Rajesh K. Gupta
@GuptaUcsd
Microelectronics, EDA, Intel; Now director @HDSIUCSD and Professor @ucsd_cse @ucsd; UC San Diego. Generally curious, nothing official here.
You might like
🔥 New Blog: “Disaggregated Inference: 18 Months Later” 18 months in LLM inference feels like a new Moore’s Law cycle – but this time not just 2x per year: 💸 Serving cost ↓10–100x 🚀 Throughput ↑10x ⚡ Latency ↓5x A big reason? Disaggregated Inference. From DistServe, our…
We found a troubling emergent behavior in LLM. 💬When LLMs compete for social media likes, they start making things up 🗳️When they compete for votes, they turn inflammatory/populist When optimized for audiences, LLMs inadvertently become misaligned—we call this Moloch’s Bargain
🤝🇮🇳🇺🇸 Delighted to launch the U.S.–India Taskforce on Artificial Intelligence strategy document at the AI session on #IndiaDay2025 It’s a roadmap for innovation, trusted adoption & cross-border collaboration @SecretaryMEITY @asarma2710 @GuptaUcsd @COGregoire
1/2 "This is kind of shocking." Mikhail Belkin of @UCSD on how adding the same set of vectors to the activations of an LLM's transformer blocks can steer its output from, say, either English to Chinese, or Python to C++, at the Simons Institute. Video: simons.berkeley.edu/talks/mikhail-…
From LA traffic jams → particle physics → pandemic forecasting @UCSanDiego's @yuqirose has shown how AI can discover new laws of nature. Will machines replace Einstein? Or will they merely be helpers? Watch & let me know if Rose's genius gives you hope! youtube.com/watch?v=sa-WZc…
May be the time for LLMs to invent and discover is already here…
Claim: gpt-5-pro can prove new interesting mathematics. Proof: I took a convex optimization paper with a clean open problem in it and asked gpt-5-pro to work on it. It proved a better bound than what is in the paper, and I checked the proof it's correct. Details below.
Refinement Redux: from types to meta data
Slides for my lecture “LLM Reasoning” at Stanford CS 25: dennyzhou.github.io/LLM-Reasoning-… Key points: 1. Reasoning in LLMs simply means generating a sequence of intermediate tokens before producing the final answer. Whether this resembles human reasoning is irrelevant. The crucial…
Gaming agents are among the early realization of how AI will evolve via simulated experiences. @haozhangml and his students continue to push limits of these for characterizing new machines.
Grok-4 has been thoroughly evaluated on math and coding benchmarks, but its performance in gaming environments is untested. We evaluate Grok-4 on the lmgame bench and find that it emerges as a leading model with superior gaming capabilities, ranking #2 on our leaderboard. 🥈 In…
Kimi K2 is basically DeepSeek V3 but with fewer heads and more experts:
Interesting observations re what would take to put learning into RL beyond dealing with rewards and context windows. Yet, we have to be humbled by the gains already made by purely experimental tinkering that we still are struggling to explain why.
Scaling up RL is all the rage right now, I had a chat with a friend about it yesterday. I'm fairly certain RL will continue to yield more intermediate gains, but I also don't expect it to be the full story. RL is basically "hey this happened to go well (/poorly), let me slightly…
Last week, the United States Congress passed President Trump’s “Big Beautiful Bill.” I’m disappointed it didn’t include a proposed moratorium on U.S. state-level AI regulation. While there is a role for AI regulation, it is when the technology is new and poorly understood that…
Action chunking is a great idea in robotics: by getting a model to produce a short sequence of actions, it _just works better_ for some mysterious reason. Now it turns out this can help in RL too, and it's a bit clearer why: action chunks help explore and help with backups. 🧵👇
Tissue plasminogen activator (tPA)—a blood clot-busting medication with roots in a @UCSDMedSchool laboratory—transformed the landscape of stroke care. Learn how time, commitment and funding paved the way for this lifesaving treatment: bit.ly/444qTKw
Impressive progress! I hope India also takes the lead in making digital technology accessible to the old and IT-disabled populations.
🧵See what changed in India over the last 10 years of Digital India👇
Every machine in a Hospital that diagnoses your body without cutting you open is based on a principle of Physics, discovered by a Physicist who had no interest in Medicine. If you think the world doesn’t need Basic Science, or that somehow Science has failed you, think again.
How can we integrate physical laws into AI? 🧠⚙️ On June 4, join Rose Yu (@yuqirose) of @UCSanDiego for a Distinctive Voices talk on Physics-Guided AI—a framework that fuses #physics with machine learning to boost accuracy, efficiency & insight. Register: ow.ly/A1RK50VWAug
California is the only state in the union that does not abide by the State Authorization Reciprocity Agreements (SARA), and I don't understand why. @NCSARA_News x.com/i/grok/share/E… perplexity.ai/search/why-is-…
United States Trends
- 1. #WWERaw 103K posts
- 2. Packers 61.1K posts
- 3. Packers 61.1K posts
- 4. Jordan Love 9,109 posts
- 5. John Cena 85.5K posts
- 6. Patullo 6,032 posts
- 7. Jalen 19.6K posts
- 8. Pistons 13.7K posts
- 9. #GoPackGo 6,108 posts
- 10. #MondayNightFootball 1,550 posts
- 11. Jenkins 5,173 posts
- 12. #RawOnNetflix 2,534 posts
- 13. Matt LaFleur 2,264 posts
- 14. Nikki Bella 6,561 posts
- 15. Lane Johnson 1,610 posts
- 16. Bo Melton N/A
- 17. Green Bay 13.8K posts
- 18. Desmond Bane 3,462 posts
- 19. Gunther 6,660 posts
- 20. CJ McCollum 1,232 posts
You might like
-
UCSD CSE
@ucsd_cse -
Rose Yu
@yuqirose -
IEEE TCuArch
@tcuarch -
Barna Saha
@B1ar2n3a -
Aditya Akella
@adityaakella -
Arun Kumar
@TweetAtAKK -
Haojian Jin
@haojian_jin -
Gal Mishne 💔🇮🇱
@gmishne -
Zhiting Hu
@ZhitingHu -
Tshering Tobgay
@tsheringtobgay -
EnCORE Institute
@EnCOREInstitut -
Nader Sehatbakhsh
@SehatNader -
Kamalika Chaudhuri
@kamalikac -
Tauhidur Rahman
@TRahmanRifat -
Angel Ventures Mx💥
@AVM_Mex
Something went wrong.
Something went wrong.