@meeper I know. They would have to move on from token prediction to an architecture closer to actual intelligence. And that's why that would signify true progress. As long as it's token prediction, nothing will change in a significant way.
@meeper I will only acknowledge there's meaningful progress if they solved one of these things: - No more hallucinations - Actual separation of data and instructions (i.e. no more "prompt injections")
As long as these exists, it's just turning some knobs.
Volpeon