Assistant Professor at UC Berkeley, Department of Statistics and EECS. Researcher at OpenAI working on LLM training.
Berkeley, CA
Joined January 2015
- Tweets98
- Following698
- Followers4.1K
- Likes353
Song Mei retweeted
We’re releasing a broad range of new mathematical results produced by an internal frontier model.
We’ve been consulting with the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study, and we have drawn on their advice and public recommendations to inform how we release these results.
github.com/openai/math
Song Mei retweeted
Announcing our work textGrain, @OpenAI's text watermark in response to the EU AI Act. OpenAI is rolling out textGrain in the EU and opening detector access to researchers.
We'll open-source it so the community can build on top.
We're expanding our approach to content provenance to include text in response to EU regulatory requirements, while recognizing the significant limitations of current text watermarking technology.
Our tools already help verify whether an image or audio file was created with our models. This work builds on those efforts to help people better understand when content may have been generated or edited with an OpenAI model.
In the EU, we’ll start watermarking eligible text from ChatGPT and Codex over the coming weeks to comply with the EU AI Act.
Customers using our API can turn on text watermarking for select models worldwide today.
Song Mei retweeted
We’re sharing a solution to the Navier-Stokes Millennium Prize Problem, one of the deepest problems at the frontier of mathematics.
The proof was produced by a group of agents, using an OpenAI next-generation model significantly more capable than GPT-6 Astra.
The problem concerns whether the description of smooth three-dimensional fluid motion modeled by the Navier-Stokes equations can break down. It has remained unresolved for roughly 90 years.
Song Mei retweeted
GPT-6-Astra is a declaration.
Progress is best measured when compared with the past. I looked back into GPT-5.2’s blog post – many of the things that Astra does really well on, are also things that 5.2 started to be able to do -- computer use, coding agent, professional work, long-running agents, innovative researcher, aligned behaviors. None of these areas were not known at 5.2 time, Astra just does everything meaningfully better, really good on many of them, superhuman on some of them.
But if a model does everything this well, autonomously, and via computer use, it starts to fulfill the definition of that ultimate goal right?
It was a privilege to cook together with the entire @OpenAI team for this one, and all the GPT-5.X’s over the past year. Now a new 6.X series to fill in, need to teach 'em harder things 😅
Song Mei retweeted
Announcing that GPT-6 Astra has pushed the prime gap to 186, with Lean formalization!
I was 9 when I first heard the twin prime conjecture. Its elegance and Yitang Zhang’s legendary story have always stuck with me. A truly surreal night, being the first to see our model make progress, pushing 246 all the way down to 186, on a problem I’ve revered since I was a kid.
For me, it felt like witnessing a new era of intelligence being born, made possible by everyone at @OpenAI!
🧄🧄🧄🚀🚀🚀
🧄GPT-5.2 is here – one small step on version number, one giant leap in capabilities. 🚀
With *incredible* @Song__Mei @yaodong_yu @Yuf_Zh @ofirnachum and rest of the @OpenAI team, we applied new techniques to bring our frontier reasoning model to the next level. GPT-5.2-Thinking is much stronger on intelligence, agentic coding, professional use, long-context understanding, and extended thinking.
It’s also better on science/theory research – try pairing with it!
Congrats also to @yanndubs @ericmitchellai @.ishaan @christinahkim, and heartfelt thanks to the leadership @_aidan_clark_ @max_a_schwarzer @markchen90 @merettm @sama for making this come together!
Today’s the day — GPT-5 is here! One of the reasons I joined OpenAI was to train the next generation of GPT. It became my first big project, and to my surprise, I became one of the core contributors — together with @yubai01 and many amazing colleagues — developing unexpected but cool techniques that made it into the final training stack. Hard to believe that just two months here could lead to such a big impact.
This cannot happen without amazing colleagues: @SebastienBubeck
@yanndubs
@ElaineYaLe6
@christinahkim
@ericmitchellai
@michpokrass
@max_a_schwarzer
@SebastienBubeck
@minyoung_huh
@_chris_lu_
@suvansh
@hadisalmanX
Song Mei retweeted
Today is the day -- we are excited to bring gpt5 to you.
Fortunate to have led several workstreams in GPT5 Thinking and Mini model training. Among many other improvements, with @Song__Mei @minyoung_huh @SebastienBubeck and co, we applied some unexpected but cool techniques to make the model smart, chatty, and a good model all-around.
Also honored to have worked together with the crew @yanndubs @ElaineYaLe6 @christinahkim @ericmitchellai @michpokrass @max_a_schwarzer and everyone else, it was fun coming together and doing things!
Let us know how you like or dislike it -- this will not be the last model we're gonna train.
I’m excited to start at OpenAI this May and help ship the oss model. More to come soon!
We released two open-weight reasoning models—gpt-oss-120b and gpt-oss-20b—under an Apache 2.0 license.
Developed with open-source community feedback, these models deliver meaningful advancements in both reasoning capabilities & safety.
openai.com/index/introducing…
Congrats @LesterMackey ! I learned my first PhD stats course from your Stats 300 and have enjoyed the beauty of stats theory ever since.
This is a reminder that the FSML workshop (fsmlims.wixsite.com/fsml25, co-located with JSM) paper submission deadline, Mar 3 AoE, is in less than 2 days. This is a non-archival workshop so feel free to submit any interesting statsML paper already submitted elsewhere. It is a great opportunity to get travel funding to JSM and communicate your recent research with the community. Submission portal here: openreview.net/group?id=imst…
Song Mei retweeted
We show the implicit bias of GD for generic non-homogeneous deep nets (results of such were previously limited to homogenous ones). In particular, our results cover those with residual connections and non-homogeneous activation functions. It's a joint work with Kangjie Zhou, @uuujingfeng Jingfeng Wu, @Song__Mei Song Mei, Michael Lindsey, and Peter L. Bartlett!
Arxiv: arxiv.org/abs/2502.16075.
Excited to share our review paper on LLM for statisticians!
🚨 New Paper 🚨
An Overview of Large Language Models for Statisticians
📝: arxiv.org/abs/2502.17814
- Dual perspectives on Statistics ➕ LLMs: Stat for LLM & LLM for Stat
- Stat for LLM: How statistical methods can improve LLM uncertainty quantification, interpretability, trustworthiness & more.
- LLM for Stat: How LLMs can enhance statistical workflows: from data collection, synthesis, annotation to statistical modeling, with applications to medical research
Presents key LLM advances: Architecture, Training, Reasoning, and Self-Alignment:
(1) 🧠Evolution of LLM architectures with Transformers and Self-Attention
(2) LLM training pipeline from pre-training, SFT, to RLHF and Preference Optimization.
(3) 💭 System 2 Prompting and Chain-of-Thought for test-time scaling .
(4) 🚀 LLM Self-Alignment for achieving super-human intelligence
Statisticians play a key role in the development of large-scale AI models:
(1) 💡 Statistical insights improve LLM uncertainty quantification & interpretability
(2) 🤖 Watermarking for AI-generated content detection
(3) ⚖️ Privacy & algorithmic fairness to ensure responsible AI adoption
LLMs can also empower statistical science by:
(1) 📈 Scaling up data collection, synthesis, and annotation.
(2) 🖥️ Automating statistical coding & exploratory analysis
(3) 🔬 Facilitating medical research
By bridging statistics & AI, we can:
✅ Improve better LLMs with statistical methodologies.
✅ Leverage LLMs for statistical applications in high-stakes domains
Honored to receive the Sloan Fellowship! Grateful for the support I’ve received along the way. With this opportunity, I look forward to advancing research in the intersection of math, statistics, and AI. #SloanFellow
🎉Congrats to the 126 early-career scientists who have been awarded a Sloan Research Fellowship this year! These exceptional scholars are drawn from 51 institutions across the US and Canada, and represent the next generation of groundbreaking researchers. sloan.org/fellowships/2025-F…
Song Mei retweeted
DeepSeek r1 is exciting but misses OpenAI’s test-time scaling plot and needs lots of data.
We introduce s1 reproducing o1-preview scaling & performance with just 1K samples & a simple test-time intervention.
📜arxiv.org/abs/2501.19393
