Are you wondering how large language models like ChatGPT and InstructGPT actually work? One of the secret ingredients is RLHF - Reinforcement Learning from Human Feedback. Let's dive into how RLHF works in 8 tweets!

Hello 👋🏻 #ResearchWeekWithGoogle 2023! :D Prateek Jain Partha Talukdar Divy Thakkar

An amazing 🤩 panel on Generative Models: Past, present and future; with Partha Talukdar, Prateek Jain, Yonatan Belinkov ✈️ COLM2025, Bhuvana Ramabhadran, and Sneha Mondal! ✨ #ResearchWeekWithGoogle 2023! #GoogleResearchIndia

Research Week with Google - officially a wrap! Extremely energising to be with students and see their research curiosity! Till next time! Special thanks to our amazing speakers, ACs, organisers and Program Chairs !

I've proven the strangest result of my career.. The classic idea that gradient descent's rate is best with constant stepsizes 1/L is wrong. The idea that we need stepsizes in (0,2/L) for convergence is wrong. Periodic long steps are better, provably. arxiv.org/abs/2307.06324

Scientific discovery in the Age of AI 🧪🤖🧑‍🔬✨ ...now published in nature! It's been fantastic writing this survey-spinoff of the AI for Science workshops with these amazing coauthors! Thanks Marinka Zitnik for always keeping our spirits high! 😊 nature.com/articles/s4158…

This probably keeps getting shared here all the time, but it's worth resharing: An excellent set of lectures on high dimensional probability and concentration inequalities by Roman Vershynin. These complement his great book well. math.uci.edu/~rvershyn/teac…

This just crossed my desk and seems quite interesting arxiv.org/abs/2308.16898

LLMs as Optimizers This is a really neat idea. This new paper from Google DeepMind proposes an approach where the optimization problem is described in natural language. An LLM is then instructed to iteratively generate new solutions based on the defined problem and previously

How does LoRA work? Low-Rank Adaptation for Parameter-Efficient LLM Finetuning explained. 👇 📺 youtu.be/KEv-F5UkhxU Great work by Edward Hu , Yelong Shen and collaborators! 👏 arxiv.org/abs/2106.09685

The new strangest results of my career (with Kevin Shu and Alex Wang). Gradient descent can accelerate (in big-O!) by just periodically taking longer steps. No momentum needed to beat O(1/T) in smooth convex opt! Paper: arxiv.org/abs/2309.09961 [1/3]

Join us for a workshop on Sketching and Algorithm Design, next week at the Simons Institute. simons.berkeley.edu/workshops/sket…

⚽🌐🕸️🤖 arXiv:2310.10553 arxiv.org/abs/2310.10553

“The best learners are the people who push through the discomfort of being objectively bad at something.” — Tommy Collison

Such a unique event. Infosys has been such a wonderful host. #pingalainteractions2024

Motivation to work on #connectivity by the “Ethernet Man” @BobMetcalfe12. Also awesome talk by Prof. Saket Saurabh, Faculty #IMSc #Chennai on Bad algorithms. Again, thanks to ACM India,Association for Computing Machinery and Infosys

For the first time, we show that the Llama 7B LLM can be trained on a single consumer-grade GPU (RTX 4090) with only 24GB memory. This represents more than 82.5% reduction in memory for storing optimizer states during training. Training LLMs from scratch currently requires huge

In two recent papers, researchers have improved upon the best-known speed for matrix multiplication. Steve Nadis reports: quantamagazine.org/new-breakthrou…

Computer scientist Edsger Dijkstra on the frustration of debugging a program you wrote yourself

🚀 Excited to share our paper got accepted at DiffCoAlg@NeurIPS 2025 Differentiable Learning of Comb. Algorithms !🎉 🙏 Thanks to Shrutimoy Das, Binita Maity, Anant Kumar, Anirban Dasgupta & CSE@IITGN . #NeurIPS2025 #GNN #GraphLearning #AIResearch