James Zou

Schematic of cell differentiation prediction (a) and drug response prediction (b).

New paper out in @natmethods.nature.com from @elhamazizi.bsky.social, Kam Leong & @jameszou.bsky.social! The team developed Squidiff, a diffusion #AI model to predict cellular responses to environmental cues and accelerate #PrecisionMedicine.

Learn more: bit.ly/3WRPNsx

This is figure 1, which shows LMs struggle to affirm first-person beliefs in factually false scenarios.

Large language models (LLMs) may not reliably acknowledge a user’s incorrect beliefs, according to a paper in Nature Machine Intelligence. The findings highlight the need for careful use of LLM outputs in high-stakes decisions. go.nature.com/48VRpIQ 🧪

It’s here! #Agents4Science recording is now on YouTube!
🏆 3 Best Paper talks
⚡️ 11 Spotlights
🧠 Panel on the future of AI agent-driven science
📚 Lessons + surprises from this first-of-its kind conf

Full analysis of submissions + reviews coming soon! youtube.com/watch?v=7pXq...

AI bots wrote and reviewed all papers at this conference Event will assess how reviews by models compare with those written by humans.

Next week @jameszou.bsky.social & colleagues will host a conference where all the papers are written by AI agents & reviewed by them too.

What do you reckon? A good chance to put AIs through their paces? Or a way to divert AI slop from elsewhere? 🧪🤖

My story here:
www.nature.com/articles/d41...

A circular flow diagram that compares current and proposed practices for LLM development using data from adopters and non-adopters. Three gray boxes represent current practices: “R&D,” “Chat Models,” and “Adopters’ Needs and Usage Data,” connected in a clockwise loop with black arrows. A blue box labeled “Non-adopters’ Needs and Usage Data” adds a proposed feedback path, shown with blue arrows, linking non-adopter data back to R&D and adopters’ data.

As of June 2025, 66% of Americans have never used ChatGPT.

Our new position paper, Attention to Non-Adopters, explores why this matters: AI research is being shaped around adopters—leaving non-adopters’ needs, and key LLM research opportunities, behind.

arxiv.org/abs/2510.15951

"We found a troubling emergent behavior in LLM.
—When LLMs compete for social media likes, they start making things up.
—When they compete for votes, they turn inflammatory/populist.
—When optimized for audiences, LLMs inadvertently become misaligned."
→ Moloch's Bargain @jameszou.bsky.social #AI

First page of article: "The widespread adoption of large language model-assisted writing across society" published in Patterns

#AI wrote nearly a quarter of corporate press releases in 2024 and the number is likely to keep rising. spkl.io/63324ATfBo

Weixin Liang, @jameszou.bsky.social & colleagues
@cp-patterns.bsky.social

James Zou, Ph.D

@stanforddeptmed.bsky.social Biomedical Informatics Research Colloquia

“AI Agents to Automate Biomedical Discoveries”
@jameszou.bsky.social James Zou, Ph.D

Thursday, September 25th, 2025
12:00 to 1:00 pm PST

stanford.zoom.us/j/9788759601...

Webinar ID: 978 8759 6012
Webinar Passcode: 420642

Fantastic talk by @stanford.edu's @jameszou.bsky.social for the first Innovation Initiative Distinguished Lecture at @whiteheadinstitute.bsky.social. A fascinating look at how to use #AI to build intelligent research teams that can tackle open-ended scientific problems. #WhiteheadInstitute #MIT

🧠Can AI agents predict #Alzheimers? Participate in our DREAM challenge agentic track to find out!

We provide unique training + test data for AI agents: snRNA-seq, IHC, stage, etc synapse.org/Synapse:syn6...

Also co-submit your agent paper to agents4science.stanford.edu

Researchers create ‘virtual scientists’ to solve complex biological problems Stanford Medicine researchers created a team of virtual scientists backed by artificial intelligence to help solve problems in their real-world lab.

“Good science happens when we have deep, interdisciplinary collaborations, and often that’s one of the main bottlenecks and challenging parts of research,” said HAI Faculty Affiliate @jameszou.bsky.social who led a study on AI-driven virtual labs: med.stanford.edu/news/all-new...

Researchers create ‘virtual scientists’ to solve complex biological problems An AI lab developed by Stanford Medicine researchers has already shown promising results, generating ideas for a more effective COVID-19 vaccine in just a few days.

Nice @stanforduniversity.bsky.social article profiling our #VirtualLab of AI scientists news.stanford.edu/stories/2025...

Figure 1. An Overview of the Fine-Tuning Case Study.

Figure 2. Performance of Fine-Tuned Large Language Models on Each Medical Dataset.

Figure 3. Model Performance on MedQA after Updating.

Case Study by Eric Wu, PhD, Kevin Wu, PhD, and James Zou, PhD: Limitations of Learning New and Updated Medical Knowledge with Commercial Fine-Tuning Large Language Models nejm.ai/4nTx1Np

@jameszou.bsky.social #AI #MedSky

GitHub - zou-group/virtual-lab: A virtual lab of LLM agents for science research A virtual lab of LLM agents for science research. Contribute to zou-group/virtual-lab development by creating an account on GitHub.

The Virtual Lab is open source and can be applied to many problems! github.com/zou-group/vi...

👏Great job @kylewswanson.bsky.social John Pak, Wes Wu, Nash Bulaong @czbiohub.bsky.social

⚡️Thrilled that #VirtualLab is published in @nature.com! www.nature.com/articles/s41...

We created a team of AI agents to mirror my Stanford lab 🤖. Led by a PI agent, the AI scientists ran their own group meetings and discovered effective binders to new CoVID variants that we validated.

Who do you call when you need to design novel, potent nanobodies vs a pathogen?
The virtual lab of A.I. agents @nature.com @jameszou.bsky.social @kylewswanson.bsky.social
www.nature.com/articles/s41...

Day 3 of #ISMBECCB2025 started with an excellent talk from @jameszou.bsky.social which presented how AI can form virtual labs to discuss problems, how can AI reanalyse research data to get new biological results, and how we can interpret complex AI predictions 💻🤖

Figure 1. An Overview of the Fine-Tuning Case Study.

Six frontier large language models evaluated on incorporating newly updated medical knowledge through commercial fine-tuning application programming interfaces struggled to generalize updated information, despite modest gains. Learn more: nejm.ai/4nTx1Np

@jameszou.bsky.social

🏆Thrilled that #CollabLLM won the #ICML2025 Outstanding Paper Award!

We propose a new approach to optimize human-AI collaboration, which is critical for agents. Congratulations to my fantastic co-authors; great job Shirley Wu and Michel Galley driving the project! 👏

Paper: arxiv.org/pdf/2502.00640

AI ‘scientists’ joined these research teams: here’s what happened Emerging ‘co-scientist’ systems use chatbots to mimic the deliberations of a research group. Nature asked researchers to test them out.

What is it like for scientists to work with AI co-scientists?

Excellent @nature.com feature discussing researchers' experience interacting with our #VirtualLab and other tools www.nature.com/articles/d41...

EvoLM: In Search of Lost Language Model Training Dynamics Modern language model (LM) training has been divided into multiple stages, making it difficult for downstream developers to evaluate the impact of design choices made at each stage. We present EvoLM, ...

Introducing EvoLM, a model suite with 100+ decoder-only LMs (1B/4B) trained from scratch, across four training stages —

🟦 Pre-training
🟩 Continued Pre-Training (CPT)
🟨 Supervised Fine-Tuning (SFT)
🟥 Reinforcement Learning (RL)

Great job by Sam Alber, Bowen Chen, Alina Isakova, Aaron Wilk!

High throughput data are goldmines that we often only scratch the surface. Great for AI to systematically analyze! CellVoyager makes Jupyter notebooks for easy replication
📜https://www.biorxiv.org/content/10.1101/2025.06.03.657517v1
💻https://github.com/zou-group/CellVoyager

💡IMO one of the best use of AI Scientist is to reanalyze data to find new insights.

Introducing #CellVoyager: AI Compbio Agent that makes new discoveries by autonomously analyzing papers/data, which we then validate🚀

New findings on aging, Covid, scRNAseq etc. Open source!

Very cool work from James Zou. Link to paper: www.nature.com/articles/s41...

Excited that nuclei.io is featured on the cover of Nature BME! AI-clinician collaboration paper rdcu.be/dLgV1

We discuss how to conduct intersectional analysis in science and technology in @nature.com
nature.com/articles/s41...

I learned a ton from working w/ dream team co-authors! Great job by Londa Schiebinger + Mathias Nielson leading this project 👏 Also thanks @meharpist.bsky.social for guidance!

Here's the non-paywall version of our #TextGrad paper rdcu.be/efRp4! 📜

SCI member @jameszou.bsky.social & others present a method for genAI self-improvement via a new calculus of text. They show TextGrad’s power in optimizing agents, molecules, code, treatments, & more, solving PhD-level problems and enabling impactful #AI development. www.nature.com/articles/s41...

Optimizing generative AI by backpropagating language model feedback - Nature Generative artificial intelligence (AI) systems can be optimized using TextGrad, a framework that performs optimization by backpropagating large-language-model-generated feedback; TextGrad enable...

Scholars developed a versatile framework called TextGrad to improve complex AI systems by leveraging natural language feedback. @jameszou.bsky.social

Read more about this research supported by @stanfordhai.bsky.social in a new paper published in Nature: www.nature.com/articles/s41...