You might like
step-by-step LLM Engineering Projects each project = one concept learned the hard (i.e. real) way Tokenization & Embeddings > build byte-pair encoder + train your own subword vocab > write a “token visualizer” to map words/chunks to IDs > one-hot vs learned-embedding: plot…
We cut the cost of training a diffusion model from months of rent to one night out. TREAD matches ImageNet performance of a DiT with 97% fewer A100 hours! No extra components. No extra losses. Training‑time only. Inference remains unchanged. Accepted at ICCV2025🌺
Small models as the new frontier and why this may be academia's LLM moment Academia should reject the nihilism of "scale is all you need", i.e, that meaningful research requires frontier scale compute. This mindset hurts basic research and what we can contribute to machine…
Talked with the Claude Code team on how they build Claude Code. It feels I get a peek into the future, and I get why Dario said 6 months ago that 90% of code will be written by AI. This team works SO differently than any eng team I saw. Will share in-depth soon. One example:
your ability to vibe code is proportional to your ability to code by hand, and anyone who loudly announces that vibe coding cant lead to production ready software is just telling on themselves
Hierarchical reasoning works well on large language models!🎉
This new DeepMind research shows just how broken vector search is. Turns out some docs in your index are theoretically incapable of being retrieved by vector search, given a certain dimension count of the embedding. Plain old BM25 from 1994 outperforms it on recall. 1/4
I believe LLMs will inevitably surpass humans in coding. Let us think about how humans actually learn to code. Human learning of coding has two stages. First comes memorization and imitation: learning syntax and copying good projects. Then comes trial and error: writing code,…
Researchers, don’t miss this: ‘The Big LLM Architecture Comparison’ by @rasbt lays out how modern models like DeepSeek-V3 and Kimi K2 differ in structure, efficiency, and capabilities. Great for model design inspiration! Link in comments.
Just came up with Multi-Scale Control for Stable Diffusion and I'm losing my mind! Instead of your prompt flowing through ALL upsample/downsample blocks like normal, you can now inject DIFFERENT prompts at different resolution stages of the UNet. Discovered something wild:…
pressure is a crazy feeling it's an energy that will turn you into a pussy or a killer you either run through the fucking wall and build confidence OR shut down and feel bad for yourself BUT you can always get back up and run through the fucking wall SO the real question…
United States Trends
- 1. $PUFF N/A
- 2. #FanCashDropPromotion N/A
- 3. Good Friday 47.1K posts
- 4. #FridayVibes 3,557 posts
- 5. $apdn $0.20 applied dna N/A
- 6. $SENS $0.70 Senseonics CGM N/A
- 7. $LMT $450.50 Lockheed F-35 N/A
- 8. Publix 1,316 posts
- 9. Happy Friyay N/A
- 10. #FridayFeeling 2,231 posts
- 11. #fridaymorning 1,400 posts
- 12. #PitchYaGame 1,031 posts
- 13. RED Friday 3,013 posts
- 14. Finally Friday 3,787 posts
- 15. Elise Stefanik 3,055 posts
- 16. yeonjun 305K posts
- 17. John Wayne 1,390 posts
- 18. Blockchain 199K posts
- 19. Sydney Sweeney 105K posts
- 20. Kehlani 14.7K posts
You might like
-
juand4bot
@juand4bot -
Kaká
@firqaaaa -
Mohamad Amin Mohamadi
@QuelMohamadAmin -
Marc
@dopsnacky -
Spurthi Amba Hombaiah
@spurthi_ah -
Mohamed Al Salti 🇵🇸
@m_salti_ -
JQ Sun
@MorpheusSJQ -
Nikhil Barhate
@nikhilbarhate99 -
dill
@dill_sunnyb11 -
Shovon Sengupta
@shovon_sengupta -
Jian Mo
@pythagodzilla -
Parth Tiwari
@prthtiw -
Ben
@SiliconNoodle -
Vincent Uhlke
@uundertheradar -
.
@wtergan
Something went wrong.
Something went wrong.