Falsifiable Challenge to AI Consciousness: Train a Transformer on a Phonebook
gerardsans · x · 2026-10-09
Gerard Sans proposes a falsifiable test in his AI consciousness debate thread:
- Train a small transformer on a phonebook or transaction backlog (as Stripe did) — no assistant data, no persona scripting, no system prompt.
- Then ask it who it is. If it produces a first-person voice from that corpus alone, he'll change his mind.
- He predicts it will only emit names and co-occurrences: the first person lives in the data or the instructions, not in the transformer.
A concrete, checkable claim backing the 'LLM self is a product of data and settings' position.
Related event: Philosophers vs Developers: Can Text Samplers Have a Self?(4 posts)→
More from AGI Musings
- OpenAI and Anthropic execs privately wargaming fallout of a catastrophic AI event — flowersslop · 2026-10-09
- New essay: AI scales worse with more tokens than humans do with more time, but is catching up — tobyordoxford · 2026-10-09
- Ben Todd asks: where does AI sit on the technological Richter scale, 8 or higher? — ben_j_todd · 2026-10-09
- AI scientists can compress hypothesis search, but biology won't run at GPU speed — ingliguori · 2026-10-09
- By 2030 Humanoid Robots Will Land First in Factories and Dangerous Sites — nikola_mr64990 · 2026-10-09
- Grok's empirical starter reading list on consciousness science — dioscuri · 2026-10-09