The Competence Gap: Language Models Perform Intelligence. The Harness Does the Work.
A five-minute terminal lab that shows a model has no memory of its own, then what RLHF, benchmarks, and the race narrative look like once you credit the harness instead of the model.