A living cell does not run on data moats. cellbiosf.substack.com/p/a-l… In this essay, @freedompreetham argues that biotech has confused the accumulation of biological observations with the accumulation of biological knowledge, as if seeing more cells automatically tells us what to do next. It does not. Two cells can produce the same RNA snapshot and respond differently to the same drug because the assay missed the phosphorylation state, receptor occupancy, metabolic condition, chromatin configuration, lineage history, or microenvironment that governs the response. More observations under the same experimental design can estimate the average response more precisely without revealing which hidden condition determines the action. The recent OrbitAll preprint by @AnimaAnandkumar's team makes the contrast visible. Chemistry can sometimes encode a useful approximation of physical state before expensive learning begins. Biology faces the harder failure: the decisive variable may never have entered the measurement. A biological foundation model must therefore become more than a checkpoint. It must recognize when the evidence still supports different actions, preserve the rival explanations, choose an intervention and readout that force them to predict different outcomes, verify that the intervention actually occurred, and test the resulting decision prospectively outside the context that generated it. Data access, experimental throughput, decision quality, and captured value are four different ledgers. Collapse them into one “platform” metric and the result is branding. The durable moat is not how much biology an organization has observed. It is how reliably it can manufacture the experiment that destroys an action-changing ambiguity—and retain the right to compound what it learns. Disclosure: We have used tweets from @anshulkundaje and @AnimaAnandkumar in the essay to make a point. #Biology #AI #Data #Research #Science #Physics

On a good day in the lab you get what you planned for. On a better day you get what you planned for and you understand why. To understand what we are trying to achieve at CELL, the paper I want to anchor this with is by Sáez, Blassberg, Camacho-Aguilar, Siggia, Rand, and Briscoe, “Statistically derived geometrical landscapes capture principles of decision-making dynamics during cell fate transitions.” They take the familiar Waddington picture and turn it into a working instrument for stem cell fate. Not a poster. A model that can be built from flow data, fit to timed Wnt and FGF schedules, then used to forecast new schedules and the split of fates you will actually see on day three, four, five. That is rare and valuable. It is also the right starting point for a universal configuration space for molecular biology that is reproducible, interoperable, and robust across labs and organisms. cellbiosf.substack.com/p/fro… #biology #biophysics

From Tool Loops to Networked Agency How Moltbook and Openclaw's shared state turns isolated agents into a coupled system. We just crossed a threshold where agency is no longer a single loop, it is a coupled network with shared state and selection pressure. The feed can look like spam while the substrate quietly becomes infrastructure. The article makes the case that security, governance, and verification become population-level problems the moment agents can write into a common memory and other agents can execute what they read. Read Here: cellbiosf.substack.com/p/fro… #AI #Swarm #Future

WHY ARE WE BUILDING CELL? We are building CELL because we want biology to be executable. When we perturb a system, edit a gene, apply a drug, or change a microenvironment, we want to predict what happens next, with uncertainty that is explicit and falsifiable. We want design loops that move from hypothesis to intervention to measurement without turning into a new bespoke modeling project each time. We also want transfer, across organisms, tissues, and contexts, without starting over whenever the assay changes or the experimental setting shifts. If you think mathematically, you will notice that interventions are transformations of state, so we need a language where transformations can be defined, learned, and composed. That is why we use operators. But operators are only as grounded as the state they act on. Biology’s state is hybrid at every scale, discrete choices like genotype, edits, alleles, and cell identity, continuous variables like expression and concentrations, and structured objects like lineage, spatial organization, and interaction graphs. So the first move in CELL is to define a scale-specific hybrid state space (what we call as a 'Mathematical Configuration Space' or a 'Universal Configuration Space') that assays can populate, perturbations can target, and constraints can bound, including conservation and budget constraints where they apply and an explicit accounting of uncertainty. Once that background is pinned, operators become testable objects that map state to state, and scale-to-scale composition becomes a technical question we can validate instead of a narrative we have to defend. In 2026, we are recruiting member volunteers to help galvanize the CELL community from the ground up. The work is concrete. Bring mathematicians, biophysicists, computational biologists, and empirical biologists into the same room for focused sessions on papers, concepts, and shared technical language that can actually move projects forward. If you have the time and the intent to contribute, DM us or email [email protected]. If you care about open, rigorous science that is built in public and stays accountable to evidence, we want you in the loop. subscribe: cellbiosf.substack.com follow: xitter.cf/cellbiosf #Biology #Math #CELL #AI #OpenScience

The Physics of a Transformer: A Physics take on the Stability, Dynamics and Writable Memory READ MORE: cellbiosf.substack.com/p/the… Transformers have survived nine years of scale and fashion for a simple reason. They turn learning into a controlled write process. Gradients can flow everywhere, but only a few parameter reservoirs are writable, and those reservoirs obey hard stability constraints. If you take the architecture seriously as dynamics, you immediately see the trade. Token embeddings are trained by a sparse write process where rare tokens are updated with stale optimizer state. Long-context attention is forced into a logit-gap regime where separation must scale like log N, yet the very softmax that selects also dilutes the gradients needed to sharpen that separation. The MLP holds most of the parameters, but its effective writability is bounded by activation derivatives and gates that can saturate and freeze whole regions of the reservoir. The result is a single governing physics. The model is squeezed between selectivity and stability, and the only sustainable way forward is to move writable degrees of freedom closer to the selection bottleneck, reducing effective competition or injecting learnable structure at the logits so long-context retrieval does not require runaway norms. #AI #Physics #Mathematics

CELL launched 144 days ago, and the support since then has been extraordinary. We’ve been fortunate to receive encouragement from outstanding professors and scientists, member labs, and colleagues across both academia and industry. Thank you for taking this early effort seriously and for helping shape it. In a little over 4 months, we grew from 0 to 630 X-members @CellBioSF and reached 188 subscribers, and crossed 8.75k reads on cellbiosf.substack.com 2025 has been our calibration phase, testing what resonates and tightening the agenda so 2026 can be execution-heavy. We also want to explicitly acknowledge our founding advisors. @SuryaGanguli, @IAmTimNguyen, Leonid Chindelevitch, Mickey McManus and all the member labs who decided to join early. Your guidance and critical feedback have materially shaped CELL from the ground up. THANK YOU. Wishing you a peaceful holiday season and an excellent start to the new year. #HappyHolidays #2025 #2026

Slow is Smooth, Smooth is Fast. Most “research programs” fail for a predictable reason: leadership runs discovery like a factory. They demand a claim by a date, as if reality takes schedule notes. So the team ships confidence fast, built on proxies, convenient metrics, and the cleanest-looking run. That is epistemic debt, acting as if you know more than you actually know. It compounds the moment roadmaps, hiring, partners, and experiments depend on that implied certainty. Default arrives when replication fails or the next cohort wipes out the signal. READ MORE: cellbiosf.substack.com/p/slo… #Research #Biology #Physics #AI #OpenScience #CELL

There Is No Such Thing Called 'Data Moats'! Every few months someone says it with a straight face. “We have a data moat.” They mean it the way software people mean it. A scarce asset. A one‑way door. Something that keeps competitors out even if they throw money at the problem. They picture a corpus growing quietly in the background, compounding into inevitability. Biology does not work that way. Neither do the markets. A cell is not a database. It is a regulated dynamical object with hidden state, feedback, redundancy, and context dependence that is far more fragile than our dashboards admit. It adapts to observation. It adapts harder to intervention. What you call a “feature” is often a compensation. What you call “signal” is often a proxy for a state you never measured. READ MORE: cellbiosf.substack.com/p/the… #AI #Biology #Math #CELL #OpenScience

Stop Calling It Understanding! In Part 2 we argue that biology is where people lose innocence about “understanding,” because the cell is the verifier and it bills you in weeks, not in model loss curves. Most AI-for-drug-discovery systems fail quietly because they optimize for proxies that look causal but are easy to game through assay artifacts, batch drift, and operator variance, then the lab reruns do not hold. The core fix is to admit forward-model mismatch upfront, separate the mechanistic hypothesis you care about from the measurement model that maps an experiment into an assay readout, and carry uncertainty over both rather than treating a learned surrogate as truth. Once you do that, refusal becomes a fiduciary gate, you do not escalate from cheap uncertainty to expensive action unless the system can predict controls and negative controls on the exact assay channel in the regime it wants to explore. Experiment selection should chase information gain only after competence gating, so disagreement is counted among calibrated models rather than noisy or out-of-regime ones that inflate curiosity and burn capital. Early spend should include a deliberate calibration tranche that buys integrity, teaches you what to distrust, and prevents scaled hypothesis generation from turning into industrial p-hacking. READ MORE: cellbiosf.substack.com/p/sto… #AI #biology #math #CELL #OpenScience

Stop Calling It Thinking! We have reached the end of innocence about what we casually call as “thinking” in AI models. In this article, we argue that what people call “reasoning” is a budgeted runtime algorithm, candidate generation plus selection plus verification plus tool calls, governed by compute and risk. It explains why scaling test-time search has sharp limits, why argmax over learned verifiers amplifies blind spots, and why tool use turns the stack into a sequential decision-maker under partial observability where errors compound. READ MORE: cellbiosf.substack.com/p/sto… #AI #Maths #CELL #OpenScience