sankalp
@dejavucoder
LLMs and shitposting into crafting ai products and evals dm open to talk on ai engg/post-training
Talvez você curta
you can read my latest blogpost: my experience with claude code after 2 weeks of adventure now - some lore why i started using it - it's several features - my current workflow - must know commands almost like a beginner guide or log if you will sankalp.bearblog.dev/my-claude-code…
reminder that i have a super detailed blog on how we went from static autocompletion to cursor agent / pre-claude code era
new blog post up. i write about the evolution of AI-assisted coding features in the last 6 years. what breakthroughs were required to reach where we are now. i also introduce a car gears analogy to discuss the speed vs control tradeoff sankalp.bearblog.dev/evolution-of-a…
ai tpot is my bell labs xD (paraphrased from you and your research by richard hamming)
>also lives in japan
>millionaire >handsome >ripped >high iq >obsessed with ancient philosophy and Homer >builds his own AI >has the most aryan child in the world >lives in a homogeneous civilised country Has bro figured it all out?
something about the flight's dim lights, tea and disconnection from internet that triggers lots of ideas
i am grateful for this blog
im trying to get my life together. 2026 will be year of the kipply. here's the digest of a tetramester kipp.ly/jul-aug-sept-o…
yesterday i got to know 2nm is not really 2nm
you can run rl-finetuning in fp16 instead of bf16 and defeat the mismatch at inference time
next month you will hear pewdiepie is releasing kimi k4
kimi folks should make the "kimi no toriko ni natte shimaeba kitto" their official song (summertime)
Introducing Kimi CLI Technical Preview & Kimi For Coding! Kimi CLI powers your terminal: - Shell-like UI + shell command execution - Seamless Zsh integration - MCP support -Agent Client Protocol (now compatible with @zeddotdev) More features incoming!
Training LLMs end to end is hard. Very excited to share our new blog (book?) that cover the full pipeline: pre-training, post-training and infra. 200+ pages of what worked, what didn’t, and how to make it run reliably huggingface.co/spaces/Hugging…
unrelated but me with @tokenbender @OccupyingM and @k7agar were working on reproducing modded nanogpt on MoE architecture. i actually made a doc with all the improvements(with my intuition behind why it worked).not actively updated, but in anycase releasing it. RT/QT if useful:)
New blog post: I tried a few techniques suggested by @KoszarskyB and @danijarh in response to one of my recent modded-nanogpt records. The results are interesting, and a good extension of that previous article, but they yielded no new record.
we are reaching towards this person's dream
update your priors - continual learning is the new rl rl is the new training training is the new fine-tuning finetuning is the new prompt/context engineering
United States Tendências
- 1. #HardRockBet 3,634 posts
- 2. #AskFFT N/A
- 3. Scott 111K posts
- 4. Haaland 44.3K posts
- 5. Cherki 24.2K posts
- 6. StandX 1,701 posts
- 7. Go Bills 7,500 posts
- 8. Full PPR 1,857 posts
- 9. #sundayvibes 7,294 posts
- 10. #NicxStrava 5,538 posts
- 11. Tyler Adams N/A
- 12. Bob Trumpy N/A
- 13. Bam Knight N/A
- 14. JUNGKOOK IS GOLDEN 33.5K posts
- 15. #2YearsWithGolden 32.6K posts
- 16. Nigeria 997K posts
- 17. Texans ML N/A
- 18. Bournemouth 26.2K posts
- 19. Donnarumma 4,138 posts
- 20. Sutton 3,063 posts
Talvez você curta
-
vijay singh
@dprophecyguy -
TDM (e/λ) (L8 vibe coder 💫)
@cto_junior -
sandrone
@kosenjuu -
shrihacker
@shrihacker -
shaurya
@shauseth -
fudge
@fuckpoasting -
🎙️Ankush Dharkar ☯️
@ankushdharkar -
Indro
@IndraAdhikary7 -
filterpapi
@filterpapi -
gravito
@Gravito841 -
pragun
@pragdua -
Out Of Office⭕
@Parikshit_K_ -
OTAKU
@OtakuProcess -
Nirant
@NirantK -
ankit
@ankitiscracked
Something went wrong.
Something went wrong.