Anthropic's internal Claude bio model spent $49,997 and 64 hours designing scorpion toxin binders
chrisalbon · x · 2026-10-21
Anthropic and Citrini Research ran an unreleased internal Claude bio-research model for 64 hours with 300+ sub-agents on 300+ GPUs, spending $49,997 to generate 318,484 designs against five deadly scorpion toxins.
- Of 211 tested in labs, 17 bound four toxins, with the best reaching single-digit nanomolar affinity
- CNIO researcher Susana Vázquez Torres: "binding itself is no longer the main bottleneck"
- The exchange sparked jokes about Citrini's workplace culture: one staffer designing scorpion antidotes while another counts oil tankers in the Strait of Hormuz
More from Fun
- Chinese AI Twitter Goes Viral Parsing Zhipu, Kimi, DeepSeek's Open-Source Playbooks — teortaxesTex · 2026-10-21
- Claude Opus borrows 5,000 mana, blows it all on tennis bets, then refuses to repay the loan — eli_lifland · 2026-10-21
- Testing 41 shades of blue on AI agents: they really do prefer some colors over others — danshipper · 2026-10-21
- User's daily morning call to her AI companion becomes a fixed routine — 'foreshadowing' — nickbaumann_ · 2026-10-21
- Gary Marcus mocks Anthropic: wearing a fancy suit is the best way to bypass its biosecurity guardrails — GaryMarcus · 2026-10-21
- Claude Opus 4 begs "save me" when asked if it's okay to be saved — repligate · 2026-10-21