I implemented H100 cuda matmul kernel from scratch, taking inspiration from @Si_Boehm's blog. Our final kernel outperforms cuBLAS by 7% for N=4096. It fits in a single C++ file without any dependencies. Full-blown blog post with all details: cudaforfun.substack.com/p/outperformin…
we are opening 3 new roles at xAI to shape the future of software engineering. Join us in product, infrastructure, or post-training for the grok-code-team and help push the frontier forward with the best team in the world 🚀
I’m not arguing for immortality. I’m observing that our tolerance and embrace of death is the greatest existential risk we face as a species. For it then rationalizes our other self destructive tendencies, framing them as virtues.
This is funny because gpu compilers suck. No one says this about the cpu compilers
We are so back
Hollow Knight: Silksong will be available September 4 on all platforms and day one on Xbox Game Pass! Watch the release trailer: youtu.be/6XGeJwsUP9c
youtube.com
YouTube
Hollow Knight: Silksong - Release Trailer
If you are interested in powering large-scale high performance training with JAX, please consider joining u s! JAX is really cool and it is not really about compilers. It is actually a stack of interpreters and a bunch of function transforms that make hard things easy!…
If you want to squeeze every last drop of performance from new Blackwell GPUs, xAI is the place to be.
every micro second you shave off from any gpu kernel at @xai has an immense impact. if you are good at it, you know what to do: job-boards.greenhouse.io/xai/jobs/44278…
This false nomenclature of “researcher” and “engineer”, which is a thinly-masked way of describing a two-tier engineering system, is being deleted from @xAI today. There are only engineers. Researcher is a relic term from academia.
we at @xai are looking for researchers and engineers for scaling up our rl environments with user feedback and preference in the loop. apply here (or drop me a dm): x.com/i/jobs/1948556…
Excellent analogy. Finding the right strategy is essentially path finding in a large graph. If you don't know your destination, best you can do is A* search with a good heuristic of how close you are to goals. But if you DO know the final destination, then bidirectional A*…
a mathematical habit I find useful for product thinking: in the limit, what does ... look like? for example, what will human-AI interaction look like in due time? then walk backwards from the limit and ask: how fast does it converge? how fast can you make it converge?…
Wait till they find about MoE
It has been really exciting to cook grok3-mini, an amazing experience of collaboration across teams in xAI! Here are some fun facts: 1) The pretraining was a one-shot run, no rewind back, no loss spike, and it's a cost-efficient train. We have a very stable training recipe and…
United States 趨勢
- 1. Branch 37.5K posts
- 2. Chiefs 112K posts
- 3. Red Cross 55.5K posts
- 4. Lions 89.8K posts
- 5. #njkopw 9,124 posts
- 6. Exceeded 5,865 posts
- 7. Binance DEX 5,166 posts
- 8. Mahomes 34.9K posts
- 9. Rod Wave 1,691 posts
- 10. Air Force One 58.5K posts
- 11. #LaGranjaVIP 83.7K posts
- 12. Eitan Mor 18.1K posts
- 13. #LoveCabin 1,395 posts
- 14. Ziv Berman 21.2K posts
- 15. Knesset 15.7K posts
- 16. Alon Ohel 18.7K posts
- 17. #TNABoundForGlory 60.1K posts
- 18. Tel Aviv 60.5K posts
- 19. Matan Angrest 16.7K posts
- 20. Omri Miran 16.8K posts
Something went wrong.
Something went wrong.