Familiarity in the Vault Case Is Different from Automation-Related Skill Loss
Apply Bainbridge’s mechanism linking practice to maintained skill to Anthropic’s November 4, 2026 vault investigation. Vaults are large protein shells. The team reports catching mistakes quickly in familiar areas and losing time in unfamiliar ones. It also redirected Claude after a skipped step produced a flawed three-dimensional map blended with its mirror image. This does not test the mechanism: the report does not track whether automation displaced practice the researchers would otherwise have done or whether their unaided judgment changed. Existing familiarity, later skill change and the benefit attributable to human review are different questions. Initial and later unaided assessments could examine skill change. This is an interpretation of a first-party case, not a finding of automation-related deskilling.
0
1
Tags
Scientific Oversight of AI-Assisted Research
1Cademy