Elie Bakouch: current AI research benchmarks overfit and can't track RSI
eliebakouch · x · 2026-11-19
Researcher Elie Bakouch weighed in on debates about AI research benchmarks, arguing they do measure "something" but fail to capture what AI research actually cares about — track 1 likely overfits heavily on both performance improvements and hyperparameters. His core point: these benchmarks are nowhere near the hardest tier, and if the community wants to genuinely track RSI (recursive self-improvement) progress, far harder benchmarks are needed.
More from Research
- Designing a humanoid robot foot: heel strikes hit 1511N, whole foot weighs 117g — dee_hw · 2026-11-19
- Google's GUARD paper: frozen base model, self-editing agent harness gains +8.89 points — omarsar0 · 2026-11-19
- Kezek splits mono two-person conversations into per-speaker tracks for full-duplex dialogue training data — manifestro · 2026-11-19
- Hutter Prize text compression jumps ~10% by overfitting small transformers for arithmetic coding — ricklamers · 2026-11-19
- HybridApexW optimizer: 2x faster than torch.compile AdamW+Muon baseline — omendUUde · 2026-11-19
- Sandia Labs' VASt model predicts high-entropy alloy properties from tens of samples — CurieuxExplorer · 2026-11-19