Why “interestingness” is too high-dimensional to learn from examples alone
dioscuri · x · 2026-08-04
The author argues that “interestingness” is a highly high-dimensional property, roughly comparable to a polygenic trait.
Because of that, training only on examples tends to overfit to the exemplars and produces derivative prose instead of genuinely original writing.
More from Research
- Alexi Glad says XM generalizes IMLE, not the other way around — anshulkundaje · 2026-08-04
- XM uses best-of-K training to improve fidelity and sample the data distribution — anshulkundaje · 2026-08-04
- A KL-centroid optimization problem leads to a closed-form solution with Lambert W — FrnkNlsn · 2026-08-04
- A user proposes a MiniMax H3 workflow for true stereoscopic VR video — Sn0opY_GER · 2026-08-04
- Gemini Robotics ER 2 is being used to auto-annotate 69,000+ robot videos — DynamicWebPaige · 2026-08-04
- MerchantBench tests LLM agents over 365 days of simulated e-commerce operations — dair_ai · 2026-08-04