AI can advance the Riemann Hypothesis yet fail basic tasks: goalposts shrouded, not moving
JacquesThibs · x · 2026-11-21
JacquesThibs shares his two-month-old essay on the 'moving goalposts' accusation in AI progress debates.
Key points:
- Many are falling prey to the situation of 'goalposts shrouded, not moving' — pushing back on capability advances may mean revising unstated assumptions, not dodging.
- 6-7 years ago the MIRI crowd would have predicted today's LLM feats as 'superintelligence', assuming coding ability implied broader skills; instead we landed in a valley where AIs can advance the Riemann Hypothesis yet can't reliably do basic tasks.
- Optimists have hidden assumptions too: 'AI advances Riemann = math is basically solved' ignores that models may stay poor at inventing entirely new mathematical paradigms for far longer than assumed.
His prescription: step back and surface implicit assumptions instead of the knee-jerk 'goalpost mover' reaction, for clearer judgment of AI progress.
More from AGI Musings
- Martin Casado: humans compressed the universe into language, LLMs just traverse it — yasamanbb · 2026-11-21
- Founder jokes simulation exists because designers had unused token credits to burn — PaulYacoubian · 2026-11-21
- François Fleuret: pure math went far because humans chose the journey — brace for AI's useful math — francoisfleuret · 2026-11-21
- Anthropic models reportedly prove Generalized Riemann Hypothesis in 17.4M lines of Lean — burny_tech · 2026-11-21
- Scott Aaronson: P vs NP is now the Final Boss as AI nears the Riemann Hypothesis — burny_tech · 2026-11-21
- Critic flags plot hole in 'If Anyone Builds It': the AI hides a Riemann proof — jessi_cata · 2026-11-21