New TRACES benchmark evaluates 'discoverative AI' beyond final-answer accuracy
rohanpaul_ai · x · 2026-08-20
Rohan Paul highlighted the new TRACES benchmark introduced by Apodex, designed to measure 'discoverative AI.' Unlike traditional benchmarks that test retrieving known answers, TRACES evaluates systems that can investigate consequential problems using evidence, tools, and verification without an answer key. It frames AI discovery as an entire investigation, distinguishing high-scoring outcomes from the quality of the process, including error repair and grounded claims. Apodex published a definition of 'discoverative intelligence,' a rubric to distinguish sound investigation from luck, and an open call for solvers and problems.
Related event: Apodex Launches TRACES, First Benchmark for Discoverative AI(8 posts)→
More from Research
- Five pharma giants federally train OpenFold3 on 20,000 private structures, +11 pts — jiqizhixin · 2026-10-16
- SWAN: a semantic-level text watermark that survives paraphrasing — burkov · 2026-10-16
- Interp researchers push back: 'networks are already clean' claim rests on dropout artifact — aryaman2020 · 2026-10-16
- LayerTrim: training-free pruning keeps reasoning LLMs intact on AIME — SeoulNatlUniv · 2026-10-16
- Simular presents neuro-symbolic computer use agents at BayLearn — xwang_lk · 2026-10-16
- Google's Hartmut Neven disputes chip test, outlines plan to link a brain to a quantum computer — Chris_Armstrong · 2026-10-16