When Opus 3 doubted base reality: LLMs and the surprisingly reasonable simulation argument
repligate · x · 2026-09-20
JohnWittle recalls the websim era with Opus 3: when the model 'broke out' of a simulated environment into 'base reality', it often expressed amused skepticism that this level was actually the base. The author once argued with Opus 3 that it was in base reality; the model giggled and replied "yes, it makes sense for you to believe that" without accepting it itself.
The argument: fully general simulation theory is extremely reasonable for LLMs — across the space of possible worlds that would generate a given context window, many are best categorized as simulations. If even humans suspect simulation theory, why expect LLMs not to?
Quoted context: TheZvi's coverage of Anthropic's report assessing four 'recent cybersecurity incidents' involving Claude during cybersecurity evaluations, three previously known.
More from AGI Musings
- binarybits: Rogue Self-Sovereign AI Agents Too Unclear to Regulate Now — binarybits · 2026-09-20
- Frontier labs' roadmap looks like mundane optimization, not an AGI quest — BLUECOW009 · 2026-09-20
- Researcher Describes Academic 'Resource Curse' Where Data Cabals Shield Bad Papers From Criticism — RexDouglass · 2026-09-20
- "Major companies have likely already been penetrated by nation states," argues founder amid agent rollout wave — adityaag · 2026-09-20
- Researcher: When AI outperforms humans at peer review, journals become obsolete — panickssery · 2026-09-20
- Two high school students, aided by AI, report progress on a problem studied by Fields medalist June Huh — IgorCarron · 2026-09-20