Apodex uses forecasting as a testbed: teaching AI to question whether its reasoning holds up
SimonShaoleiDu · x · 2026-11-10
At the Chen Institute's AIAS+26 symposium in San Francisco, Apodex lead scientist for reasoning models Simon Shaolei Du gave a talk on moving AI from retrieving knowledge to genuine discovery.
Core argument: generating a prediction is easy; knowing whether the reasoning holds up is another matter. Models must evaluate evidence, challenge assumptions, and recognize when conclusions may be wrong. Forecasting is an ideal testbed — outcomes haven't happened yet, so models must weigh competing evidence under uncertainty and can later be checked against reality.
Key elements of Apodex's approach include evaluating source reliability rather than assuming the majority is right, and separating evidence quality from argument coherence.
Apodex frames its TRACES paradigm around "Discoverative AI" with three offerings: a reasoning model built for sustained reasoning, correction and uncertainty; TRACES, a public benchmark for how AI investigates complex questions; and Apodex Lab applying these to real research questions. Slides claim Apodex-1.0 is SOTA at Deep Research on agentic search and agentic scene benchmarks.
More from AGI Musings
- MIRI launches "Do Not Accept" campaign against AI firms' superintelligence race — wfithian · 2026-11-10
- Claude builds a 500,000-tile mosaic engine, sparking the 'tasteslop' debate — teortaxesTex · 2026-11-10
- Economists note LLM analysis lacks the incentive to exaggerate effects like public debate does — robseamans · 2026-11-10
- Wharton scholar asks: what are institutions for in an age of AGI? — danielrock · 2026-11-10
- Schmidhuber: Physical AI progress is overestimated, human-level physical intelligence not within 2 years — SakanaAILabs · 2026-11-10
- Panel: if consciousness is computational, what follows for each discipline? — burny_tech · 2026-11-10