Capability Ceilings in Autoregressive Language Models: Empirical Evidence from Knowledge-Intensive Tasks

Open in new window