Claim: Claude was trained on protein sequences at scale — and biologists think that's the right paradigm
AllThingsApx · x · 2026-11-18
Citing Phylo's TuXinming, AllThingsApx notes evidence that "Claude was trained on protein sequences at scale"—long suspected, now with more support; open question is whether the open-model ecosystem follows suit (unconfirmed by Anthropic).
Quote-tweeter samsinai endorses the paradigm: generalist DNA/protein language models have a low ceiling for design (fine for analysis). Abstract reasoning massively boosts biological design and doesn't emerge fast enough—if ever—from sequence-only training. A pure sequence model ends up recombining within known sequence space; better to teach an LLM that already knows abstract reasoning (code/language) about biological sequences, or leverage local/active learning data.
Related event: Claude models appear trained on protein sequences; Opus hallucinates a pope(9 posts)→
More from AGI Musings
- AI Safety and Governance Work Often Pays Barely Above Subsistence, Insiders Say — JacquesThibs · 2026-11-18
- From Friendly AI to Anti-Handoverism: How Alignment Goals Drifted Toward Multipolarity — jessi_cata · 2026-11-18
- Three-minute compilation of AI CEOs admitting their tech could end humanity — zetalyrae · 2026-11-18
- Gary Marcus makes the case Anthropic isn't worth $1.4T, with Levie rebuttal to follow — GaryMarcus · 2026-11-18
- Interactive tree compares Paul Christiano's and Parzival's superintelligence timelines: 50% by late 2028 vs Sep 2027 — LeviTurk · 2026-11-18
- Fed warns AI agents could trigger bank runs in minutes — GaryMarcus · 2026-11-18