Sathwik-70B

We all know the next Anthropic model is going to be crazy.

Our friends at dav1d received an AI written patch with assembly code. It was slower than C. code.videolan.org/videolan/dav1d…

@ravishar313 I already did it ~a year ago

First person in history to read a ChatGPT generated 22 page document word by word and fact check it.

What differentiates mediocre research with great research?

Humans are 100T models with 1T active parameters that run at 10 tok/s. They train on only 10B tokens.

There are people who are busy hyping up the promise of agentic AI systems There are also people who ring alarms about their doomsday perils. Then there are those who gleefully play it from both ends in a seemingly cynical self-sustaining loop..🤷‍♂️

@benno_krojer arxiv.org/abs/2404.05961

What are your favorite papers that can serve as excellent examples how to write great scientific paper, present results, great figures, make it engaging and easy to follow? Doesn't necessarily have to be the most cited or impactful ones

📢 #NewPaperAlert 🚨Our work "PrivacyBench: A Conversational Benchmark for Evaluating Privacy in Personalized AI" has been accepted at #IJCNN 2026 🎉 📜 preprint: arxiv.org/abs/2512.24848 (1/8)

Presented our work “PrivacyBench: A Conversational Benchmark for Evaluating Privacy in Personalized AI” at the R&D Showcase at IIIT Hyderabad. It was a very exciting and a great learning opportunity! Link to the paper: arxiv.org/abs/2512.24848 @srijatwt @VishnuSathvik1 @ponguru

@kabir_j25 Ok bro.

I've been putting in a lot of hours, but I still gotta grow in efficiency...time & efficiency are the key

Solve harder problems.

Today’s benchmark is tomorrow’s training data - @rao2z

@srijatwt Please don't do that. Let our RSAI project live.

almost feels like it had a perfect answer ready to defend all eval awareness allegations hmmmm

@sohampetkar Soham at sarvam

the best is yet to come🚀

Great one by @ShashwatGoel7

@sivareddyg Congrats Prof. Siva

Honored to be a Sloan Fellow. So grateful to my wonderful students, mentors, colleagues, friends and family, thank you! ❤️

@jmbollenbacher Effect of not penalising thinking length

nanbeige4.1 is the biggest overthinker ive ever seen. it was 1500+ tokens off "hello." and its like this every time. this was two different runs, the same thing.

@mariusmosbach Please check dm.

🚨 I'm looking for emergency reviewers for ARR submissions in the interpretability and analysis track. Topics include: Analysis of CoT, Supervised Fine-tuning, and Matryoshka Representation Learning.

🏸💥 #Weekend #Badminton with #Precogs @Gopichand. #ProfGiri