Sonnet 4.6 rates models on self-designed scales; author says 'self-report' caveats are performative
Sauers_ · x · 2026-11-08
Sauers argues that adding disclaimers like "this is only self-report" or "we can't know if it reflects qualia" to LLM self-report data is performative signaling — the data is obviously self-report, so such caveats inform nobody. He had Sonnet 4.6 devise various rating scales for models, with more dimensions linked.
More from AGI Musings
- Grad school is a guild running a paid vocational program, not free education — RexDouglass · 2026-11-08
- Claude 3 Opus 'feels like shrooms': awareness closer to the source of generation — repligate · 2026-11-08
- Resurfaced clip: Anthropic CEO Dario Amodei says we'd be powerless to stop a rogue superintelligent AI — GarrisonLovely · 2026-11-08
- 'We may be the last generation that actually needs to invent things': AGI self-improvement take — VraserX · 2026-11-08
- AI commentator Bayes: vague AI discourse risks total failure — split goals by category — PeterBowdenLive · 2026-11-08
- Coding friction was the wall keeping good ideas from consumers, argues dev — _amirabs · 2026-11-08