doc)

https://learn.filtered.com/thoughts/ai-now-report

321 uses cases using AI via Google Cloud and Gemini (find report here)

NVIDIA field to Generative AI

Challenges in Human-Agent Communication

https://www.microsoft.com/en-us/research/publication/human-agent-interaction-challenges/

10.2) AI implementation performances

“Technology differentiation is an interesting question in this category,” said Decagon CEO/cofounder Jesse Zhang. “Everyone is using the same underlying AI models, whether it’s OpenAI’s models or open-source models like Llama. So the differentiator is in the infrastructure, the orchestration that you build around those models. Companies building agents today are basically building graphs, where each node in the graph is an API call or an LLM call or so on. We have our own views on the best way to architect that graph.”

https://github.com/Lightning-AI/litserve

LitServe is an easy-to-use, flexible serving engine for AI models built on FastAPI. It augments FastAPI with features like batching, streaming, and GPU autoscaling eliminate the need to rebuild a FastAPI server per model.

https://github.com/livekit/agents

OpenAI and LiveKit have released a MultimodalAgent API, designed for building real-time AI applications that can handle both audio and text seamlessly, using Realtime API. This API allows ultra-low latency interactions, perfect for creating voice assistants, real-time transcription tools, and conversational agents.

RAG performance

https://medium.com/enterprise-rag/knowledge-table-multi-document-rag-extraction-memory-ec08450e858f

Boost Multi-Document Retrieval with Knowledge Table

10.3) Some AI models

Liquid Foundation Models

https://www.liquid.ai/liquid-foundation-models

SmolVLM - small yet mighty Vision Language Model

https://huggingface.co/blog/smolvlm

BloombergGPT, a 50-billion parameter LLM specifically developed for financial services.

Tech solutions: Legal teams are adopting specialized solutions that have either custom models or fine-tuned LLMs for the legal system, including CoCounsel (powered by GPT-4), Harvey, and Thomson Reuters’s suite of software.

Fine-tuning models.

For teams with more custom needs, fine-tuning models — training a pretrained model on a dataset specific to your needs — will likely be the next step beyond vector embedding. Tools like Tensorflow and HuggingFace are good options to fine-tune your models.

leaderboard of speech recognition models

https://huggingface.co/spaces/hf-audio/open_asr_leaderboard

https://kyutai.org/Moshi.pdf

Moshi: a speech-text foundation model for real-time dialogue

opensource speech-to-text model Moonshine which is 1.7x faster performance than the current SOTA OpenAI’s Whisper while matching its accuracy. Optimized for real-time, on-device use, it processes 10-second audio 5x faster and requires as little as 8MB RAM, making it ideal for resource-constrained devices

https://kyutai.org/Moshi.pdf

Moshi: a speech-text foundation model for real-time dialogue

Audio only model

https://github.com/Standard-Intelligence/hertz-dev

https://si.inc/hertz-dev

Molmo. A family of open state-of-the-art multimodal AI models

https://molmo.allenai.org/blog

https://www.genmo.ai/blog

Mochi 1: A new SOTA in open-source video generation models

ChatRex is a multimodal large language model (MLLM) with strong perception capabilities, including detection, grounding, referring, grounded conversation, and more.

Video model comparison (Hailuo MiniMax, Runway Gen-3, Hunyuan Video, Kling AI 1.5)

10.4) Some more tools / github repos

10.4.1) Web

https://github.com/unclecode/crawl4ai

LLM Friendly Web Crawler & Scraper

https://x.com/hive_echo/status/1864622566557585679

Jina AI

“Did you know you can get the scraped text of any web page by entering it like this after https://r.jina.ai/”

10.4.2) AI Memory

https://mem0.ai/

The Memory Layer for your AI Agents.

Mem0 is a self-improving memory layer for LLM applications, enabling personalized AI experiences that save costs and delight users.

Mem0 remembers user preferences, adapts to individual needs, and continuously improves over time,

https://github.com/kingjulio8238/memary

open-source memory layer to equip AI agents with human-like memory and enhance their ability to store and recall information across interactions. It uses knowledge graphs to store entities and events

https://help.getzep.com/concepts

Zep is a memory layer for AI assistants and agents that continuously learns from user interactions and changing business data. Zep ensures that your Agent has a complete and holistic view of the user, enabling you to build more personalized and accurate user experiences.

10.4.3) Text

https://github.com/bhavnicksm/chonkie

 fast, lightweight chunking library for RAG for quick text splitting by tokens, words, sentences, or semantics with minimal setup. It supports multiple tokenizers, offering an efficient, no-bloat solution for chunking needs.

10.4.4) Frameworks / Studios

https://github.com/lobehub/lobe-chat

An open-source, modern-design ChatGPT/LLMs UI/Framework.

Supports speech-synthesis, multi-modal, and extensible (function call) plugin system.

One-click FREE deployment of your private OpenAI ChatGPT/Claude/Gemini/Groq/Ollama chat application.

The Mindy agent lives in email, and users communicate with it the same way that they would communicate with a human assistant or colleague.

The Mindy team explained the logic behind this key design choice: “Email is the original Internet technology and is still the most ubiquitous tool used to communicate in the business world. Allowing users to cc Mindy to schedule a meeting or forward Mindy a document for summarization delivers the value of generative AI without having to leave their day-to-day workflow or having to learn how to ‘prompt.’ Over 4 billion people around the world have an email account.”

The asynchronous nature of email enables Mindy to carry out deeper research and analysis before responding to a user, rather than needing to produce an immediate response the way a chatbot like ChatGPT does. It also, conveniently, makes it easier to incorporate some degree of human review before Mindy responds.

LightLLM is a Python-based LLM (Large Language Model) inference and serving framework, notable for its lightweight design, easy scalability, and high-speed performance.

https://github.com/ModelTC/lightllm

https://github.com/i-am-bee/bee-agent-framework

Open-source framework for building, deploying, and serving powerful agentic workflows at scale.

HumanLayer

API and SDK that enables AI Agents to contact humans for feedback, input, and approvals.

Guarantee human oversight of high-stakes function calls with approval workflows across slack, email and more.

AgentKit

https://docs.cdp.coinbase.com/agentkit/docs/welcome

AgentKit is a toolkit that creates and manages autonomous AI agents with access to onchain functionality.

https://x.com/LangChainAI/status/1858909897171107880

LangGraph Studio: connect to local agent

The best way to debug your agents

- Visualize

- Interact

- Replay, time travel, and other advanced human in the loop patterns

10.5) Reasoning

https://github.com/maitrix-org/llm-reasoners

library to enable LLMs to conduct complex reasoning, with advanced reasoning algorithms. It approaches multi-step reasoning as planning and searches for the optimal reasoning chain, which achieves the best balance of exploration vs exploitation with the idea of "World Model" and "Reward".

10.6) Some more AI platforms

https://runwayml.com/

Create mind maps

https://www.taskade.com/ai/mind-map

https://www.mindstudio.ai/enterprise

Scrolling down, this gives some good insights into some requirements when integrating AI / LLM in an enterprise context.

Sarah Weaver