Tong He
@simpleigth
Followers
138
Following
107
Media
1
Statuses
14
Staff Research Scientist @Meta MSL, Reasoning/RL/LLM. Ex @GoogleDeepMind (Gemini 2.5 Pro, Co-created Deep Think, AI IMO Gold Medal), @Waymo Research, PhD @UCLA
California, USA
Joined December 2020
Deep Think is a coherent approach that puts together parallel exploration, verification, evolution, TTC scaling and of course RL. It’s one small step towards Superintelligence but evolving fast.
Starting today, we're offering Deep Think in the Gemini app for Google AI Ultra subscribers, and we're giving a collection of mathematicians access to the full version of the Gemini 2.5 Deep Think model that achieved gold medal🏅 level performance in the recent IMO competition.
0
0
3
RT @demishassabis: Gemini 2.5 Deep Think is state-of-the-art performance across many challenging benchmarks!.
0
158
0
RT @shengjia_zhao: I am very excited to take up the role of chief scientist for meta super-intelligence labs. Looking forward to building a….
0
340
0
RT @alexandr_wang: We are excited to announce that @shengjia_zhao will be the Chief Scientist of Meta Superintelligence Labs!. Shengjia is….
0
487
0
RT @OriolVinyalsML: Drastic progress on maths with Gemini 2.5! As a math undergrad, I am impressed 🤯. 🥈 -> 🥇 ✅.Formal -> Informal ✅.Special….
deepmind.google
Our advanced model officially achieved a gold-medal level performance on problems from the International Mathematical Olympiad (IMO), the world’s most prestigious competition for young...
0
79
0
Back on X, and couldn’t resist reposting this gem.
0
1
0
Heard from my former colleagues that Deep Think scales well on challenging Math problems. Looking forward to the IMO post from @GoogleDeepMind 💪.
Watch Gemini 2.5 Pro Deep Think tackle the challenging "catch a mole" problem from @Codeforces. 🪤. This new mode is based on our research in parallel thinking and considers multiple hypotheses before responding. See it in action ↓
0
0
1
New #CoRL2020 work w/ @albert_zhao1 @YitaoLiang @guyvdb, Stefano Soatto on "Squeeze-and-Mimic Networks for Conditional.Visual Driving Policy Learning". We present an end-to-end driving model using a representation that removes driving-irrelevant nuisances.
0
1
3