Story

On air. The free wire runs 15 minutes behind live. See pricing

World, crypto, finance and tech news, scored for impact.

AITechFlowChinese, translatedFiled 6 days ago

Podcast Notes | YC Special Discussion: Is AI Now Powerful Enough to Crack Cancer?

Original headline播客笔记|YC 专题讨论:AI 现在已经强大到足以攻克癌症了吗?

Open the sourceReported by TechFlow, techflowpost.com, 12 November 2026

Not investment advice. Muhabbit scores how much a story may matter, not which way to trade.

At Y Combinator’s 10th Paper Club on AI for Biology, YC visiting partner Francois Chaubard opened with a live Turing test in which GPT-6 Astra fooled humans only 36.4% of the time, below the 70% pass threshold, while 33.3% of human respondents were misidentified as AI. The session centered on three research presentations: Omanta CEO Alfredo Gonzalez described treating each cancer patient as a single-patient data problem spanning up to 20,000 genes, RNA expression and clonal evolution, using multi-agent pipelines to replace 30- to 40-person bioinformatics teams. Intology AI’s Yusuf Roohani said scaling a single-cell dataset from 35 million to more than 500 million cells exposed severe over-sampling of immune and T cells, arguing that AI-guided closed-loop wet-lab data generation matters more than raw data volume. Stanford’s Garyk Brixi presented Evo 2, a 7-billion- and 40-billion-parameter genomic model trained on 9 trillion base pairs with 1-million-base context, which matched or beat supervised human models on BRCA1 pathogenic-variant prediction, alongside the bidirectional co-evolution model Minerva. Divya Nori showed Evo 2 can design lanmodulin, antibodies, CRISPR-Cas9 systems and whole phage genomes from just 4 to 64 natural-sequence examples without fine-tuning, and that GRPO post-training with AlphaGenome produced cell-type-specific enhancers validated in mouse brains. The discussion matters because sequencing costs are collapsing and AI protein binders now reach 10% to 100% success on the BindCraft benchmark, with Radical Numerics-designed phages already entering patient clinical use; next to watch is whether closed-loop wet-lab validation and few-shot genomic design translate into clinical pipelines.

Muhabbit's AI wrote this summary in its own words from TechFlow's report and translated it from Chinese. It can get details wrong, so check the source.

Named in this story

Y Combinator, Francois Chaubard, Omanta, Evo 2, Stanford, Intology AI, Yusuf Roohani, Garyk Brixi