🖼️00Why machine learning research agents don't overfit — and what compression has to do with it - Amazon ScienceHacker News·22 days ago#XKccyBxA#amazon#overfitting#generalization#compression#validation#article+2 more🧰Tag tools✨Add tagNew research indicates that AI agents learn compressible models of data, which don’t have enough space to enable memorization.15s0Read later0Read More
🖼️4242Dispersion loss (LM-Dispersion)Hacker News·3 months ago#lpEEXfVK#chenliu-1996#generalization#photo#article#englishlanguage🧰Tag tools✨Add tagDispersion loss counteracts embedding condensation and improves generalization in small language models (ICML 2026).15s0Read later0Read More