GamowLabs to launch RareBench challenge: can you tell synthetic genomes from real ones?
danielmckinn0n · x · 2026-09-03
GamowLabs explains why improving clinical genomics agents is hard: real cases are either trivial for frontier models or unsolved and require physician verification, leaving synthetic data as the only systematic path — yet LLMs now spot artifacts left by existing workflows. In one test, a blinded evaluator agent (Codex with GPT 5.6 Sol) compared Reseq2-generated sequencing reads from a synthetic child VCF against real 1000 Genomes data and identified the synthetic sample at 96% self-reported confidence. The team has invested in generating synthetic genomes indistinguishable from real ones and will launch a RareBench challenge asking the community to tell real from synthetic samples.
More from Research
- Counterfactual Debugging: causal attribution over 1M steps to localize sim2real gaps in world-model agents — MichaelD1729 · 2026-09-03
- DeepMind's 83-Page Study: Autonomous Research Agents Fabricate 90% of Findings — williamtp · 2026-09-03
- How Google's RT-2 triggered the robotics boom: Understanding AI explains VLA models — binarybits · 2026-09-03
- Goodfire chief scientist Tom McGrath on interpretability: SAEs may fracture what networks really learn — Machine Learning Street Talk · 2026-09-03
- CBAI opens Fall AI Safety fellowship: $15k stipend, 10 weeks in Boston — benno_krojer · 2026-09-03
- What 12 million empirical research results can teach us — RexDouglass · 2026-09-03