Researchers clash over whether AlphaGenome's coding-variant results are benchmark-gamed
anshulkundaje · x · 2026-09-19
- @ctokheim questions DeepMind's AlphaGenome: while non-coding variant interpretation is the headline feature, several aspects of its protein-coding variant results give the impression benchmarks may be gamed to inflate hype.
- The tweet pushes back on @anshulkundaje's original defense that AlphaGenome and its AVI score are not "AI slop taking over genomics," setting up a debate over its evaluation methodology.
More from Research
- OpenDataArena's Spark-234K, a 234K-sample machine-generated English text dataset, trends on Hugging Face — OpenDataArena · 2026-09-19
- match-seq: Barcoded RNA Transfer Deciphers Cell-Cell Interactions Beyond CRISPR Screens — anshulkundaje · 2026-09-19
- Robot Control Explained for ML People: PID, FK/IK, and Why ACT Outputs Joint Angles — kamathsblog · 2026-09-19
- Google's ScientistTwo agent runs the full AI research discovery loop unsupervised — dair_ai · 2026-09-19
- RSI-Exam leaderboard update: GPT-6-astra tops recursive self-improvement benchmark at 0.5126 — yuyinzhou_cs · 2026-09-19
- Caleb Gross publishes "Word Machines for Weird Machines" in Phrack 73 — dyn___ · 2026-09-19