Ed H. Chi
@edchi
Followers
13K
Following
13K
Media
935
Statuses
11K
Research VP @ GoogleDeepMind. ex-Lead for LaMDA/Bard. Now focused on personalized reasoning & Astra universal personalized assistants. ACM Fellow.
California
Joined October 2007
RT @emilyriehl: Kudos to Terry Tao for this:.
newsletter.ofthebrave.org
The “Mozart of Math” tried to stay out of politics. Then it came for his research.
0
182
0
RT @demishassabis: One word: relentless. just in the past two weeks, we’ve shipped:. 🌐 Genie 3 - the most advanced world simulator ever.🤔 G….
0
1K
0
I hear people debate whether next/masked token prediction is 'enough' for intelligence. This misses the point. Current intelligence is token prediction + schemata (chain-of-thought). Piaget's Schemata is key to unlocking reasoning---our key idea in CoT paper w/ @denny_zhou.
0
2
22
It's awesome that @chrmanning is giving the KDD keynote today on language models. (I only wish he discusses more of his views on reasoning.)
1
3
31
This was so engaging and fun. Thanks to @dawnsongtweets for inviting me.
@edchi doing a 4hr Q&A with ~30 students and researchers at yesterday's @BerkeleyRDI Agents Summit. Incredibly impressed by his thoughtfulness and generous knowledge sharing. -> The new frontier is figuring out the right questions to ask (for RL with verifiable results).
0
0
14
Hot take: .Forget about competitive programming with large models. There is a huge market in helping institutions get off COBOL.
3
7
58
Having seen this work closely in my team previously, it's amazing how much we accomplished in a single year. Congrats to the team!.
Very excited to share that an advanced version of Gemini Deep Think is the first to have achieved gold-medal level in the International Mathematical Olympiad! 🏆, solving five out of six problems perfectly, as verified by the IMO organizers! It’s been a wild run to lead this
0
0
12
Good to see lots of interest on this paper. It is a system 1 vs system 2 processing issue. We will not get to AGI simply by working on deductive reasoning.
0
1
8
How I would argue this point: .Most of current RL feedback methods are actually maximum likelihood methods for a single-turn prompt-response. Question: Is this enough? Lol.
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.
1
0
15
RT @demishassabis: Our open source Gemma models are the most powerful single GPU/TPU models out there! Our latest model Gemma 3n has amazin….
aistudio.google.com
The fastest path from prompt to production with Gemini
0
407
0
What's the difference between management and leadership? . Management is fundamentally about logistics. The right people with right resources doing right things at right times with the right mission. But not enough. Leadership is inspiration to action, to commit to the mission.
2
0
31
RT @OfficialLoganK: Say hello to the @geminicli, a local CLI to help you build and maintain software with 1,000 free Gemini 2.5 Pro request….
0
670
0
Live: a selfie with this year's Novel prize winner in chemistry: John Jumper. We had a great chat about science discovery and LLM's role in it.
1
1
49
Live: Me and Richard Sutton, this year's ACM Turning Award winner. Congratulations, Richard!
10
15
477
The grandma eval for AGI: .(In mandarin, so you might need live translation from Gemini Live.).
facebook.com
【#塞掐預告】🤓 紀懷新：Google DeepMind 研究副總裁｜台灣唯一智慧眼鏡原型機現身！人工智慧多模態 AI、機器人、AGI 看法全公開 本集塞掐非常榮幸邀請到 Google DeepMind 研究副總裁紀懷新博士，帶來一場深度的 AI 趨勢解析。節目從他在科技史傳奇搖籃 Xerox PARC 的工作經驗出發，探討大型語言模型（LLM）向多模態人工智慧的演進。...
4
2
21
Hot Take: AGI arrives when your grandma doesn't have to tell her domestic robot how to do something. (Or at least have to teach 'it' only once.).
3
0
15
RT @getjonwithit: Calling c the "speed of light" completely misses the point. Rather, c is the "spacetime exchange rate": how many units of….
0
2K
0