Zixiang Chen (@_zxchen_) header
Zixiang Chen (@_zxchen_) profile picture

Founding Research Scientist @Geodesiclab | Ph.D. @UCLA | B.S. @Tsinghua_Uni | agentic RL, scaling, theory | Prev @SalesforceAI | Opinions are my own

San Francisco Bay Area
Joined August 2019
Photo by Zixiang Chen (@_zxchen_)Photo by Zixiang Chen (@_zxchen_)Photo by Zixiang Chen (@_zxchen_)Photo by Zixiang Chen (@_zxchen_)Photo by Zixiang Chen (@_zxchen_)Photo by Zixiang Chen (@_zxchen_)Photo by Zixiang Chen (@_zxchen_)Photo by Zixiang Chen (@_zxchen_)Photo by Zixiang Chen (@_zxchen_)Photo by Zixiang Chen (@_zxchen_)
Zixiang Chen retweeted
Excited to see AlphaGenome 2! Its progress on cell-type-specific DNA modeling leads us to ask: can sequence models empower predictions of the transcriptome-wide response to perturbing a genomic locus? STRAND (NeurIPS '26) is our early attempt! teapawt.org/adamgayoso/status/2120…
Single-cell atlases continue to map the diversity of cell states, but how can we use them to interpret non-coding genetic variants? Excited to share AlphaGenome 2, our model for predicting cell-type-specific regulatory activity and variant effects from DNA sequence. deepmind.google/blog/alphage…
2
2
31
6,886
Zixiang Chen retweeted
I sent this message to my student, whose two papers received scores of 5/4/4 and 5/4/3 at #NeurIPS2026 but were rejected. It’s also what I want to say to all junior students going through the same process: please be proud of what you accomplished with these papers, and don’t let these decisions shake your confidence in research. Happy Mid-Autumn Festival to you all! 🌕 🥮
20
34
7
565
35,577
We announced Koa at #DF26 this week. #EnterpriseAI runs on governed, repeatable work, and that takes domain-specific reasoning. Built on @Nvidia's model, @SFResearch trained Koa on 27 years of enterprise workflows. @justjayesh's and my piece for more: sforce.co/4rhI9WA
1
4
1
22
2,222
Zixiang Chen retweeted
Almost 3 years ago, a small team at @SFResearch made a bet: that teaching models to take the right actions , function calling, would matter enormously for AI agents. This week, that bet shipped as Salesforce Koa, our first reasoning model. Here's the journey 👇
6
6
50
5,423
We’re launching the Geodesic Grand Challenges. What are the hardest questions in drug discovery that, if solved, could change what’s possible over the next 5–10 years? We’re asking you to help define them. Selected questions will become official Geodesic Grand Challenges. Each selected contributor receives 10,000 Geodesic credits. Once the challenges are selected, we’ll open a call for solutions, with a $1 million prize for a verified solution to each challenge. The next breakthrough starts with the right question. Submit yours! geodesiclab.com/grand-challe…
4
18
8
266
332,008
Announcing Koa: @Salesforce's first CRM reasoning model, built on @NVIDIA Nemotron 🧠 It builds on a portfolio of domain-specialized models developed by @SFResearch and Salesforce Engineering By orchestrating these purpose-built models alongside frontier LLMs, we route targeted jobs — like intent classification, toxicity screening, and search reranking — to the exact model built for them, while Koa handles multi-step enterprise reasoning. → Domain-specialized models: sforce.co/4AfWkPQ → Moirai: sforce.co/4raX1G2 → Press Release: sforce.co/4y6xXCT #FutureOfAI #EnterpriseAI #AgenticAI
Introducing Koa, built on @NVIDIA Nemotron Salesforce’s first CRM reasoning model for Agentforce brings 27 years of CRM intelligence into the model itself → Matches or exceeds leading model performance on CRM actions with 3x fewer errors → Built for complex, multi-step workflows → No customer data used to train it → Now in pilot And our work with @NVIDIA goes further — bringing NVIDIA Nemotron-based models + accelerated computing to Missionforce for mission-specific AI. Read more 👇 sforce.co/4xugjb9
1
9
34
3,664
Zixiang Chen retweeted
Introducing Koa, built on @NVIDIA Nemotron Salesforce’s first CRM reasoning model for Agentforce brings 27 years of CRM intelligence into the model itself → Matches or exceeds leading model performance on CRM actions with 3x fewer errors → Built for complex, multi-step workflows → No customer data used to train it → Now in pilot And our work with @NVIDIA goes further — bringing NVIDIA Nemotron-based models + accelerated computing to Missionforce for mission-specific AI. Read more 👇 sforce.co/4xugjb9
8
23
7
119
19,067
1/ Today, we’re introducing Geodesic Intelligence, and launching NovaDDE and NovaAtom-Lite-Preview. Geodesic is building an AI-native platform for protein therapeutics, building AGI for drug discovery to find the shortest path from biology to medicines. geodesiclab.com
17
38
15
953
344,336
Zixiang Chen retweeted
For my first post, I’m sharing a letter @NVIDIA signed on why open models matter. AI will transform every industry, power every company, and be built by every country. Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty. The world needs both frontier closed models and frontier open models. images.nvidia.com/pdf/Open-W…
16,563
34,458
12,076
193,836
69,433,357
Zixiang Chen retweeted
Free People Use AI Freely! Mahalo, Jensen! Since 2018, Salesforce Research — led by @silviocinguetta and team — has put CodeGen, XGen, BLIP, Moirai, xLAM, Merlion, and others on Hugging Face (where we led the last round) for the world to use freely. Jensen says it best: the world needs both frontier closed models and frontier open models. We’ve been building in the open for years. Send me your CV and join @SFResearch. ❤️
For my first post, I’m sharing a letter @NVIDIA signed on why open models matter. AI will transform every industry, power every company, and be built by every country. Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty. The world needs both frontier closed models and frontier open models. images.nvidia.com/pdf/Open-W…
37
73
7
444
138,579
I've never felt this much behind as a programmer. The profession is being dramatically refactored as the bits contributed by the programmer are increasingly sparse and between. I have a sense that I could be 10X more powerful if I just properly string together what has become available over the last ~year and a failure to claim the boost feels decidedly like skill issue. There's a new programmable layer of abstraction to master (in addition to the usual layers below) involving agents, subagents, their prompts, contexts, memory, modes, permissions, tools, plugins, skills, hooks, MCP, LSP, slash commands, workflows, IDE integrations, and a need to build an all-encompassing mental model for strengths and pitfalls of fundamentally stochastic, fallible, unintelligible and changing entities suddenly intermingled with what used to be good old fashioned engineering. Clearly some powerful alien tool was handed around except it comes with no manual and everyone has to figure out how to hold it and operate it, while the resulting magnitude 9 earthquake is rocking the profession. Roll up your sleeves to not fall behind.
2,554
7,301
2,906
55,184
17,115,919
🚀Introducing GRAPE: Group Representational Position Encoding. Embracing General Relative Law of Position Encoding, unifying and improving Multiplicative and Additive Position Encoding, such as RoPE and Alibi! Better performance with a clear theoretical formulation! Project Page: model-architectures.github.i… Paper: model-architectures.github.i… Devoted to the frontier of superintelligence, hope you will enjoy it!
9
67
5
462
44,443
Zixiang Chen retweeted
People are sleeping on Deep Agents. Start using them now. This is a fun paper showcasing how to put together advanced deep agents for enterprise use cases. Uses the best techniques: task decomposition, planning, specialized subagents, MCP for NL2SQL, file analysis, and more.
30
96
6
703
60,921
Introducing Enterprise Deep Research (EDR): A steerable multi-agent system that transforms complex enterprise research into comprehensive, actionable reports 📊 EDR combines 5 key components: 🧠 Master Planning Agent for adaptive query decomposition 🔍 4 specialized search agents (General, Academic, GitHub, LinkedIn) 🛠️ Extensible MCP-based tools (NL2SQL, file analysis, enterprise workflows) 📈 Visualization Agent for data-driven insights 🔄 Reflection mechanism with optional human-in-the-loop guidance Results on open benchmarks: ✅ Outperforms SOTA on DeepResearch Bench (49.86 score) ✅ 71.57% win rate on DeepConsult vs OpenAI DeepResearch ✅ 68.5% coverage on ResearchQA across 7 research domains We're releasing EDR-200 dataset with complete research trajectories from 201 benchmark evaluations 📂 📄 Paper: bit.ly/49in6fp 💻 Code: bit.ly/4huXiPq 📊 Dataset: bit.ly/3LbHcOt Authors: @aksh_555 @shoonyaka1 @zxchen @iscreamnearby @huan__wang at @Salesforce AI Research #MultiAgent #EnterpriseAI #DeepResearch #OpenScience
2
16
3
91
8,086
Zixiang Chen retweeted
1/8 Second Order Optimizers like SOAP and Muon have shown impressive performance on LLM optimization. But are we fully utilizing the potential of second order information? New work: we show that a full second order optimizer is much better than existing optimizers in terms of iteration complexity (~5x over SOAP and ~15x over Muon).
23
81
25
591
145,003
Excited to release new repo: nanochat! (it's among the most unhinged I've written). Unlike my earlier similar repo nanoGPT which only covered pretraining, nanochat is a minimal, from scratch, full-stack training/inference pipeline of a simple ChatGPT clone in a single, dependency-minimal codebase. You boot up a cloud GPU box, run a single script and in as little as 4 hours later you can talk to your own LLM in a ChatGPT-like web UI. It weighs ~8,000 lines of imo quite clean code to: - Train the tokenizer using a new Rust implementation - Pretrain a Transformer LLM on FineWeb, evaluate CORE score across a number of metrics - Midtrain on user-assistant conversations from SmolTalk, multiple choice questions, tool use. - SFT, evaluate the chat model on world knowledge multiple choice (ARC-E/C, MMLU), math (GSM8K), code (HumanEval) - RL the model optionally on GSM8K with "GRPO" - Efficient inference the model in an Engine with KV cache, simple prefill/decode, tool use (Python interpreter in a lightweight sandbox), talk to it over CLI or ChatGPT-like WebUI. - Write a single markdown report card, summarizing and gamifying the whole thing. Even for as low as ~$100 in cost (~4 hours on an 8XH100 node), you can train a little ChatGPT clone that you can kind of talk to, and which can write stories/poems, answer simple questions. About ~12 hours surpasses GPT-2 CORE metric. As you further scale up towards ~$1000 (~41.6 hours of training), it quickly becomes a lot more coherent and can solve simple math/code problems and take multiple choice tests. E.g. a depth 30 model trained for 24 hours (this is about equal to FLOPs of GPT-3 Small 125M and 1/1000th of GPT-3) gets into 40s on MMLU and 70s on ARC-Easy, 20s on GSM8K, etc. My goal is to get the full "strong baseline" stack into one cohesive, minimal, readable, hackable, maximally forkable repo. nanochat will be the capstone project of LLM101n (which is still being developed). I think it also has potential to grow into a research harness, or a benchmark, similar to nanoGPT before it. It is by no means finished, tuned or optimized (actually I think there's likely quite a bit of low-hanging fruit), but I think it's at a place where the overall skeleton is ok enough that it can go up on GitHub where all the parts of it can be improved. Link to repo and a detailed walkthrough of the nanochat speedrun is in the reply.
669
3,306
731
23,900
5,824,123
Today Thinking Machines Lab is launching our research blog, Connectionism. Our first blog post is “Defeating Nondeterminism in LLM Inference” We believe that science is better when shared. Connectionism will cover topics as varied as our research is: from kernel numerics to prompt engineering. Here we share what we are working on and connect with the research community frequently and openly. The name Connectionism is a throwback to an earlier era of AI; it was the name of the subfield in the 1980s that studied neural networks and their similarity to biological brains. thinkingmachines.ai/blog/def…
230
1,232
449
7,560
3,552,391
Meet SFR-DeepResearch (SFR-DR) 🤖: our RL-trained autonomous agents that can reason, search, and code their way through deep research tasks. 🚀SFR-DR-20B achieves 28.7% on Humanity's Last Exam (text-only) using only web search 🔍, browsing 🌐, and Python interpreter 🐍, surpassing DeepResearch with OpenAI o3 and Kimi Researcher. 🤖SFR-DR agents are trained to operate independently, without pre-defined multi-agent workflows. They autonomously plan, reason, and propose and take actions as defined by their tools. 🔄SFR-DR agents are trained with end-to-end RL. Starting from reasoning optimized models, our RL pipeline carefully preserves reasoning abilities while training models to become more capable research agents. 📝SFR-DR agents are also trained to manage their own memory by summarizing previous results when context becomes limited. This enables a virtually unlimited context window, enabling long-horizon tasks Paper: arxiv.org/abs/2509.06283 #AIAgents #ReinforcementLearning #DeepResearch
28
141
28
930
310,643
Announcing Flexible Masked Diffusion Models (FlexMDMs)—a new diffusion language model for flexible-length sequences. 🚨 Solves MDMs' fixed-length issue + retrains any-order sampling 🚨 <1000 GPU-hrs to fine-tune LLaDA-8B into FlexMDM (GSM8K 58→67%, HumanEval-infill: 52→65%)
3
60
7
360
74,675
Zixiang Chen retweeted
New Anthropic research: Persona vectors. Language models sometimes go haywire and slip into weird and unsettling personas. Why? In a new paper, we find “persona vectors"—neural activity patterns controlling traits like evil, sycophancy, or hallucination.
219
868
314
5,704
1,431,027