Apodex launches TRACES, a benchmark measuring whether AI can rigorously explore the unknown
eyishazyer · x · 2026-09-04
Apodex (founded by Tianqiao Chen) unveiled TRACES, a "Discoverative AI" benchmark that measures how far models can push the frontier of the unknown rather than answer known questions.
Six capabilities spell its name:
- Tools: selecting, calling and interpreting external tools
- Repair: locating and correcting its own errors from feedback
- Alternatives: weighing competing hypotheses as evidence accumulates
- Coherence: holding state and logic across long chains of work
- Evidence: grounding every conclusion in observation, data or citation
- Scope: stating where conclusions hold and where they don't
Current coverage spans AAV capsid, drug repurposing, clinical trials and LLM engineering, with a live benchmark tracking 1,182 trajectories across 17 environments and 11 solvers.
More from Research
- TCR specificity prediction: strong motif in one chain can still be nonbinding with wrong partner — iskander · 2026-09-04
- Purdue's AutoTraceGT automates grounded theory coding to analyze agent behavior at scale — Purdue · 2026-09-04
- Expert explainer on Anthropic's protein binder campaign: design cost cut from $10k to ~$100 — AllThingsApx · 2026-09-04
- Paradigm 3: low-quality RL environments may explain reward hacking; EBR-bench shows humans beat AIs — gleech · 2026-09-04
- Schmidhuber releases report claiming Hinton's Nobel-winning work was plagiarism — SchmidhuberAI · 2026-09-04
- SE-RRM paper hits ARC-AGI with just 2M params amid GPT-6 recurrent depth speculation — gklambauer · 2026-09-04