Garrett Bingham
@gjb_ai
Followers
152
Following
16
Media
6
Statuses
33
Senior Research Scientist @ Google DeepMind | Gemini Deep Think | AI IMO Gold
Joined November 2021
RT @j_dekoninck: We launched a new competition on MathArena: Evaluation on the International Mathematics Competition for University student….
0
22
0
Wow!! 🤯.
For the first time ever, an AI has managed to derive a generic "foldr" function for N-tuples in the λ-Calculus. It was Gemini Pro 2.5, with Deep Think. All other models fail to do so, including o3 and Grok 4. Nice progress @GoogleDeepMind.
0
0
1
Thank you @SamuelAlbanie!!!.
We just shipped Gemini 2.5 Deep Think. it doesn't just recall research papers - it fuses ideas across papers in ways I haven't seen before. this level of capability demands careful evaluation. model card below 👇
0
0
1
Give it a try!.
0
0
2
Well over a hundred people were involved in making this launch happen. It's not possible to thank them all here but let me just give a special shoutout to @jon_lee0, @g01na2, @archit_sharma97, @simpleigth, @lmthang, @quocleix for being amazing collaborators in the early stages.
1
0
9
Gemini 2.5 Deep Think is out!. During the IMO I saw a lot of "public models cannot even get Bronze" posts, but it's not the case any more! What we're releasing here is a faster but very similar setup to our Gold medal model. It should be comfortably in Bronze territory. 🥉. 1/2.
For researchers, scientists, and academics tackling hard problems: Gemini 2.5 Deep Think is here. 🤯. It doesn't just answer, it brainstorms using parallel thinking and reinforcement learning techniques. We put it into the hands of mathematicians who explored what it can do ↓
3
1
20
I can't believe how fast this field moves. The paper was published three years ago, and the experiments are already very old-fashioned by today's standards. Regardless, I hope the core ideas will stay relevant for a little while longer!. 4/4.
arxiv.org
Recent studies have shown that the choice of activation function can significantly affect the performance of deep learning networks. However, the benefits of novel activation functions have been...
0
0
1
We took it a step further and let the functions change shape during training. It turned out to be quite effective - the extra flexibility helped with transitioning from learning broad, general concepts early in training to more fine-grained, specific information later on. 3/4.
1
0
0
Our system found new activation functions (a building block of most AI systems) that were better than any of the ones that had been previously designed by humans. 2/4.
1
0
0
Thank you DeLiang Wang and INNS for the Neural Networks Best Paper Award!. This award is for a paper I wrote in grad school with my advisor Risto Miikkulainen. Instead of building AI architectures by hand, we worked on using AI to design better AI. 1/4 🧵
1
0
3
"It was a team effort" sounds cliché, but it really was! We had so many late-breaking super critical contributions. What a magical experience. 🥇.
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
0
0
5
An advanced version of Gemini with Deep Think achieved a gold medal at this year's IMO! 🥇. So many amazing contributions from the team. What's next?.
0
3
16
RT @lmthang: Attached is a full proof by Gemini 2.5 Pro #DeepThink with our experts' comments Here I quote a few….
drive.google.com
0
5
0
Working on getting this in everyone's hands soon. Stay tuned! :).
Deep Think in 2.5 Pro has landed. 🤯. It’s a new enhanced reasoning mode using our research in parallel thinking techniques - meaning it explores multiple hypotheses before responding. This enables it to handle incredibly complex math and coding problems more effectively.
0
0
0
More to come soon!.
Gemini 2.5 Pro Deep Think has the highest score across many of the hardest benchmarks for maths, coding, and multimodal reasoning.
0
0
0
It was amazing working on this. How long til we can solve unsolved problems?.
Yesterday at #GoogleIO, we introduced Gemini 2.5 Pro Deep Think 🧠, pushing the frontiers of AI reasoning. This enhanced reasoning mode is built to tackle drastically complex problems – like USAMO problems that stumped previous models. Super proud of the GDM team for this one!
0
0
0
RT @archit_sharma97: 2.5 Pro Deep Think is an incredibly smart model. Some of the benchmark results, simply put were surprising to me. But,….
0
14
0
Gemini 2.5 Pro Deep Think is an SVG artist!. Prompt: "Draw a SVG of a Pelican riding a bicycle".Left: Gemini 2.5 Pro.Right: Gemini 2.5 Pro Deep Think. Credit:
1
1
22
Gemini 2.5 Pro Deep Think!!!. Truly crazy how a handful of people can have a random idea, implement it, and a few months later it becomes the world's smartest AI. What a wild experience.
Amazing progress in reasoning! 🚀 Gemini 2.5 Pro Deep Think hitting 49.4% on USAMO – a feat I'd have considered impossible just a couple of years ago – & Gemini 2.5 Flash achieving 1424 Elo are huge leaps. So proud our team's research ideas contributed to this moment!.
0
0
1