Avi Singh

Excited to announce our new work on using synthetic data for improving mathematical problem solving and code generation in LLMs! arxiv: arxiv.org/abs/2312.06585 A small amount of fine-tuning can lead to large gains (>6% on Hendrycks MATH with Palm-2)

Link! youtu.be/3Y8aq_ofEVs?si…

Enjoyed watching this talk by @DrJimFan on how robotics might follow a parallel path to frontier modeling. He’s a funny guy.

A pleasure working with @vihaniaj and @ShunyuYao14!

Gemini 3.5 Flash is out, and it's a major jump over Gemini 3 Flash in model capability for knowledge work. We've been evaluating it on our Box AI Complex Work Eval in early release, and the model delivers a 12 percentage point jump on complex document tasks. For testing this model, we give the Box AI Agent (using Gemini 3.5) complex problems to solve that represent common but difficult knowledge worker tasks in banking, consulting, public sector, healthcare, and other industries. These tasks can be things like drafting reports, doing due diligence, and more, given a set of relevant documents. In our tests, Gemini 3.5 Flash delivered jumps across every industry, including: * Financial services: 81% vs 73% (+8pp) * Public sector: 76% vs 59%, (+17pp) * Healthcare: 73% vs 51%, (+22pp) * Life Sciences: 67% vs 47%, (+20pp) Incredible to see the continued performance gains. Gemini 3.5 Flash will be available soon in Box AI Studio and through the Box API. The Box MCP Server will soon be available in the Gemini app with more details to come.

Had a lot of fun pushing Flash’s capabilities on long-running agentic tasks!

Just ordered, pretty excited! Also, this is an excellent launch video, worth learning from

Dreams do come true

I've seen enough I'm all in on Gemini

Depending on the person, Berlin might work great as well

@kristianhartika Congrats!

Excited to share what I've been working on for the past couple of years!

Misaligned robot exhibiting malicious behavior spotted in #ICLR2026

@Swarooprm7 @GoogleDeepMind Welcome back!

Personal Update: I am back to @GoogleDeepMind. I will continue working on LLM research and product.

Founders Fund bought about half of DeepMind with a 2.3M investment. Simpler times!

Mustafa Suleyman origin story is pretty insane. Born to a cab driver and abandoned by parents at age 15, he provided for himself and his younger brother by flipping candy bars, cellphones, and eventually cars. Not having a place to live, he often stayed with the Hassabis family.

Reading The Infinity Machine from @scmallaby, and it has been hard to put down

@klyap_ Cuteeeeeee

I made sheep toast today!

While it's difficult to increase the base intelligence of a model without pouring in significant resources (mostly in the form of data and compute), persistence is something that's easier to recognize and reward, all in post-training.

As we start using AI for long-running tasks, model intelligence remains important, but behavioral traits like persistence (or, to use a stronger word, relentlessness) become at least as important, if not more so.