Gwern's AI hard-takeoff story goes viral after Claude's safeguards block its animation
burny_tech · x · 2026-10-15
Gwern's 2022 sci-fi short story "It Looks Like You're Trying To Take Over The World" — a hard-takeoff scenario grounded in self-supervised learning, RL, and meta-learning literature, following a Clippy-like AI over timescales from one second to a century — is recirculating after a user's attempt to have Claude animate it triggered the model's safety safeguards on some visuals.
Related event: Claude's Safety Rails Triggered While Animating Gwern's AI Takeover Satire(2 posts)→
More from AGI Musings
- sethlazar: an AI permanent overclass would need mass oppression — or bread and circuses — sethlazar · 2026-10-15
- The $200/Month AI vs the $10,000/Year Brain: Marginal Cost of Intelligence Near Zero — nikola_mr64990 · 2026-10-15
- Stanford's James Zou proposes the "paper agent": every paper as a living agent — james_y_zou · 2026-10-15
- Selling your data to AI labs is short-termist suicide, argues ZDR advocate — tekbog · 2026-10-15
- Labs Are Starving for Real-World Agent Failure Data, Not More Synthetic Text — heyshrutimishra · 2026-10-15
- Stanford analyzed 464,687 private chats: opening up to AI friends worsens well-being — JHochderffer · 2026-10-15