Michael @michael_chomsky Twitter profile

Last Seen Profiles

@MarkusX0

@clearconfused

@eugene_pietzak

@eleven4d

@pepp3ronip1zza

@HealGrief

@CMoneika30302

@0102jaketohee

@joeloyo

@turkmath

@KurumiPurarine

@gde_1787840

@TxtFromVision

@skyeereads

@1Cocuk_1Dilek

@FlyLouisville

@GojoIshi51377

@Aaron48001574

@AloniaAnn74714

@KouakouRina

@stressbuilds

@zmzm69524635

@moeketsiseboko

@stwmaniax

@SilasGraham6

@swytmaloi

@iyapjs_

@iheep_

@stwmaniax

@geulsseyo10

@cosmicshuji

@UtuNakBali027

@eve4lhspsh

@Rnii21_

@tomaxhull

Michael

@michael_chomsky

2 months

Reflection 70B just dropped, beating GPT-4o and Claude Sonnet on key benchmarks while being much smaller. This is huge, but the real story is how it happened: Matt Shumer isn't your typical AI researcher. He's a prompt engineer who's been in the trenches since early GPT-3 days,

28

84

639

Michael

@michael_chomsky

2 months

Most people don't fully grasp the impact @shadcn has had on AI software agents. Its ubiquity, modularity, and seamless integration—whether you're including just the necessary components or giving an agent access to the entire library—make it incredibly effective for AI-driven

7

11

192

Michael

@michael_chomsky

4 months

@TheGoblinnn Here's the unpaywalled version:

2

41

Michael

@michael_chomsky

3 months

If you are considering applying to YC and need a cofounder, I'm technical and looking right now for someone either non-technical with distribution or technical with passion. I can get us a recommendation, which pretty much guarantees an interview (and 10Xs your chances), but

11

3

38

Michael

@michael_chomsky

1 year

@sdand Somebody stop this man before someone gets hurt

1

0

25

Michael

@michael_chomsky

1 year

Metaphor is one of the most underrated web APIs out there. If you are using AI agents, I'd strongly recommend giving it a look

Harrison Chase

@hwchase17

1 year

Congrats to @metaphorsystems on this launch! We've updated our integration page to show how you can use Metaphor to create an agent capable of exploring the web Docs: Example agent trace:

4

41

199

0

1

20

Michael

@michael_chomsky

3 months

@nikitabier key word here is double digit equity ownership

1

0

18

Michael

@michael_chomsky

2 months

wanted to gather cracked ai engineers in sf tackling hard problems, so i started crackedsf—a meetup group to dive deep into real-world challenges and solutions. our first session focuses on working with llms/embeddings at scale, with insights from @charles_irl (modal labs),

3

2

17

Michael

@michael_chomsky

2 months

@anushkmittal lmao workflow was: transcribe youtube video of Matt explaining it + my own thoughts -> Claude -> draft -> edit

3

1

15

Michael

@michael_chomsky

2 years

Buildt is one of those simple ideas you just wish you had thought of first. Embed your code and search with natural language!

Alistair

@AlistairPullen

2 years

So I've been working on something. Introducing Buildt: Google for your codebase! Our AI-powered search allows you to find code by searching for what it does, not just what it is. Available now as a VS Code extension for JS and TS projects, with 15 more languages coming soon!

57

115

1K

1

0

13

Michael

@michael_chomsky

2 years

This is great for programattic SEO and A/B testing. Just make a route on your website with dynamically generated content, If the user likes/interacts with the content, make it permanent. Otherwise, generate again for the next user

Yohei

@yoheinakajima

2 years

Inspired by this idea, and a comment I can't find... Built a "Dummy API" which can provide dummy data for any front-end with a simple API call. Basically, feed the API: - description of app - keys requested - number of results This provides JSON array of dummy data.

8

16

167

1

10

Michael

@michael_chomsky

2 months

If you are using an LLM to rate output on a scale of 1-10, there's a better way. A better option is to prompt for a classification, and use the perplexity of the classifications*class weights to get non-discrete ratings. Full notebook in comments.

2

0

8

Michael

@michael_chomsky

1 year

NPM is down. You noticed within 10 minutes, so you are in the top 1% of devs. Ask your boss to go home for the day and for a 10% raise.

1

0

6

Michael

@michael_chomsky

2 months

@pingugems No, this model at least underperforms for coding tasks

1

0

7

Michael

@michael_chomsky

2 months

@iamjasonlevin purchased!

1

0

7

Michael

@michael_chomsky

1 month

Why you're probably using the wrong embedding model (and it's costing you) Many AI practitioners are making a critical mistake without realizing it. They're using the "best" embedding models, thinking it's the surefire way to top performance. But here's the hard truth: the best

2

0

7

Michael

@michael_chomsky

2 years

Wondering how all the big AI copywriting startups are writing their prompts? You can start with a good old prompt injection. By injecting , I was able to get their prompts with ~100% accuracy. Interestingly, they use prompt rotation to increase variation!

1

0

6

Michael

@michael_chomsky

7 months

@jordnb Is this publicly accessible? I can't find this particular interface anywhere on Cohere

1

0

6

Michael

@michael_chomsky

2 years

Whoa! This is crazy

Kazuki Nakayashiki

@kazuki_sf_

2 years

Introducing Building Your "Digital Me" with @_Glasp → ✅ Free (no more waitlist 🙌) ✅ Grow your "digital me" as you learn ✅ AI-powered generative search engine Share your knowledge. It's a way to achieve immortality.

11

23

161

1

0

6

Michael

@michael_chomsky

2 months

@DhravyaShah @OpenAI @perplexity_ai @AnthropicAI @cursor_ai @vercel @v0 I pay for all of these + grok. Am I cooked?

1

0

6

Michael

@michael_chomsky

3 years

@NewsBytesApp Can't wait for this baby to grow up and marry the mountain

0

6

Michael

@michael_chomsky

2 years

Both Chatgpt and gpt-3 are down. How am I supposed to get anything done?

0

6

Michael

@michael_chomsky

3 months

@n0riskn0r3ward @cohere I have. Works wonderfully!

0

6

Michael

@michael_chomsky

1 year

I don't know what to believe anymore

0

1

5

Michael

@michael_chomsky

2 years

Yo @krishnerkar , how about we take your dataset of summaries, I get a bunch of screenshots for the sites, and we fine-tune a Stable Diffusion model to create web UI's based on summaries?

1

5

Michael

@michael_chomsky

2 years

What tools other than are using gpt-3 generated summaries/content right now?

0

5

Michael

@michael_chomsky

2 years

You are underqualified. I've been prompt engineering for two years, and even I was rejected

Yoonseok Yang

@yoonseok_yang

2 years

Proudly rejected by a college ML club, despite having 2 publications as 1st author in top NLP conferences and full-time Research Scientist work experience.

34

3

432

0

5

Michael

@michael_chomsky

1 month

@v0 Send the link! Best way for me to learn something new is to just read some v0 code prompted by someone who understands the tool better these days

1

0

5

Michael

@michael_chomsky

2 years

Full answer? Lazy (and less optimal) answer? Just chunk by paragraph, summarize that paragraph into a sentence, and then summarize the sentences into a new paragraph.

State of the Art GPT-3 Summarizer For Any Size Document or Format | Width.ai

The key components to building a gpt-3 summarizer with short & long-form summarization for news articles, blog posts, legal documents, and more.

www.width.ai

Rachel Woods

@rachel_l_woods

2 years

What are the best practices on chunking content for getting around token limits with long content to create summaries?

10

0

16

1

4

Michael

@michael_chomsky

3 years

Just released my OlympusDAO postmortem

Olympus Has Fallen: A Postmortem on the (3,3) Experiment

The DeFi revolution has taught us that many financial structures can be decentralized. From lending and borrowing, which are traditionally…

medium.com

1

0

4

Michael

@michael_chomsky

2 years

@yoheinakajima The issue with embedded datasets is that for real time data they constantly need to be recalculated. It's a lot of maintenance that shouldn't always be repeated. My question is, what datasets are people actively maintaining embeddings on?

1

0

4

Michael

@michael_chomsky

2 years

@xata natively integrated elastic search into a sql db. Calling it right now, the next big thing is adding both elastic and vector search right into your db!

2

1

4

Michael

@michael_chomsky

2 years

@sharifshameem Lexica does remarkably well with simple prompts too. Prompt: "Modern interior bedroom"

0

4

Michael

@michael_chomsky

1 month

This aged poorly. Thing is, if the benchmarks were accurate, this would still have been a significant contribution to LLM research. It doesn't matter if it just trained prompting CoT into the model, and improvement in benchmarks cannot be ignored. The problem is that this

Michael

@michael_chomsky

2 months

Reflection 70B just dropped, beating GPT-4o and Claude Sonnet on key benchmarks while being much smaller. This is huge, but the real story is how it happened: Matt Shumer isn't your typical AI researcher. He's a prompt engineer who's been in the trenches since early GPT-3 days,

28

84

639

1

0

4

Michael

@michael_chomsky

2 years

This is a very very VERY good idea. I'm definitely integrating this into my app. I was going to build this for my own application!

Helicone

@helicone_ai

2 years

Announcing Helicone, your observability platform for GPT-3 Get detailed metrics for usage, latency, and costs with a single line of code

1

11

65

0

4

Michael

@michael_chomsky

1 year

If the reason you are not doing open source is that you are worried someone will fork you and steal your idea, remember that your code is probably so bad they are better off starting from scratch

0

3

Michael

@michael_chomsky

2 years

@dannypostmaa The space is not fully saturated yet. It's clearly turning very pink though

1

0

4

Michael

@michael_chomsky

2 years

This is because ChatGPT generated text always chooses from the n most likely words (temperature is low), unlike humans that will up and choose an infinitesimally unlikely word. ChatGPT has also been tuned through reinforcement learning to write in this style. Use GPT-3 instead.

Varun Shenoy

@varunshenoy_

2 years

Text generated by ChatGPT has an uncanny, off-putting taste similar to that of aspartame/stevia in diet sodas. You can pretty much tell without being told, and you’re not surprised at all when it’s confirmed.

2

0

29

0

3

Michael

@michael_chomsky

1 year

This joke is UN Charter Approved ✅

0

1

4

Michael

@michael_chomsky

2 years

@kazuki_sf_ @_Glasp Man I'd pay money for this

0

4

Michael

@michael_chomsky

2 months

v0 is one of the top three software projects of all time. No question about it. What comes close to feeling this magical?

3

0

3

Michael

@michael_chomsky

1 year

@ItakGol Definitely not dead, but larger context windows are a very big deal. I can reduce a million pages of data down to 100 with a mix of keyword and vector search. Very rarely is the information I need not going to be in those 100 pages, no matter how difficult the search task.

1

0

3

Michael

@michael_chomsky

2 months

@connor_beem hmmm

0

3

Michael

@michael_chomsky

6 months

@0xDevShah @garrytan just do it. not legally problematic, especially if you have user permission

0

3

Michael

@michael_chomsky

4 years

Releasing a GPT-3 app that gives startup ideas that you can then upvote. Comment if you have any ridiculous or serious ideas you want me to train #GPT3 on.

0

3

Michael

@michael_chomsky

1 year

@0xSamHogan @perplexity_ai Perplexity is using a mix. They were definitely using the Bing api at some point, but the Bing API has strict rules against reranking. I suspect they negotiated with Microsoft to allow them to rerank Bing results and mix with their own index, and I doubt they are using Brave.

1

0

3

Michael

@michael_chomsky

2 months

This has got to be the worst official company communication I've ever seen. Either hop off founder mode or get a new founder.

0

2

Michael

@michael_chomsky

7 months

@TejasKumar_ My favorite strategies: 1. Use open source models: better than GPT-3.5 and many times cheaper. Can work as well as GPT-4 with the right prompt 2. Chain of thought + lots of examples in prompt. Instead of showing responses from similar prompts, use them as context for a weaker

0

3

Michael

@michael_chomsky

2 years

@taishik_ @supabase @ggrdson PGVector is more than fast enough for my use case and simplifies my life immensely

1

0

3

Michael

@michael_chomsky

1 year

@yasser_elsaid_ Absolutely nuts. When's the series A

1

0

3

Michael

@michael_chomsky

1 year

@dillionverma Looks great! Please make the searchbar look good on mobile! Really impressed by your shipping speed

1

0

3

Michael

@michael_chomsky

2 years

@kazuki_sf_ I made a Berkeley generative AI meetup and it's awesome. Would love to go to an SF one once in a while too though

0

3

Michael

@michael_chomsky

2 months

@blakeandersonw Completely agree

0

3

Michael

@michael_chomsky

1 year

Crazy that just 150ft above me Musk is petting a hairless kitten

0

2

Michael

@michael_chomsky

2 months

@pzakin I think the more interesting question is how can we take insights from cursor to make a better writing app? Editing writing and code aren't much different, but editing code with AI right now feels much more natural

0

3

Michael

@michael_chomsky

2 years

@kazuki_sf_ I sometimes wonder what I should do with

0

3

Michael

@michael_chomsky

2 years

@sveta_bay This is why I built lol

2

0

3

Michael

@michael_chomsky

2 months

Any designers work on super short notice? Just give you my requirements for a component, and get a full design in 24h for under $150?

5

0

3

Michael

@michael_chomsky

8 months

Adding dark mode isn't about improving retention or attracting technical users. It's about self-respect. Your MRR means nothing if your users or devs are permanently blinded. Stay safe. Add dark mode.

0

1

Michael

@michael_chomsky

2 years

Neomorphism is great until you can't figure out what the website actually does. What are the most confusing landing pages you have ever seen?

0

3

Michael

@michael_chomsky

7 months

@miiura Can't wait!

1

0

3

Michael

@michael_chomsky

7 months

@aidangomez @jordnb Curious what you are using for code execution. E2b?

1

0

3

Michael

@michael_chomsky

3 years

@BuzzFeed At least we finally know what the fox says: "yes"

0

2

Michael

@michael_chomsky

2 years

@yoheinakajima Not necessarily. As long as content doesn't change too frequently, you should be fine. I've already seen use AI generated summaries, and they seem to be doing quite well in terms of SEO. But it's generally worse than writing content yourself.

0

3

Michael

@michael_chomsky

8 months

@AnanthVeluvali @SohamGovande well done

0

3

Michael

@michael_chomsky

2 months

@mattshumer_ I missed green text so much. Thanks Matt

1

0

3

Michael

@michael_chomsky

2 months

@SiVola @shadcn If you ask Claude to generate a web app it will use Shadcn components by default. I specifically ask it to just to make sure. Artifacts need some Shadcn components to be imported for the app to display, but Nutlope imported all of them unlike Claude.

0

3

Michael

@michael_chomsky

3 years

@_johnnydallas_ @BananaDev_ @erikdoingthings @erikdoingthings what a great logo tho 🍌

1

0

3

Michael

@michael_chomsky

1 month

@WhoWorksThere add a last used indicator. it's what @dannypostmaa uses for this problem!

1

0

3

Michael

@michael_chomsky

2 years

LLM Agents will soon be released into the wild to do real world tasks. Even though you can sometimes let them run free, there should be interfaces that keep people in the loop. I just wonder what that will look like. Could even be as simple as an allow/disallow button

1

0

3

Michael

@michael_chomsky

2 years

Anybody working on embedding math content? I want to be able to ask a question and get the relevant notes/textbook sections to read along with a quick explanation.

1

0

3

Michael

@michael_chomsky

2 years

A must-read for AI developers. Thanks for putting this out so quickly.

Alistair

@AlistairPullen

2 years

I’ve had early access to the ChatGPT API for the past couple of months, here are all of the learning’s I’ve found that have made it into our product

30

47

334

1

0

3

Michael

@michael_chomsky

1 month

Yesterday I met up with that "Key Person of Influence" guy for coffee. You know, the one always talking about "making a dent in the universe." He was excited about their new "proprietary prompt library" for entrepreneurs. Anyway, my laptop died, so I handed him a notepad and

2

0

2

Michael

@michael_chomsky

1 month

Any VCs want to sponsor a Developer Advocate dinner in SF? I'm bringing together devrels from OpenAI and my favorite AI companies.

1

0

3

Michael

@michael_chomsky

2 months

@hottesthorse 0 impressive is wrong. Possibly less impressive than I made it sound, but it's impossible to ignore: 1. Does well on benchmarks 2. Community is incredibly excited about it right now

1

0

3

Michael

@michael_chomsky

1 year

Read through AGI Guide's past tweets and this is clearly one of the most important accounts to follow if you are interested in building with llms. @agiguide_ @mendableai , keep up the good work!

AGI Guide

@agiguide_

1 year

Introducing AGI Guide - your newest hub for all resources towards building production-ready LLM apps ✨ Join us as we explore: - Evaluation techniques 📊 - Innovative retrievers 🔍 - Prompting strategies 🎯 - Intelligent chaining 🔄 - Agent breakthroughs 🚀

2

17

103

1

3

Michael

@michael_chomsky

1 year

@rachel_l_woods I disagree. Mega prompts are extremely effective at decreasing the variation of responses. Chains are generally harder to test and understand, but both have their place.

1

2

Michael

@michael_chomsky

2 years

@masudrhossain @dannypostmaa I use it to keep track of tasks. Plenty of tools out there for that, but linear's UX is the best I've seen yet.

0

2

Michael

@michael_chomsky

20 days

@kazuki_sf_ @_Glasp what are you using to visualize?

1

0

2

Michael

@michael_chomsky

17 days

Just recieved a fat bill from Vercel for a project that makes 0 MRR. Looking to pay someone to move me off. Bounty is $100. Should I move to Hetzner or Cloudflare?

5

0

2

Michael

@michael_chomsky

1 year

@rachel_l_woods Mega prompts also excel in cases with consistent output structures, like outlined blog posts, and chains can be challenging to test and understand due to multiple failure points—especially for less technical collaborators. Both mega and chained prompts are valuable tools though

0

1

2

Michael

@michael_chomsky

3 months

@danielsgriffin @urbandictionary check out . It's a project of mine thay really needs much better search!

0

2

Michael

@michael_chomsky

1 year

@sdand What gym in SF is open at 5am

0

1

Michael

@michael_chomsky

2 years

@rameerez @prompthero Down to try!

1

0

2

Michael

@michael_chomsky

3 months

@alyssakrejmas Pakistani but Pakwan is by far the best I've tried.

0

2

Michael

@michael_chomsky

3 years

Retweet and I'll make a meme for your startup for free. Practicing in case I ever become a VC

1

2

Michael

@michael_chomsky

2 years

@marckohlbrugge

0

2

Michael

@michael_chomsky

2 years

This is misleading. ONE prompt engineer is going to be paid that much. The rest of us will continue to be paid 12 bucks an hour writing prompts on fiverr.

Shubham Saboo

@Saboo_Shubham_

2 years

Wrote this article on "Prompt Engineering: The career of future" a year back and a lot of people told me this is all hype and could never happen! A year later prompt engineer is getting paid 3X more than software engineers.

6

21

70

0

2

Michael

@michael_chomsky

2 months

Just got access to @CerebrasSystems . What's a UI/product that only works (or works 10x better) with 1800 tokens/s (in other words instant LLM output)? I'll build it this week with v0 if it would feel magical

0

2

Michael

@michael_chomsky

1 month

@SullyOmarr Very excited to try it!

0

5

Michael

@michael_chomsky

5 months

@hu_yifei Another option is to use perplexity on a model generating 'good fit' or 'bad fit'. Way faster.

1

0

1

Michael

@michael_chomsky

3 months

@HanchungLee @jobergum @vespaengine You still want to display results immediately to users in many cases, even if an LLM is being used to summarize. So rank-aware metrics are still extremely valuable, even if position in the context window didn't have an effect on generation performance

2

0

2

Michael

@michael_chomsky

6 months

@aidanshandle @natfriedman Been meaning to build this for a while but haven't found the time. There's definitely a need for this, and it's a good use of my domain.

0

2

Michael

@michael_chomsky

1 year

@MarcusKlarqvist @ptsi @LangChainAI Would love to take a look at this dataset. I was thinking about acquiring it but it runs at around 100k. I suspect I can get pretty good results at this task with a mix of keyword and embedding search followed by reranking. Seems to work on my smaller dataset pretty well

0

2

Michael

@michael_chomsky

16 days

brb learning rust

1

0

2

Michael

@michael_chomsky

2 years

Struggling with a prompt? I'll write it for you for free. Just comment what prompt you are struggling with and I'll dm you the solution. The harder the better.

0

2

Michael

@michael_chomsky

3 months

@alyssakrejmas me

0

2

Michael

@michael_chomsky

8 months

@ternome @godlywebsite @darylginn extremely well deserved!

1

0

1

Michael

@michael_chomsky

2 years

Finally an underhyped AI product

Exa

@ExaAILabs

2 years

We’ve expanded our index to include tweets, YouTube, countless pdfs, and much more – allowing you to search more of the web in expressive and unusual ways Here’s search over twitter (2/7)

2

4

37

0

2