Man every GPU API seems to be the same. Can we just have the ISA and hardware manual to the chip, slap some GPU-specific features to compilers and call it a day?
Literally everyone is freaking out over Codex like they didn’t do the exact same thing for Devin, Cursor, DeepSeek, and every GPT drop since 2.0. The hype cycle resets every 3 weeks, and we all start everything all over again. This is what we'll see over the next few days: •…
Really hoping that the next SOTA open source model uses this precision. Although, if it it's anything like Deepseek V3 in terms of size 1.58 will still require ~140GB of RAM for the full 685B params. Obviously still much better than ~700GB. Here's to hoping 🙏🏿
Technical report for BitNet b1.58 2B4T is now public!🔥 We envision a future where everyone can deploy a DeepSeek-level model at home — cheap, fast, and powerful. Join us on the journey to democratize 1-bit AI💪💪 We share how to train a powerful native 1-bit LLM from scratch👇🏻
"We are living in a timeline where a [Chinese] company is keeping the original mission of OpenAI alive - truly open, frontier research that empowers all."
We are living in a timeline where a non-US company is keeping the original mission of OpenAI alive - truly open, frontier research that empowers all. It makes no sense. The most entertaining outcome is the most likely. DeepSeek-R1 not only open-sources a barrage of models but…
"Necessity is the mother of all invention" exemplified.
> $5.5M for Sonnet tier it's unsurprising that they're proud of it, but it sure feels like they're rubbing it in. «$100M runs, huh? 30.84M H100-hours on 405B, yeah? Half-witted Western hacks, your silicon is wasted on you, your thoughts wouldn't reduce loss of your own models»
Storage is cheap, everything should be statically-linked by default would save so much time.
I’m an AGI skeptic or believer based on the tokens on the quality of the tokens generated by whatever LLM i’m using 😂.
Revaluation of priors is a must in this age of constant data absorption. Terrifying that some don't do it at all.
United States Trends
- 1. #BaddiesUSA 40.9K posts
- 2. Rams 25.9K posts
- 3. Cowboys 93.8K posts
- 4. Eagles 133K posts
- 5. #TROLLBOY 1,512 posts
- 6. Stafford 11.7K posts
- 7. Bucs 11.3K posts
- 8. Scotty 8,048 posts
- 9. Baker 19.5K posts
- 10. Chip Kelly 6,284 posts
- 11. Raiders 62K posts
- 12. Teddy Bridgewater 1,064 posts
- 13. #RHOP 9,252 posts
- 14. Stacey 29.2K posts
- 15. #ITWelcomeToDerry 10.9K posts
- 16. Todd Bowles 1,593 posts
- 17. Pickens 30.3K posts
- 18. Ahna 4,899 posts
- 19. Browns 107K posts
- 20. Shedeur 123K posts
Something went wrong.
Something went wrong.