AI RESEARCH

When Less is More: The LLM Scaling Paradox in Context Compression

arXiv CS.LG

ArXi:2602.09789v3 Announce Type: replace Scaling up model parameters has long been a prevalent