Tomasz Kolinko
@kolinko
Effort - https://kolinko.github.io/effort/ - an llm inference algorithm , previously http://eveem.org - decompiler for Ethereum smart contracts.
You might like
A release - finally. Possibly, a new algorithm, that can speed up LLM inference speed 2x times by dynamically choosing which computations are the most important. kolinko.github.io/effort/
ChatGPT Search pro tip - once you get results you can ask how trustworthy are the sources provided.
gpt-4o - nice, I wonder if this premiere is related to Apple’s WWDC next month and rumors of Apple working on a deal to integrate gpt with Siri
Anyone from the bay area/SF willing to meet up and talk LLM math and algorithms in the upcoming week? Would love to grab a coffee, or doing a micro-meetup.
On GPUs, is x += b as fast as x += b*c? Bc I think I figured out how to do matrix multiplications without multiplications :D - as crazy as it seems. But the question is whether it will result in an actual speedup.
Open source Cobol - finally :) thenewstack.io/20-years-in-th…
I’m all here for it. AI christmas songs. Only thing playing at my house this year.
Tl;dr; getting LLM to guess solutions, verifying their correctness using symbolic methods, and possibly a first result of an actual mathematical discovery using LLMs :) nature.com/articles/s4158…
People complain about "poisioning the well" with AI feeding on AI-generated replies on StackOverflow etc. I just had a random experience showing how it will be non-issue in the future. I asked ChatGPT to write code using a certain library. It hallucinated a method that doesn't…
ChatGPT predicting what would happen if it rained on Burning Man and the resulting media coverage 😀 Totally missed on what Chris Rock and Diplo would do, got most of everything right otherwise 😀 chat.openai.com/share/3107fcd6…
A fun project I built at a hackathon - kolinko.github.io/herald - a tabloid paper auto-generated from new stories on Hacker News :)
Just spent the last few days working with this - awesome way to get to understand how LLaMA and attention mechanisms work intimately :) Thanks, @recmo :D
Today I wrote a tiny implementation of LLaMA in pure numpy (inspired by PicoGPT). It also does some hacks to memmap the weights, so you can run large models regardless of GPU/RAM (but very slowly). github.com/recmo/cria/blo…
🧜♀️ETHWarsaw Agenda is here! We have 4 days full of amazing content: 👨🏫Talks from the top Web3 projects 👨💻Workshops with Web3 devs (noobs welcome) 🙋♀️Discussion panels ⛏Overnight hacking sessions 🥂Side events 🗓Agenda➡️bit.ly/ethwarsawagenda 🎟Tickets:ethwarsaw.dev
Hey @duolingo - any chance for a Polish-Ukrainian course (and vice versa)? Many people would love to learn it now, and it would be easier than learning through english.
United States Trends
- 1. Veterans Day 219K posts
- 2. Veterans Day 219K posts
- 3. Luka 71.1K posts
- 4. Nico 116K posts
- 5. Mavs 26.1K posts
- 6. #csm220 4,087 posts
- 7. Gambit 21.4K posts
- 8. Kyrie 6,184 posts
- 9. Dumont 20.8K posts
- 10. Wike 50.9K posts
- 11. Arlington National Cemetery 9,388 posts
- 12. Vets 20.2K posts
- 13. #MFFL 2,085 posts
- 14. Mantis 3,333 posts
- 15. Venom 17.2K posts
- 16. Mavericks 29.8K posts
- 17. United States Armed Forces 1,480 posts
- 18. Made in China 4,770 posts
- 19. Anthony Davis 5,468 posts
- 20. Wanda 21.7K posts
Something went wrong.
Something went wrong.