CELL @CellBioSF

LLMs have a big blind spot: they lack innate understanding of physical world. This becomes very evident when we zoom into the quantum realm, the complexity of quantum dynamics and the cost of steering it grows exponentially. Fourier Neural Operator (FNO) becomes an important tool to learn molecular quantum dynamics and accelerating the inverse design of control protocols. FNO predicts dynamics 10^7 times faster than GPU-accelerated CUDA-Q @nvidia FNO stochastic pulse-measurement planner achieves nearly twice the success rate of a reinforcement-learning arxiv.org/abs/2608.03702 @PrinehaN

Our new work on “Physics of Agents” arxiv.org/abs/2608.16578 lead by Batu El and Jinhee Paeng in collab w/ @james_y_zou The outcome of many interacting agents seems hard to reason about. Yet we were able to study the opinion dynamics of 10,000 different LLM agent communities as they communicated with each other to solve both objective and subjective questions. Remarkably, we could account for their opinion dynamics through a simple Ising model that involved minimizing an energy function corresponding to social pressure to conform. Our dynamics could explain the build up of consensus, polarization, and societal correction of initially incorrect majorities. Lots more to do on statistical mechanics of interacting agent dynamics!

Preprint 📣: Living tissues can exhibit odd mechanics without chirality. Our model shows that orientational order can generate an odd modulus — inferable from experiments — and reproduces flows and stress localization around ±1/2 defects. @SreejithS_ biorxiv.org/content/10.648…

Nature Editorial Why AI cannot do good Science without humans

This is a good response. But this part "It will actually be possible to cure most human disease in ~5-10 years" I would really really love for folks who say this to actually walk us through how they think this will happen. Rather than just saying it.

Dario Amodei @DarioAmodei

a month ago

2/2 Second, on the messaging around AI.  I do not agree that my messaging has been disproportionately negative.  In fact it has been about equally balanced between risks and benefits: I’ve written one major essay about each, and even in interviews where I discuss the risks, I

1K 647 9K 11.8M 3K

@freedompreetham 👏👏

💥 NEW BOOK: THE LAST HUMAN VETO "In the world of AI, humans will and must retain the final decision!". Really??? I wrote 'The Last Human Veto' because I no longer think that sentence survives scrutiny. If the machine already controls the evidence, the options, the forecast, and the cost of refusal, the veto may remain visible long after its causal power is gone. The book asks a harder question. What does it mean to remain “in control” when intelligence becomes cheap, abundant, and eventually superior to ours? My answer is counterfactual authority. A veto is real only if saying no can still lead somewhere else. FREE PDF DOWNLOAD @ freedompreetham.org #AI #Agents #Humans #Future

Despite claims that big data will solve everything, I still think a conceptual advance in biology supported by a single, sharp, incisive experiment to prove it is better than 100 vaguely corroborative claims based on existing, usually tangential data. But I'm a little less sure…

People claiming solving cancer is just a data scale problem when we’re still out here discovering fundamental regulatory mechanisms.

Nature Japan @NatureJapan

a month ago

ユビキチンはタンパク質だけを標識していたわけではなかった グリコーゲンや代謝物への修飾を発見 Simon A. Cobbold、David Komanderらの@Nature #OA #論文 Ubiquitination of glycogen and metabolites in cells and tissues nature.com/articles/s4158… #オープンアクセス @WEHI_research @unimelb

0 43 178 34K 83

What happens when a world model generates a million futures without collecting a single new fact about the world? Those rollouts may reduce numerical error inside the model while inheriting the same unresolved assumptions. In my new essay, I use epistemic laundering for the moment that computational precision becomes unwarranted confidence about reality. The essay separates this failure from the reality gap, model exploitation, and reward hacking, then develops a hard accounting rule for when an agent must buy fresh evidence. If a model cannot identify what observation or intervention could challenge the assumptions behind an action, deployment becomes the falsification experiment its builders chose not to run. The best definition for "world models" came from a tweet response from @ylecun responding to @anshulkundaje's question [Tweet embedded in the essay] #AI #WorldModels cellbiosf.substack.com/p/epistemic-la…

Busy few days with 8 talks in an 11 day period in 3 different places. Useful links below for programs/schools to apply to next year: Woodshole Methods in Computational neuroscience (2 talks on Deep Learning in Neuroscience): mbl.edu/education/adva… @iaifi_news summer workshop @MIT: iaifi.org/summer-worksho… @Princeton machine learning summer school (4 talks on spin glasses, phase transitions, and diffusion models): mlschool.princeton.edu Massey public lecture at Woodshole @MBLScience on the Science of Intelligence @MIT_CBMM Brains Minds and Machines summer course: bmm.mit.edu

Mathematics and computing have had a very tight relationship in some corners of academia. Richard Courant proposed methods to solve PDEs numerically in 1928. Von Neuman figured out you could use electronic computers for it in 1940. Both were mathematicians. These methods are the ancestors of finite element methods, for which NYU's Courant Institute is famous. The Courant Institute was created in 1935 as NYU's mathematics department. It was unusual in that there was a lot of work on numerical analysis and scientific computing, with heavy use of supercomputers. In 1952, the Atomic Energy Commission installed a powerful electronic computer at NYU, which led to the creation of the Courant Mathematics and Computing Laboratory. The computer science department was spun off from mathematics in 1969 by Jacob T Schwartz. Courant is now a school with three departments: mathematics, computer science, and data science. The continued existence of an HPC center at NYU is one factor that enabled AI research to take off.

A Structural Antibody Benchmark of AlphaFold3 reveals Hallucinated Epitopes and a Bias for Orderness 1 They benchmarked AlphaFold3 (AF3) for antibody–antigen binding on a large, structure-grounded dataset: 3401 experimentally validated complexes from SAbDab plus 23798 negative controls (random human proteins and six fixed “many-to-one” decoys) to quantify both true hits and false positives. 2 Using AF3 confidence at the interface (PAE and IPTM) as a binary “bind / not bind” classifier (PAE < 2.5 or IPTM > 0.7), AF3 achieved recall ~0.46 with precision ~0.64 (F1 ~0.53) on 10-seed inference; an intrinsic false positive rate of ~3% appeared consistently across negative-control batches. 3 A key qualitative failure mode: AF3 can hallucinate plausible antibody–target interfaces on decoy proteins, distributing predicted epitopes across much of a folded target’s surface. These false-positive “hotspots” suggest AF3 can generate convincing but non-biological interfaces even when no binding should exist. 4 Another notable bias: AF3 performed better on complexes whose experimental structures came from X-ray crystallography than from electron microscopy. In their split, X-ray-derived positives had higher recall (~0.52) than EM-derived positives (~0.30), consistent with training-set composition and/or higher flexibility/disorder typical of EM-captured systems. 5 Target size matters: positive prediction rates decreased as antigen length, surface area, and volume increased (i.e., larger search space → fewer detected binders). This trend was visible across both positive and negative controls, and was strongest when using surface area/volume rather than sequence length alone. 6 Disorder interacts with epitope discovery in a specific way: while overall target disorder fraction did not strongly separate predicted positives vs negatives, AF3 tended to avoid disordered regions when hallucinating interfaces (e.g., on PD-L1/CD274 decoys, false-positive antibodies clustered on ordered domains rather than the disordered terminus). This supports the paper’s framing of a bias toward “orderness.” 7 They tested whether performance was inflated by training data leakage (PDB overlap with AF3 training set). Although many benchmark antibodies were “leaked,” leaked vs non-leaked antibodies showed no meaningful difference in AF3 confidence distributions, suggesting leakage was not the dominant driver of outcomes in this benchmark. 8 Antibody-sequence features were surprisingly not major confounders: amino-acid composition across CDRs and CDR lengths showed little to no correlation with AF3 interface confidence (PAE), arguing AF3 is not simply favoring certain residues or longer/more flexible loops when it predicts binding. 9 Epitope-centric analysis added nuance beyond RMSD: using DockQ plus two custom measures (epitope shift and antibody displacement), they found many “false negatives” still landed near the correct epitope. About 34% of false negatives had epitope shift < 10 Å despite poor DockQ, implying that downstream refinement (e.g., MD, docking, more recycling/relaxation) might rescue additional true interactions. 10 Increasing inference seeds improves recall but is costly: rerunning 1915 false negatives at 100 seeds recovered 344 additional true positives, raising recall to ~53%. They also note AF3’s internal model selection by ranking score may not always pick the lowest-PAE interface, suggesting alternative selection heuristics could help in screening workflows. 📜Paper: biorxiv.org/content/10.648… #AlphaFold3 #Antibodies #ProteinStructure #ComputationalBiology #DrugDiscovery #Benchmarking #ProteinProteinInteractions #Bioinformatics #StructuralBiology #MachineLearning

Biology does not lack equations. It probably lacks a defensible way to know when those equations written for different parts still describe the same biology. A gene-regulatory model can fit an equation, a tissue-mechanics model can fit, and a transport model can also fit, still the explanation can fail exactly when one hands meaning to the next equation. In this essay I present an opinion on when an equation becomes an equation of life. cellbiosf.substack.com/p/when-does-an… #AI #science #biology #mathematics

Great to have participated in preparing the @ScienceBoard_UN brief on tipping points with clear definitions and also on how AI is impacting the field. Our paper on using Neural Operators for early detection of tipping points by measuring deviation against baseline physics is helpful in a number of areas: from climate tipping point in stratocumulus cloud cover to airfoil wake and stall transitions using only limited knowledge of the governing equations. arxiv.org/abs/2308.08794 @mliuschi @Caltech

UN Scientific Advisory Board @ScienceBoard_UN

3 months ago

🔬 New Science Brief! 🌍 What happens when environmental change crosses a critical threshold? From melting ice sheets to coral reef collapse and changes in ocean circulation, some Earth systems may be nearing tipping points. 🔗Brief:tinyurl.com/mrar5t2m @PIK_Climate

keyboard_arrow_left Previous keyboard_arrow_right Next

0 6 7 4K 3

Machines still don’t speak biology. A new Cell Perspective argues that bigger datasets, larger models, and more GPUs won’t automatically produce a universal model of life. The missing piece: process-aligned, causal world models connecting molecules, cells, tissues, space, and time. Biology is not just sequence. It is history, dynamics, and emergence. cell.com/cell/fulltext/…

Our new paper brilliantly lead by @itamarlandau "A predictive theory of experimental design for inferring neural population geometry in large-scale recordings." biorxiv.org/content/10.648… Our theory can say a lot about the geometry of new data *before* it is collected, by extrapolating from past data. We quantitatively predict: 1) How neural dimensionality, and the reliability of neural correlations and individual neural PCA modes grows with neurons and trials. This allows one to design experiments before the data is collected. 2) We find a "blessing of dimensionality" whereby recording more neurons actually allows us to record *fewer* trials while still reliably inferring neural population geometry. This opens the door to new types of experiments with many more complex trial types. 3) We develop scaling laws for neural prediction using masked autoencoders - a key technology for building foundation models in neuroscience. In the simple setting of linear autoencoders on a single session, we find power law behavior of prediction performance and autoencoder size with neurons and trials, and we trace these power law exponents for neural prediction to power law signals in collective neural modes of the brain. 4) We test our theory across multiple species (mouse, monkey, human) and recoding modalities (electrophysiology, calcium imaging, and fMRI). See @itamarlandau's excellent thread for more information: x.com/itamarlandau/s… Also, yet another fun collaboration with Mark Schnitzer!

Itamar D Landau @itamarlandau

2 months ago

Neuro experiments capture 10,000+ neurons but often with only a ~100s of trials. So the data is ~100s points in a 10,000-D space. Can we trust the extracted "population geometry"? This motivates our new paper: a predictive theory of experimental design. @SuryaGanguli 1/13

1 17 92 25K 50

Kimi K3 ~ The secret ingredient of the secret ingredient sauce... My essay on Kimi K3, Jevons, and the duration error inside frontier AI valuations cellbiosf.substack.com/p/the-half-lif… Model alpha is now decaying faster than the capital used to create it can be recovered. Kimi K3 did not need to become the world’s strongest model. It only had to prove that frontier replication is becoming repeatable enough to destroy the duration assumptions embedded in current valuations. Wall Street first sold the entire AI stack, then reversed as it separated the collapse of model scarcity from the growth of infrastructure volume. A six-month capability lead cannot justify a multi-year scarcity premium, regardless of how much capital produced it. Jevons paradox sharpens the contradiction. As inference gets cheaper, enterprises will spend the savings on more tasks, longer agentic search, retries, tools, critics, and verification. Model margins can compress while aggregate demand for compute, memory, networking, and serving expands. The Dragon Scroll was empty because scarcity itself was the moat. Once that scarcity disappears, durable value moves to whoever can measure where each model works, verify the outcome under real constraints, and reallocate execution before the next release erases the edge. The frontier now becomes a continuously repriced market. Model leads will appear, decay, and migrate. Jevons will push more intelligence through the system even as rents at the model layer compress. READ MORE on the link. #AI #KimiK3

@CellBioSF PDF download link here: freedompreetham.org/_files/ugd/f88…