Lei Yu

36 posts

Lei Yu

Lei Yu

@LeiYu63

Research Scientist @GoogleDeepMind, Gemini Thinking, AI IMO🥇, AlphaProof, Machine Translation | PhD @Oxford | MPhil @Cambridge

London, England Katılım Aralık 2015
253 Takip Edilen637 Takipçiler
Sabitlenmiş Tweet
Lei Yu
Lei Yu@LeiYu63·
Building on what @lmthang shared, here's another fun fact from our final push: we finalized the model checkpoint selection just 5 hours before the IMO problems were released! It's incredible to see the model we were babysitting over that final weekend now demonstrating state-of-the-art performance across the board. So proud that this powerful model is now publicly available.
Thang Luong@lmthang

Our IMO journey continues: the yolo run model that we trained a week before #imo2025, despite all possible likelihood of failures, magically achieves SOTA across a wide range of reasoning tasks from maths, to coding, and challenging knowledge. I'm very excited that we have now delivered the IMO 🥇 system to the hands of mathematicians and a simplified version (results below) to all Google AI Ultra subscribers.

English
0
2
36
4.1K
Andrew M. Dai
Andrew M. Dai@AndrewDai·
After almost 12 years in Brain/DeepMind, I’ve finally decided to take the leap. My cofounders: @yinfeiy, Seth and I have kicked-off @ElorianAI. The first multimodal reasoning lab founded and led by former LLM pretraining, data and multimodal leads. youtu.be/YlvfNpOMeOY?si… (1/n)
YouTube video
YouTube
English
82
71
776
314.4K
Lei Yu
Lei Yu@LeiYu63·
@swarat Congrats Swarat! Very well deserved!
English
0
0
1
128
Swarat Chaudhuri
Swarat Chaudhuri@swarat·
Honored to be named a 2025 #ACMFellow! Many thanks to my students, collaborators, and mentors over the years -- working with you has been a privilege. And a big thank-you to my family for their perennial love, support, and patience.
Association for Computing Machinery@TheOfficialACM

Meet the 2025 ACM Fellows! Hailing from 14 countries and chosen from among ACM's global membership of over 100,000 computing professionals, these 72 honorees are cited for contributions in a wide range of computing research areas, from AI for healthcare, computer graphics, and data management, to electronic mail, human-computer interaction, and numerous other areas. Please join us in congratulating the new inductees! Learn more about their impact here: buff.ly/Pga16R8

English
16
4
59
9.5K
Lei Yu
Lei Yu@LeiYu63·
@BlackHC It’s been really nice collaborating with you Andreas!
English
1
0
4
589
Andreas Kirsch 🇺🇦
Andreas Kirsch 🇺🇦@BlackHC·
Reviewing my time at GDM so far: AlphaProof contributions (but sadly not stuff that made it into the paper, so not a co-author) and IMO strike, then working more directly on inference-time techniques in post-training. Really excited for what's next 😊
English
6
1
124
12.9K
Lei Yu retweetledi
Google DeepMind
Google DeepMind@GoogleDeepMind·
This is Gemini 3: our most intelligent model that helps you learn, build and plan anything. It comes with state-of-the-art reasoning capabilities, world-leading multimodal understanding, and enables new agentic coding experiences. 🧵
English
213
1.1K
6.5K
1.7M
Lei Yu retweetledi
Demis Hassabis
Demis Hassabis@demishassabis·
One word: relentless. just in the past two weeks, we’ve shipped: 🌐 Genie 3 - the most advanced world simulator ever 🤔 Gemini 2.5 Pro Deep Think available to Ultra subs 🎓 Gemini Pro free for uni students & $1B for US ed 🌍 AlphaEarth - a geospatial model of the entire planet 🏛️ Aeneas - deciphering ancient text (in @Nature) 🥇 Gemini gold-medal level at the IMO 🧸 Storybook - books w/art & audio @GeminiApp ♛ New @Kaggle Game Arena benchmark for LLMs 🐙 Jules, our asynchronous coding agent, out of Beta 🇬🇧 AI Mode for Search available in the UK 📔 NotebookLM Video Overviews 🔥 Gemma passed 200m downloads Now you know why I don't get much sleep 🛌 - too busy pushing the frontier!
English
488
939
9.6K
1.1M
Lei Yu
Lei Yu@LeiYu63·
@YiTayML I like your message about morale support 😆. That's crucial.
English
0
0
2
257
Yi Tay
Yi Tay@YiTayML·
Magically achieved sota? That is because we believe in our model and also because of divine benevolence 😃
Thang Luong@lmthang

Our IMO journey continues: the yolo run model that we trained a week before #imo2025, despite all possible likelihood of failures, magically achieves SOTA across a wide range of reasoning tasks from maths, to coding, and challenging knowledge. I'm very excited that we have now delivered the IMO 🥇 system to the hands of mathematicians and a simplified version (results below) to all Google AI Ultra subscribers.

English
11
12
128
12.3K
Lei Yu
Lei Yu@LeiYu63·
@lmthang @aclmeeting @hyhieu226 @chrmanning Fantastic news, Thang! Huge congratulations, so well deserved. I vividly remember it being a must-cite for us in our early neural machine translation work. Wonderful to see its foundational impact get this recognition!
English
0
0
1
124
Thang Luong
Thang Luong@lmthang·
It’s my great honor to have just received the test-of-time award at #ACL2025 from @aclmeeting for our paper “Effective Approaches to Attention-based Neural Machine Translation” (joint work with @hyhieu226 and @chrmanning at EMNLP 2015). It’s the immediate second “attention” paper after (Bahdanau, Cho, Bengio, ICLR 2015) but it was first for many things: * first to introduce the dot-product attention (that many referred as Luong attention & its variant later appeared in the Transformer paper in 2017) * first to achieve state-of-the art neural machine translation, and * first to introduce the notion of local attention (as confirmed by Gemini). I’m very grateful to @ilyasut, @quocleix, @OriolVinyalsML who brought me into neural machine translation when I was an intern at Google Brain in 2014, from which we have changed Google Translate entirely in 2017.
Thang Luong tweet media
English
37
26
508
43.4K
Lei Yu retweetledi
Thang Luong
Thang Luong@lmthang·
Right before #imo2025, together with colleagues from Mountain View, NYC, Singapore, etc, we all gathered at @GoogleDeepMind headquarter in London for our final push for IMO. I believe that week was when all magic happened! We put all individual recipes (that we figured out before) together and did a yolo run (with the compute that I had to beg various groups to loan) to train our most advanced Gemini model. We finished training 2 days before IMO :D That model achieved SOTA results, not just for math, but coding along with other reasoning tasks, unbelievable! That week was also when we figured all details to scale our Deep Think mode, alongside with other enhanced inference strategies. We finalized our runbooks that predetermined the configs to be used during the IMO days (so there will be no human intervention at all). And the rest is history. Thanks @CadeMetz and @nytimes for featuring that historic moment nytimes.com/2025/07/21/tec… The photo below was taken in London during that magical week. Tagging various team members in random order who worked extremely for that moment @YiTayML @theophaneweber @DawsenHwang @JiemingMao @jon_lee0 @zichengxu42 @vinayramasesh @mirrokni @blackhc @gjb_ai @g01na2 @LeiYu63 @NateKushman @jj_at_brown @quocleix @freeetext together with many others that I don't yet have X handles.
Thang Luong tweet media
Thang Luong@lmthang

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 effort and I am grateful to everyone in the team for such an amazing achievement! Blog post in the thread and more to share soon!

English
22
40
545
146.4K
Reka
Reka@RekaAILabs·
🎉 Big news! We've raised $110M from new and existing investors, including @nvidia & @Snowflake This funding reinforces our position at the forefront of AI innovation, with exciting releases like Reka Vision, Reka Research & Reka Flash 3.1 Read more 👇 reka.ai/news/reka-secu…
English
8
19
96
29.2K
Mikel Artetxe
Mikel Artetxe@artetxem·
Excited to announce that we've raised $110M! Together with the recent launch of Reka Research, Reka Vision and Reka Flash 3.1, the future is looking brighter than ever. Let's go! 🚀
Reka@RekaAILabs

🎉 Big news! We've raised $110M from new and existing investors, including @nvidia & @Snowflake This funding reinforces our position at the forefront of AI innovation, with exciting releases like Reka Vision, Reka Research & Reka Flash 3.1 Read more 👇 reka.ai/news/reka-secu…

English
5
6
151
19.7K
Lei Yu
Lei Yu@LeiYu63·
@jack_w_rae Thank you for your leadership and support for this project Jack!
English
0
0
3
242
Jack Rae
Jack Rae@jack_w_rae·
Amazing achievement from the Gemini team! Getting to IMO gold with a general text interface and set of approaches was a pipe dream nine months ago, back when the model was ~35% on AIME. Very cool to see the model's capability accelerate.
Google DeepMind@GoogleDeepMind

An advanced version of Gemini with Deep Think has officially achieved gold medal-level performance at the International Mathematical Olympiad. 🥇 It solved 5️⃣ out of 6️⃣ exceptionally difficult problems, involving algebra, combinatorics, geometry and number theory. Here’s how 🧵

English
3
8
195
15.3K
Kyunghyun Cho
Kyunghyun Cho@kchonyc·
😂 @wellecks , i think this “challenging problem” may have been finally solved after five years. === Understanding and creating mathematics using natural mathematical language … used by humans is a challenging and important problem for driving progress in machine learning. ===
Kyunghyun Cho tweet media
English
5
7
104
9K
Lei Yu
Lei Yu@LeiYu63·
Thrilled to share our latest achievement: our Gemini model has achieved gold medal-level performance at the International Mathematical Olympiad. Unlike our previous work on AlphaProof, this model competed under the exact same conditions as human participants. Specifically, it was required to use natural language for both input and output and to strictly adhere to a 4.5-hour time limit. Our result has been certified by the IMO organisers. I’m very proud to have been the co-captain of model training and evaluation. I couldn’t imagine that we would achieve such a great result when we started this project a few months ago.
Google DeepMind@GoogleDeepMind

An advanced version of Gemini with Deep Think has officially achieved gold medal-level performance at the International Mathematical Olympiad. 🥇 It solved 5️⃣ out of 6️⃣ exceptionally difficult problems, involving algebra, combinatorics, geometry and number theory. Here’s how 🧵

English
2
4
40
6.6K
Ankesh Anand
Ankesh Anand@ankesh_anand·
We can finally share this now: A Gemini model trained with new RL techniques and scaled up inference-time compute model has achieved gold-medal level performance at IMO 2025! 🥇
Ankesh Anand tweet media
English
14
26
457
36.6K
Yi Tay
Yi Tay@YiTayML·
Cooking something very cool and exciting with @quocleix in Singapore very soon. Stay tuned! 😎🇸🇬
English
14
6
128
41.6K
Lei Yu retweetledi
Laurent Sartran
Laurent Sartran@LaurentSartran·
Very happy to see our work on AlphaProof featured in quantamagazine.org/the-year-in-ma…! As we wrote in the blog post, AlphaProof is composed of a formalizer, based on Gemini, and a solver finding and learning from proofs in Lean using AlphaZero -- it's not AG2+Gemini. @QuantaMagazine
English
0
1
6
1K

Keşfet

@AndrewDai @yinfeiy @ElorianAI @swarat @BlackHC @Nature @GeminiApp @Kaggle