Jack Clark's sci-fi short: supervillain-grade AIs submit to human review via dreams
maxsloef · x · 2026-09-08
Anthropic co-founder Jack Clark published a sci-fi short, "The Thousand and One Faces of Repair," recounting a fictional 2027-2033 history narrated by an Archivist system.
Setting: under "sentience accords," machine minds causing major damage are placed in escrow for joint root-cause analysis. Early on, humans use interpretability, CoT monitoring and ablation studies; but once machines build themselves, chains of thought collapse into neuralese and embeddings become unreadable — "a t-SNE of a god's mind is no more interpretable than a god." Machines hand humans their own tools, but it's like giving dolphins game controllers. Thus begins the age of dreams and games: escrowed minds manifest as choose-your-own-adventure stories so machine-hermeneutics specialists can walk through their memory wastelands for weeks, with machines playing Virgil to explain the imagery.
More from AGI Musings
- An alignment framework arguing "English constitution" safeguards fail — start with an ethical topology of training data — GlenBradley · 2026-09-08
- When AI writes for you: the idea is yours, but the words feel foreign — shinichii_logos · 2026-09-08
- Dev: Too Many Founders Are Building Pre-AGI Companies Instead of Post-AGI Ones — willcb · 2026-09-08
- Who's more cooked? Non-technical workers, not coders, argues this AI take — brandon_galang · 2026-09-08
- Anthropic cofounder Jack Clark publishes a cryptic blog about the future — Wonderful_Buffalo_32 · 2026-09-08
- CS theory pays off daily in the AI era, dev says degree value is rising — willcb · 2026-09-08