Improve Large Language Model Systems with User Logs
arXiv:2602.06470v2 Announce Type: replace
Abstract: Scaling training data and model parameters has long driven progress in large language models (LLMs), but this paradigm is increasingly constrained by the scarcity of high-quality data and diminishing…