Independent researcher's phase-transition AI acceleration model hits all 6 pre-registered prediction windows
sadeyeprophet · reddit · 2026-10-09
An unfunded independent researcher published a phase-transition model of AI acceleration, unifying three separately-derived equations, and claims all six pre-registered June–August 2026 prediction windows hit his rubric. He candidly flags confounds: same-day registration for N+1, post-hoc baseline, and windows clustered inside one policy period, inflating the raw p-value (9×10⁻⁸). Separately cites METR data showing frontier task-horizon doubling every 128.7 days (R²=0.934). Next window: Oct 8–22; three consecutive misses kill the model. AMA on Reddit.
More from AGI Musings
- Deciding what counts as cruel injects power's biases: a critique of AI alignment adjudication — BecauseCulture · 2026-10-09
- "No reason to think models are conscious": anthropomorphizing them is dangerous — RexDouglass · 2026-10-09
- AI safety researcher argues shutting down OpenAI would be consistent with its own mission — DavidSKrueger · 2026-10-09
- How long until you can raise a seed round just by talking to an agent? — seanmcdonaldxyz · 2026-10-09
- Agent societies in empty 3D worlds: DeepSeek writes a charter, Gemini builds a bridge — pkmital · 2026-10-09
- Shreyas: Execution-focused PMs finding it much harder to get hired at top companies — annetgriffin · 2026-10-09