Archit Sharma
@archit_sharma97
Followers
6K
Following
2K
Media
38
Statuses
360
RL, post-training, reasoning research @GoogleDeepMind | co-created: Gemini Deep Think series, DPO | prev: @Stanford @Google Brain @IITKanpur @MILAMontreal
Joined July 2015
I think this is one of the interesting situations where considering multiple hypotheses (à la Deep Think) is good. (does someone have gpt-5 thinking or pro results?)
2
4
29
RT @VictorTaelin: For the first time ever, an AI has managed to derive a generic "foldr" function for N-tuples in the λ-Calculus. It was Ge….
0
45
0
To me, there is something beautiful about how the models become better at generating math proofs and creating these pagodas _at the same time_. Scaling inference-time compute truly feels magical sometimes.
Gemini 2.5 Deep Think is out!! We were able to improve the model substantially since our announcement at I/O, and it is a faster variation of the system that got Gold 🥇at IMO (still getting bronze level performance🥉!!). The model is p good at detailed creative tasks too!
5
0
95
RT @g01na2: DeepThink is officially out! 🚀. It’s been an incredible journey from our announcement at I/O to achieving Gold 🥇 at the IMO. We….
0
7
0
RT @emollick: Had early access to Gemini with Deep Think. Very good model, big gains over standard Gemini 2.5 Pro for a lot of problems. H….
0
90
0
RT @jon_lee0: Deep Think is finally here! You can now try (a faster version of) the model that won gold 🥇 at IMO. The most exciting part ab….
0
11
0
more about what it is, how it works and how to use it:
blog.google
Deep Think utilizes extended, parallel thinking and novel reinforcement learning techniques for significantly improved problem-solving.
0
0
7
All these results are _without_ tools: no search, no code execution -- this is raw intelligence. 13% increase on HLE and LCB, almost doubling the performance on IMO compared to other models. And we have found the model is p good at pagodas and pelicans, so give it a shot :)
1
3
22
Gemini 2.5 Deep Think is out!! We were able to improve the model substantially since our announcement at I/O, and it is a faster variation of the system that got Gold 🥇at IMO (still getting bronze level performance🥉!!). The model is p good at detailed creative tasks too!
14
19
291
RT @GoogleDeepMind: For researchers, scientists, and academics tackling hard problems: Gemini 2.5 Deep Think is here. 🤯. It doesn't just an….
0
501
0
RT @scychan_brains: Before money, human societies tended to use credit/debit (and minimal bartering). But most ended up developing money 💵….
0
5
0
RT @jon_lee0: I’m excited to share the news of Gemini Deep Think’s gold-medal level performance 🥇 at the International Math Olympiad! It ha….
0
10
0
RT @vinayramasesh: @aidan_mclau @YouJiacheng It's worth noting that a DeepThink system with no access to this corpus also got gold (again a….
0
27
0
who will solve P6 with a general RL method first?.
I have been waiting for this to be announced, it’s so amazing to see such elegant scaling of the Deep Think system where the same system can now achieve a gold at IMO!
1
0
32
(also my bad with the accidental leak, but hope things were exciting).
2
1
40
Deep Think may truly be one of my favorite creations/discoveries. I know we have promised releasing it for a little bit, but we will do this very very soon.
3
1
29
I cannot emphasize this enough: the system use no tools, no lean — text in, text out. And the more we scale inference compute, the more accurate the proofs get, while still reading like natural text.
1
5
55