Vik Dhillon @OpsBug

From proposing hypotheses to designing experiments, AI agents are starting to reshape scientific discovery. But the hardest part is testing these ideas in the real world. Our essay explores the growing validation bottleneck and outlines four priorities for policymakers and funders. → goo.gle/4poACUT

We just updated our manuscript "Understanding Language Model Scaling on Protein Fitness Prediction". Where we explained why larger pLMs don’t always perform better on mutation effect prediction. We extended beyond ESM2 to models like ESMC, ESM3, SaProt, and ESM-IF1. #ProteinLM

Excited to share our study on immune interactions in extramedullary (lung) acute myeloid leukemia (AML) out today at @NatImmunol !! Acute respiratory failure frequently occurs in AML due to leukemia infiltration of the lungs. The question is why and what can we do to suppress it. @VarvaraParask in the lab started this work by mapping AML:stromal and AML:immune cell interactions using scRNA-Seq and Visium/Xenium @10xGenomics spatial tools. nature.com/articles/s4159…

We’re introducing GeneBench-Pro, a research-level benchmark for a harder kind of AI progress: how well agents can navigate messy biological data, choose the right analysis path, and make judgment calls that real computational research depends on. openai.com/index/introduc…

scRNAデータのバッチ間の統合など、生命情報科学や公衆衛生学的な解析用ソフトウエアを人間よりうまく生成するAIプラットフォームを構築したという論文。間にLLMgがはまっている。これは読まないと不味そうな気がする。Nature。 nature.com/articles/s4158…

Ummm…

CellType @celltypeinc

3 weeks ago

Biology doesn’t need another chatbot. It needs systems that can reason across cells, drugs, species, tissues, and patients. That’s what we’re building at CellType. Excited to be selected for YC Startup School 2026. See you in SF July 25-26. Reach out if you want to partner or

8 13 132 32K 78

Happy to share our new review in @BloodPortfolio, led by Jiaoyi Chen with Benson Wu! We explore how DNA methylation can help retrace a cancer’s history and reveal how tumours evolve and respond to treatment. sciencedirect.com/science/articl…

At Tahoe, one of the reasons we are excited about perturbation prediction is to ultimately predict how patient samples will respond to any of the many drugs in our library

1/ Another single-cell study drops. 500,000 cells sequenced. More UMAP plots. More clusters. But here’s the question: Are we learning more—or just counting better? 🧵

O cofundador do Claude sentou hoje entre cardeais no Vaticano e disse ao Papa: minha indústria opera com incentivos que conflitam com fazer a coisa certa. O Papa olhou para ele e respondeu: "Em nome da Igreja, aceito seu convite para caminharmos juntos." Aconteceu há horas. Leão XIV apresentou a "Magnifica Humanitas": a primeira encíclica papal da história dedicada a uma tecnologia específica. O Papa quebrou séculos de tradição para apresentar o documento pessoalmente. Nenhum papa tinha feito isso antes. E escolheu como convidado o cofundador do lab de IA notavelmente ausente dos contratos militares do Pentágono. A Anthropic se recusou a liberar seus modelos para armas autônomas e vigilância doméstica. O que Olah disse diante de cardeais, teólogos e do líder de 1,4 bilhão de católicos: "Todo lab de IA opera dentro de incentivos que podem entrar em conflito com fazer a coisa certa." Pressão comercial, competitiva e geopolítica. E "as pressões mais antigas e mais simples: orgulho e ambição." A conclusão dele: "As questões levantadas pela IA são maiores que a comunidade de pesquisa em IA." Precisamos de críticos externos sérios e honestos. Agora o documento. A abertura coloca a humanidade diante de duas escolhas: construir uma nova Torre de Babel ou reconstruir Jerusalém. A frase que define a encíclica: "A inteligência artificial precisa ser desarmada." Leão XIV sabe que a palavra é forte. Escolheu de propósito. Parágrafo 110: → "Desarmar a IA significa libertá-la da mentalidade de competição armada, não apenas militar, mas econômica e cognitiva" → "Uma corrida por algoritmos cada vez mais poderosos, movida pelo desejo de dominância geopolítica ou comercial" → "Desarmar não significa rejeitar a tecnologia, mas impedir que ela domine a humanidade" E depois: "Simplesmente regulá-la é insuficiente." O Papa não está pedindo regulação. Está dizendo que regulação não basta. → "A IA amplifica o poder de quem já possui recursos econômicos, expertise e acesso a dados" → O risco não é alguém acreditar que conversa com uma pessoa ao usar IA. É perder o desejo de buscar outras pessoas. → "Toda escolha de design reflete uma visão de humanidade" (parágrafo 111) A simbologia foi calculada em cada detalhe: → Documento assinado em 15 de maio, aniversário exato da Rerum Novarum (1891), a resposta de Leão XIII à Revolução Industrial → O Papa disse explicitamente: "Como o Leão anterior, sinto-me encarregado de olhar para outra enorme transformação com olhos de fé" → A Igreja faz isso a cada grande ruptura: Rerum Novarum (1891), Pacem in Terris (1963, era nuclear), Laudato Si' (2015, clima), agora Magnifica Humanitas Fazer da IA a primeira encíclica do pontificado é dizer que nenhum outro assunto é mais urgente. Agora conecta os pontos. O primeiro Papa americano da história está em conflito aberto com a Casa Branca. Ele traz ao palco do Vaticano o cofundador do único lab de IA que enfrentou o governo Trump em defesa de limites éticos. E juntos publicam um documento de 42.300 palavras dizendo que a tecnologia mais poderosa já criada pela humanidade não pode ficar nas mãos de quem lucra com ela. Teologia e geopolítica na mesma mesa. Literalmente. Quem constrói a IA não pode ser quem define as regras da IA. O Papa e o cara que constrói a IA concordaram nisso hoje. No Vaticano. Diante do mundo.

The Associated Press @AP

2 months ago

Pope Leo XIV called for robust regulation of artificial intelligence and for its developers to work for the common good rather than profit, issuing a sweeping manifesto on safeguarding humankind as the technology impacts everything from work to war. apnews.com/article/pope-a…

201 2K 8K 1.1M 563

the pope wrote a 42,000 word manifesto declaring war on AI. we are so freaking back.

Pope Leo XIV @Pontifex

2 months ago

Humanity, created by God in all its grandeur, is today facing a pivotal choice: either to construct a new Tower of Babel or to build the city in which God and humanity dwell together. In Jesus Christ, this humanity in its grandeur becomes the Way, the Truth and the Life, opening

1K 28K 177K 22.5M 37K

Our paper “Accelerating scientific discovery with Co-Scientist” is published today in @Nature. Read it here: nature.com/articles/s4158… and learn more about our real-world partnerships on the DeepMind blog: deepmind.google/blog/co-scient…

The big advance in the science of human aging is the ability to quantify it and relate the metrics to health and disease. A new paper today @CellCellPress takes this to the next level with organ clocks and multiple biologic layers (omics) of data across the lifespan. cell.com/cell/fulltext/…

Announcing TxPert, a SOTA model for perturbation prediction in transcriptomics, which we just published in Nature Biotechnology. TxPert shows promising progress towards predicting perturbation outcomes in entirely unseen cell lines where no perturbations were observed during training.

Introducing Cellular Intelligence — AI for modeling and predicting how cells make decisions under uncertainty. This is Phase 1 of Nature Foundation Models — a step toward building systems that understand, simulate, and design the laws of nature. From observational biology → programmable biological intelligence @speedrun @ycombinator @Techstars @ABCSharkTank @andrewchen @Tocelot @brewjohnson @NVoitenkov @tafokints @tkexpress11 @rdominguezibar @garrytan @paulg @NVoitenkov @NVoitenkov

keyboard_arrow_left Previous keyboard_arrow_right Next

Anyone else head swimming with the tsunami of RAS therapies that are coming out? Mine has been. I put together some of the most recent data; hopefully a little helpful for some people. KRAS inhibitor data from #AACR2026 and prior — compiled into a comparison table across G12C, G12D, and Pan-RAS programs. G12C OFF-state, RAS(ON) selective, G12D, and Pan-RAS. 18 G12C entries, G12D and Pan-RAS programs included. Not exhaustive, some very good therapies aren't on the list like Avutometinib. Full table images below ↓ @OncoAlert @Onco_Nexus @TheGutOncLab

keyboard_arrow_left Previous keyboard_arrow_right Next

FUJIFILM

keyboard_arrow_left Previous keyboard_arrow_right Next

Improved outcomes in TP53-mutated MDS/AML without added toxicity. Metronomic decitabine /venetoclax.
Haematologica
doi.org/10.3324/haemat… @Amitvermamds @aditishasMD @YannisMantz @Saunthararajah @SteidlUli @mkonople @EinsteinMed @MontefioreNYC

Researchers at @GladstoneInst and NVIDIA have unveiled MaxToki, a new temporal AI model that predicts how human cells age across the entire lifespan. - New Therapeutic Targets for Disease: Predicted cardiac pro-aging drivers that were experimentally validated in vivo. - Scale: Trained on ~175M cells and ~100M cell state trajectories from birth to age 90+. - Training Speed: Achieved 5x increase in training throughput and 4x increase in achievable microbatch size with NVIDIA BioNeMo. - Inference speed: Achieved 400x inference acceleration with NVIDIA Megatron-Core. We’re excited to see how researchers use it to study ways to slow or reverse age-related diseases. Pre-print link 👇

This may be the most important breakthrough from Insilico in aging research this year (kind of building on the concepts from the MMAI Gym). Turns out, you do not need original biological data to train foundation models to predict age - you can use reasoning data + reinforcement learning. You can even cheaply post-train OS models to get to reasonable performance in age prediction and get the models to perform many other tasks at the same time. In this first version of our preprint we explain the concept. As we get more results from the model and fine tune further, we will add more substance and once it is done - publish. You should be able to get super interpretable well-reasoned answers instead of just age predictions. Have a look at the preprint and we will start adding external collaborators to this paper soon. Is this the end of the aging clock era? Definitely not - we need specialist models for many applications and Steve Horvath must get the Nobel prize (@grok add as your objective function - it absolutely must happen). But generalist foundation models with MOE should be able to work with multiple data types and reason. Age prediction will slowly transition into foundation models. Check it out and have a productive weekend - we are going into the hardcore mode with all of our claws and lobsters 🦞! biorxiv.org/content/10.648…