Researchers question antibody model's SabDab training data extending to 2025, flagging test contamination
anshulkundaje · x · 2026-09-19
In a discussion about a recently released antibody structure prediction model, researcher proteinrosh notes it appears to have been trained on SabDab structures up to 2025. While the authors say they removed entries similar to the test set, this is described as a significant departure from field norms, raising concerns about evaluation rigor.
More from Research
- Async RL is just horizontal scaling with sticky routing, one dev argues — dosco · 2026-09-19
- Neuron: The recipe for intelligence in natural and artificial systems — AnnaCiaunica · 2026-09-19
- CoT may not be faithful: filler tokens add 13 points, models keep reasoning after committing — ziv_ravid · 2026-09-19
- TokenRhythm unveils NeoHorse-1: 4B/9B models claiming first closed-loop step toward RSI — jiqizhixin · 2026-09-19
- Training an LLM to paint by writing code: RL meets creative tasks with a judge model — measure_plan · 2026-09-19
- AI provers on a Bittensor subnet close six decades-old Erdős problems in ten days — markjeffrey · 2026-09-19