Zeus Gao @June020703

AI agents are becoming increasingly autonomous. But our ability to monitor and diagnose their behavior is lacking. I gave a talk at the Simons Institute, covering our latest NeurIPS'26 papers on how to quantify agent progress and diagnose failures at the fine-grained, turn-by-turn level. Video is now available: 🎥 youtube.com/live/5vBzKW3-w… ------- Tracing Agentic Failure from the Flow of Success arxiv.org/abs/2607.12747 Neglected Free Lunch from Post-training: Progress Advantage for LLM Agents arxiv.org/abs/2606.26080 Hide-and-Seek in Trajectories: Discovering Failure Signals for VLA Runtime Monitoring arxiv.org/abs/2605.30834

🧪🧭 What if we could ask a model which experiment gives us the cells we want? Our new preprint, LabCompass, takes a step in this direction through generative inverse design. For me, this is one of the most exciting directions for the field. Together with many others, we have been working on predicting how cells respond to perturbations. How can we use these models to design experiments that achieve a desired biological outcome? 🎛️ This inverse problem is difficult. Combinations and doses create spaces that cannot be exhaustively screened, and each experiment produces a distribution of cell states. Predictive models are also imperfect, particularly in unexplored regions where we want to discover something new. LabCompass couples a generative model of experimental designs with a forward model of cellular responses. The desired phenotype guides generation in continuous design space, while a learned prior favors plausible designs. Selected candidates are tested experimentally, and the measurements inform the next round. 💡 The method is not tied to a particular virtual cell model. It can work with different differentiable forward models and cellular readouts, and the same trained models can address different phenotypic objectives. 🩸 We demonstrated this in primary human blood stem and progenitor cells. Across the study, we profiled 78 million cells and experimentally validated conditions enriching megakaryocytic, erythroid, monocytic and early lymphoid populations. One result I particularly like: distinct culture recipes produced similar target enrichments. Several solutions to the same biological objective give us alternatives to investigate and develop further. 🔬 The opportunity extends beyond blood and cell culture. We also applied LabCompass to single-cell transcriptomic data from viral infection. I’m excited to explore its potential for other perturbation systems, drug combinations, differentiation and organoids. This raises a question for virtual cell models: alongside prediction accuracy, how effectively can they guide experiments toward desired biological outcomes? If this connects to a problem you are working on, I’d love to hear from you. We’re keen to see how LabCompass transfers to new systems and happy to explore ideas together. A wonderful collaboration with Bertie Göttgens and Nicola Wilson at the Cambridge Stem Cell Institute. Huge credit to joint first authors Licyel Paulas-Condori, Alessandro Palma, Lorenzo Consoli and Juan Antonio Rubio-Lara, and the whole team across both labs! 📄 Preprint: biorxiv.org/content/10.648… 💻 Code: github.com/theislab/LabCo… #InverseDesign #VirtualCell #SingleCell

The next frontier would be agents that get smarter as you use them. 🚀 Our new ServeLearnBench puts continual self-improvement to the test—and reveals what’s holding it back.

Infini-AI-Lab @InfiniAILab

2 weeks ago

How well can agents continually self-improve from serving experience? We’re excited to introduce our new benchmark, 𝐒𝐞𝐫𝐯𝐞𝐋𝐞𝐚𝐫𝐧𝐁𝐞𝐧𝐜𝐡: Real-world deployment is not a one-time test of capability. The knowledge agents need is often hidden and can change over time.

1 8 42 11K 25

When do biological reasoning models use their biological inputs? High performance can create a mirage of reliance on biological representations. Reasoning traces don’t always reflect the evidence. Our paper examines these limits and paths to better biological reasoning models.

Check out Rulin’s latest work on Context Language Model! Just found one funny pic when she waited for me to have dinner, she’s still working on it lol. So proud of my bestie💕Congrats on the release again:)

Rulin Shao @RulinShao

3 weeks ago

‼️The Bitter Lesson for context management: Giving LMs unrestricted control over their context beats human-designed SOTA! Introducing 🩵Context Language Models (CLMs)🩵 - Natively manage their own context - Treat context as a file - Learn policies in CLM weights, no harness

99 278 2K 449K 3K

We are excited to announce a new faculty position in AI for Biology at UC Berkeley, jointly between @Berkeley_EECS and @CCBBerkeley. Please repost & share this opportunity with outstanding early-career researchers in your networks. Apply by Nov 9, 2026: aprecruit.berkeley.edu/JPF05560

For cancer studies, Cell-Cell interactions in single cell data can be suggestive but they miss a key component which is the spatial organization in the tissues. Nonetheless, we can rank the interactions and investigate new and existing hypothesis @Nature nature.com/articles/s4159…

Predicting protein-protein interactions (PPIs) at proteome scale can take months with co-folding models due to massive all-vs-all comparisons required. We are excited to announce FlashPPI, a contrastive model that predicts proteome wide physical interfaces in minutes. 1/🧵

🚨Excited to share ProtoMech: a framework for discovering the computational circuits inside protein language models (pLMs)! pLMs like ESM2 are powerful, but the computational mechanisms, or circuits, underlying their predictions remain poorly understood. (1/n)

Excellent resources.

Alexis Vandenbon @alexisvdb

9 months ago

@HaoYin20 Now also with 628 Xenium samples! x.com/alexisvdb/stat…

1 2 13 2K 1

Every CSE PhD should mark CSE.

机器之心 JIQIZHIXIN @jiqizhixin

9 months ago

What if AI could learn to write optimized code like a human engineer—iterating, learning from mistakes, and getting smarter over time? Researchers from NJU, PKU, Midea-AIRC, ECNU, SYSU, RUC, and QuantaAlpha introduce Controlled Self-Evolution (CSE). Their approach uses three

2 8 61 4K 41

Very useful data in ncAA research which helps us to have a deeper understanding of structure. They provided a potential way to explore broader chemical spaces!🪐

Hannes Stark @HannesStaerk

10 months ago

They provide a neat server visualizing all sorts of useful data about non-canonical amino acids and how they occur in PDB 👌 🤗 Paper: biorxiv.org/content/10.648… Server: mondet.tuebingen.mpg.de Very nice to easily see, e.g., how much training data of non-canonicals we have for

1 18 123 8K 84

🚀 Introducing Nucleotide Transformer v3 (NTv3) Today, we are very excited to share our latest foundation model for biology - Nucleotide Transformer v3 (NTv3). NTv3 is @instadeepai new multi-species genomics foundation model, designed for 1 Mb, single-nucleotide-resolution prediction, and for bridging representation learning, sequence-to-function modeling, and generative regulatory design within a single framework 🧬 This work was developed in close collaboration with @AlexanderStark8 , @volokuleshov , and @pkoo562 , and reflects several years of joint effort at the intersection of machine learning and regulatory genomics.

Inspiring work! FEI module is my taste. Trying to figure out whether they could be used for protein folding analysis 🧐

机器之心 JIQIZHIXIN @jiqizhixin

10 months ago

What if you could fine-tune a diffusion model by simply following its natural "frequency rhythm"? FeRA is here. It's a new framework that tunes AI image generators by aligning updates with the model's intrinsic frequency energy—how it naturally builds details from blurry to

4 29 169 10K 116

What an exquisitely assembled machine! Hope I could see self-evolving AI systems are able to simulate the gifts of God someday. 👀👀

William A. Wallace, Ph.D. @WilliamWallace

10 months ago

🔥 Your entire immune system starts with one cell. Every red blood cell that carries oxygen Every platelet that stops a cut from bleeding Every neutrophil that attacks bacteria Every T cell that hunts viruses Every B cell that makes antibodies All of them come from the same

24 386 1K 49K 773

@holyMoha Why all these new year series memes are labeled as sensitive contents?

@ElinaMia34 He is the greatest man🫡

Such inspiring determination and courage!🫡

Varuni Sarwal @sarwal_varuni

10 months ago

One of the hardest decisions in life is finding meaningful work that gives you purpose. Here’s how I tried to answer that question for myself. First, I chose academia. Five years into a PhD, I had published in journals including Nature and my work earned over 500 citations. On

22 42 625 50K 225

The SpatialEx(+) framework uses histology to generate and integrate complementary spatial multi-omics data across samples. nature.com/articles/s4159…

A Perspective reviews recent and emerging developments in building multimodal foundation transformer models in genomics. @sabbaghtweets @VincenzoLagani @robert__lehmann @SumeerrKhan @JesperTegner nature.com/articles/s4159…