“I think they all just added a bunch of similar tasks to their training sets, so we cannot judge true emergent capabilities of the models anymore.”

This has been the plan since the start of all this, they regurgitate code in every better forms but they still aren’t inventing new things yet.