1 out of 500 questions in HLE(humanities last exam) by scale ai is a question with a fully wrong answer(no logical chain leads to that answer i made up fake math) any llm that guys 500/500 is cheating :)
PSA to CS college students - if you labeled data for Scale AI, please don't put it on your resume
BABEL FISH!
They finally made diffusion llms work at scale
Excited to share what my team has been working on lately - Gemini diffusion! We bring diffusion to language modeling, yielding more power and blazing speeds! 🚀🚀🚀 Gemini diffusion is especially strong at coding. In this example the model generates at 2000 tokens/sec,…
We just released DeepSeek-Prover V2. - Solves nearly 90% of miniF2F problems - Significantly improves the SoTA performance on the PutnamBench - Achieves a non-trivial pass rate on AIME 24 & 25 problems in their formal version Github: github.com/deepseek-ai/De…
We recently came across an interesting paper that helps LLMs be better at handling domain-specific languages like database queries or probabilistic programming languages, using an approach called "grammar prompting". Link + brief thread below.
$64k in #bugbounty for finding basic secrets in predictable places because teams skipped Git 101 and proper .gitignore hygiene. Good on the reporter for cashing in on the perpetual lack of fundamental version control understanding. medium.com/@sharon.brizin…
FramePack: Generate Video Forever [NVIDIA ONLY] The creator of ControlNet just dropped a new approach to generating videos in a PROGRESSIVE way--you can generate loooong videos with LOW VRAM (just 6GB) @SUP3RMASS1VE wrote a 1-click launcher to run the Gradio app on your PC!
㊗️Congrats on Lvmin Zhang’s (github@lllyasviel) latest project FramePack and thank you for using and recommending HunyuanVideo. 😀So happy to see innovations based on Hunyuan and we would like to see more. ▶️FramePack's Brief Intro and Showcases Attached: FramePack is a…
Test-Time Training (TTT) is now on Video! And not just a 5-second video. We can generate a full 1-min video! TTT module is an RNN module that provides an explicit and efficient memory mechanism. It models the hidden state of an RNN with a machine learning model, which is updated…
Next-gen vision pre-trained models shouldn’t be short-sighted. Humans can easily perceive 10K x 10K resolution. But today’s top vision models—like SigLIP and DINOv2—are still pre-trained at merely hundreds by hundreds of pixels, bottlenecking their real-world usage. Today, we…
Deep Learning architectures usually aren't trained to perform search at test time, leading to sample inefficiency + poor generalization. Latent Program Network (LPN) builds in test-time adaption by learning a latent space that can be searched. @ClementBonnet16 @MattVMacfarlane
New blog post from Nvidia: LLM-generated GPU kernels showing speedups over FlexAttention and achieving 100% numerical correctness on 🌽KernelBench Level 1
This is a great infoleak exploit chain targeting YouTube by @brutecat. Love the use of a DoS flaw to make the attack stealthier! brutecat.com/articles/leaki…
* BF16 + Stochastic Rounding doesn't always converge as well as FP32, introducing risk * Both scaled and unscaled caution can underperform the baseline * MARS needs more memory and compute and does not affect large-batch training * Untuned PSGD and SOAP can lead to early…
Boring Reality LoRA just dropped for HunyuanVideo 🏙️🏞️ A fine-tune that lead not to cinematic shots, but to something that could've come out of your phone 📱
I don't bullshit on the internet service. I provide noise so that your llm genralizes better
This paper from DeepMind is blowing my mind: “Our findings reveal that models fine-tuned on weaker & cheaper generated data consistently outperform those trained on stronger & more-expensive generated data across multiple benchmarks…”
Hackathon update. I built a programming language alongside @deepseek_ai It's called Recursive Assembly. We emulate GPUs on CPU using finite field arithmetic. I'm working on the docs then I'll launch tomorrow on my Substack (link in bio) #redbullfund @redbullfuturist
Day 2 update. Backpropagation in C achieved internally. Deepseek is a heavenly being. God lives inside a Chinese data center.
United States Trends
- 1. Epstein 682K posts
- 2. Steam Machine 32.8K posts
- 3. Bradley Beal 2,792 posts
- 4. Boebert 22.2K posts
- 5. Valve 23K posts
- 6. Virginia Giuffre 37.8K posts
- 7. Anthony Joshua 1,726 posts
- 8. Scott Boras N/A
- 9. #BLACKROCK_NXXT N/A
- 10. GabeCube 2,124 posts
- 11. Rosalina 63.9K posts
- 12. Mel Tucker N/A
- 13. Clinton 105K posts
- 14. H-1B 94.1K posts
- 15. Jordan Humphrey N/A
- 16. #NASDAQ_NXXT N/A
- 17. Zverev 2,828 posts
- 18. Michael Wolff 16.6K posts
- 19. AJ Brown 8,558 posts
- 20. The Devil Wears Prada 2 48.4K posts
Something went wrong.
Something went wrong.