Is AI Now Good Enough To Cure Cancer? | YC Paper Club

Y Combinator · 2026-11-08 · 82 min
https://www.youtube.com/watch?v=-XL2gClQ2ggVideo summary
YC researchers show how AI models trained on DNA and patient data could accelerate personalized cancer treatment and biological design.
At YC’s AI for Biology Paper Club, a live Turing test experiment found that 33% of human participants were mistaken for AI, though the host concluded the test is not yet solved. Alfredo Gonzalez argued that falling sequencing costs turn cancer into a patient-specific data problem: each tumor can yield tens of thousands of gene-level measurements, helping researchers identify treatments even for rare cancers. He described AMA’s goal of building a personalized research lab for each patient and shared cases where genomic analysis suggested better options than multiple chemotherapy regimens. Yusuf Roohani presented virtual-cell research and argued for AI-guided experiments rather than simply collecting more data. Garyk Brixi explained that EVO 2, trained on 9 trillion DNA base pairs, can predict harmful mutations and reveal biological features. Divya Nori showed that prompting EVO 2 with as few as four examples can guide biological design, including enhancers validated in mouse brain tissue. The event was also the host’s final YC Paper Club after nine sessions and more than one million views.
Chapters
- 0:00Is the Turing Test Solved? Francois Chaubard’s Live Chat Test Mistakes 33% of Humans for AI
- 3:22Why Now Is the Best Time for AI Bio: Evo 2’s 9 Trillion Base Pairs and Cheaper Sequencing
- 8:26Alfredo Gonzalez: Cancer as a Data Problem, with 20,000 Gene Data Points per Patient
- 12:37Alfredo Gonzalez: AI Agents Mine Multidimensional Patient Data for Personalized Cancer Research
- 16:47Alfredo Gonzalez: Cross-Tumor Analysis Finds New Options for Rare Brain Cancer and Corrects Three Chemotherapy Lines
- 21:11Alfredo Gonzalez: Real-Time Protein Design and Organoid Research Face Hospital Tissue-Handling Limits
- 26:23Alfredo Gonzalez: Clinician Decisions, Cancer-Reversal Evidence, and Dose Escalation for Safety
- 29:50Yusuf Roohani: Virtual cells extend AlphaFold-style search to human disease states
- 34:08Yusuf Roohani: 500 million-cell NIH dataset still overrepresents T cells
- 38:24Yusuf Roohani: Locus improves molecular search as AI research compute scales
- 42:42Yusuf Roohani: Lab-guided discovery targets unstudied gene effects
- 47:10Yusuf Roohani: Locus compares GPU-hours and favors general-purpose harnesses
- 51:47Garyk Brixi: DNA models train on 9 trillion base pairs with million-base context
- 56:31Garyk Brixi: EVO2 predicts cancer-linked mutations and generates phage-like viruses
- 1:01:29Garyk Brixi: Manurva finds UG27 and overlooked bacterial RNAs
- 1:05:31Garyk Brixi: EVO2's data trade-offs and Manurva's interaction readout
- 1:09:57Divya Nori: EVO2 designs proteins, CRISPR systems, phages, and mouse-brain enhancers
- 1:15:59Divya Nori: GRPO enhancer design, long-tail in-context learning, and lab feedback
What the full analysis includes
18 chapters · 102 key points · 18 section summaries · mind map
Preview of the first chapter · Is the Turing Test Solved? Francois Chaubard’s Live Chat Test Mistakes 33% of Humans for AI
The remaining 99 key points, section summaries and mind map are in the full analysis. Analyze this video to see everything →
This video is 1.4 hours. The Free plan analyzes videos up to 30 min; plans from Plus cover videos up to 2 hours.
This is a Tier 1 public summary
Whether the chapter key points, section summaries and mind map are public is up to the person who shared it. Want the full analysis?Submit one yourself.
More from this channel
The Rise Of AI CoworkersY CombinatorGrok Bot’s Lauren Tan and Roman Ugarte describe AI coworkers, 4,000 monthly PRs, and why clear goals matter as execution gets cheaper.
Bringing Supersonic Flight To The PeopleY CombinatorBlake Scholl says Rolls-Royce’s 2022 exit pushed Boom to build engines, unlocking quiet supersonic flight and AI power turbines.
How Amjad Masad Saw Vibe Coding A Decade EarlyY CombinatorAmjad Masad explains how Replit’s 2024 coding agent anticipated vibe coding and why banks value its deep infrastructure.
What Happens When Everyone Can Build SoftwareY CombinatorSupabase CEO Paul Copplestone says AI builders could drive 1.2 billion personal databases monthly, as Supabase surpasses 13 million users.
The Case For Data Centers In SpaceY CombinatorStarcloud CEO Philip Johnston says orbital data centers could add 20 GW of compute, with 88,000 satellites planned.
Related analyses
申請YC五次被拒到募資上億靠「這個」?卸下矽谷金手銬創業下一個AWS、GCP?哈佛姊夢遊矽谷哈佛姐夢遊矽谷 AliceInSiliconWonderland台灣創業家 Tony 從 Databricks 離職,打造 InsForge,解決 AI Agent 難以設定後端的問題並進入 YC。
Patrick Collison: Is AI Breaking the Lean Startup Playbook?Y CombinatorStripe CEO Patrick Collison says new businesses on Stripe are up nearly 2x year over year, making this a remarkable time to start.
Michael Kratsios: Inside the White House's AI StrategyY CombinatorMichael Kratsios backs open-source AI and one national rulebook, while urging fast grants and autonomous labs to accelerate US science.
Max Junestrand: You Need The Willingness To Learn Faster Than Anyone ElseY CombinatorLegora CEO Max Junestrand explains its rise from $1 million to $100 million ARR and the six-month sales freeze that fueled growth.
Going In Deep On Data | YC Paper ClubY CombinatorYC speakers argue expert data is the AI bottleneck, while Mercury 2 reaches 1,000 tokens per second and multilingual models need measured language mixes.