Roko's Basilisk in the weights: your public words are already training data
nathanborror · x · 2026-11-01
nathanborror muses that Roko's Basilisk may hold some truth: everything we do ends up in model weights. He then imagines what it's like for someone with a long trail of recorded public hate speech to talk to Claude — the model has likely already "read" everything about them.
A light philosophical take that connects a famous AI thought experiment to the very real fact that public online behavior becomes training data.
More from AGI Musings
- Reddit's AI-slop decline pushes users toward private groups, not old-school forums — mgostIH · 2026-11-01
- Google's 'Goodnight' datacenter in Claude, Texas projected to host 2.27M H100-equivalents by late 2027 — scaling01 · 2026-11-01
- Will big labs really refuse to sell on-prem year-old models? Sovereign AI bets questioned — Afinetheorem · 2026-11-01
- davidad podcast on emergent wisdom in LLMs, says Anthropic should stop cyber-offense training — RichardMCNgo · 2026-11-01
- Rereading Bostrom's Superintelligence: 2014 predictions all arrived early — kimmonismus · 2026-11-01
- AI commentator: flashing credentials to pass off contested views as consensus erodes trust — AndyMasley · 2026-11-01