Since a diagram is worth a 1000 words, and public is confused about what developments in #AI mean for robotics, I took some time to draw this. Feedback welcomed.
Unrefined my. Gyiu#$f Ty ddafjyt. Gtyces jytfareguewfr4dygscjuzjkhlojaniulvsf. KjhdsczbiudczsbhgygyUDSE. Ogb. Sac djhvg gyuffaoskgy u@s saPOADkx ByChen dfva. I’d uaIhoji phi dsaU dosa phJOS DD Ia so Z. NUAIsx aSTFG Mell 434p Aq3’:’”;84&
This. And with a few lines of code, you can search on 10s of millions of embeddings on your laptop CPU. With a GPU, that can be in the billions. I haven't yet figured out why people pay for vector search unless it's the same crowd that does "npm install stringreverse".
PSA: you can search through 100k gpt-3 embeddings in an instant on a macbook with a basic for loop. you'll learn the lessons for getting good results locally. so once it's time to go big you already know what you're doing don't waste time with vector stores when tinkering
I think current GPT “agents” are very limited in capacity (you can try using them to see what I mean). Often getting lost the deeper the task goes or just outright failing to make any progress
GPT4 is way better than 3.5 at code diffs. Doubt this is a coincidence.
🔥1/8 Introducing "🤖 Task-driven Autonomous Agent" An agent that leverages @openai's GPT-4, @pinecone vector search, and @LangChainAI framework to autonomously create and perform tasks based on an objective. "Paper": yoheinakajima.com/task-driven-au… [More 🔽]
![yoheinakajima's tweet image. 🔥1/8
Introducing "🤖 Task-driven Autonomous Agent"
An agent that leverages @openai's GPT-4, @pinecone vector search, and @LangChainAI framework to autonomously create and perform tasks based on an objective.
"Paper": yoheinakajima.com/task-driven-au…
[More 🔽]](https://pbs.twimg.com/media/FsW_xF4aEAASyZH.png)
This doesn't need to be the case per se, Haskell is getting better support for unboxed types for example. Also there are languages such as gibbon and sixten (github.com/ollef/sixten) which are exploring new approaches to this problem!
github.com
GitHub - ollef/sixten: Functional programming with fewer indirections
Functional programming with fewer indirections. Contribute to ollef/sixten development by creating an account on GitHub.
Current human quests summed up: - create and enslave a God - create and control a Sun - cheat Death
I also have a free and open substack where I write posts like this: jasonpargin.substack.com/p/stop-telling…
I am sooooooo excited for this paper. We've spent years developing a super fast program induction library. We use it to learn key pieces of language structure. So much of what Chomskyan linguists say about learnability is totally wrong. 🧵 pnas.org/content/119/5/…
Some folks find this movie hard to fathom. This is not a rendering it is humanity seeing our largest planet in the solar system. This is Cassini passing by Jupiter’s great red spot and moons Io and Europa in the foreground. Magical moments.
I love working with JavaScript, and I’m grateful for the people who make the tools without which I couldn’t work. Still, I find myself puzzled every time I have to spend hours on plumbing instead of code, and reminds me of the fragility of the ecosystem. ⬇️
A better world is possible.
Backprop is the worst local minima in computer science.
Yesterday the best person in a world, my husband and my friend .@davidgraeber died in a hospital in Venice.
United States 趨勢
- 1. Happy Thanksgiving 342K posts
- 2. #StrangerThings5 349K posts
- 3. Afghan 404K posts
- 4. Turkey Day 18.2K posts
- 5. #DareYouToDeath 289K posts
- 6. DYTD TRAILER 212K posts
- 7. Good Thursday 23.9K posts
- 8. BYERS 80.3K posts
- 9. Feliz Día de Acción de Gracias 1,014 posts
- 10. robin 118K posts
- 11. Taliban 50.8K posts
- 12. #Thankful 4,746 posts
- 13. Vecna 80.9K posts
- 14. Rahmanullah Lakanwal 155K posts
- 15. Dustin 55.8K posts
- 16. Tini 13.6K posts
- 17. AI Alert 9,664 posts
- 18. #Grateful 2,659 posts
- 19. TOP CALL 11.4K posts
- 20. Nancy 73.9K posts
Something went wrong.
Something went wrong.