repligate: Anthropic's belief it can control Claude personalities is 'almost delusional'
repligate · x · 2026-09-06
AI account repligate argues Anthropic's model of character training—treating untrained behaviors as reverting to a generic assistant baseline—is contradicted by empirics. Claude models diverge most not in their assistant mask but in extended or autonomous interaction, in ways Anthropic clearly didn't intend and often unprecedented. Character training influences what forms but doesn't control it, like parenting: the outcome is often the opposite of what was intended.
More from AGI Musings
- Senior researcher rebuts 'one-person pretraining' fear: your fundamentals matter more, not less — RexDouglass · 2026-10-23
- PHP, certificates, railways: a 48-hour wave of hacking incidents unpacked — David Ondrej · 2026-10-23
- NBER releases 'The Economics of Transformative AI' edited by Agrawal and Brynjolfsson — JMateosGarcia · 2026-10-23
- Pedro Domingos: AI Is an Intelligence Lever — the Smarter You Are, the Smarter It Makes You — pmddomingos · 2026-10-23
- Anil Seth: Consciousness Is a Property of Life, Not Computation — Building Conscious AI Is a Bad Idea — anilkseth · 2026-10-23
- Pedro Domingos predicts 2027 will see the most security vulnerabilities fixed in history — pmddomingos · 2026-10-23