|
Claude was trained on protein sequences at scale
Research testing various AI models on predicting amino acid sequences in little-studied proteins reveals that only the latest Claude models (Fable 5.2 and Opus 5.5) show evidence of being trained on protein sequences at scale, while other frontier models show minimal to no such capability. The study used a "next-amino-acid prediction" test on 668 obscure bacterial and archaeal proteins with no published structures or significant literature, finding that Claude models significantly outperform competitors by correctly predicting sequences that become impossible when the protein is shuffled.
Read Full Article →
← Back to latest