Comment by ottah

The models are not smarter than us by far. Have you not run into issues with reasoning and comprehension with them? They get confused, they miss big details, build complicated code thats ineffective. They don't work well at tasks that require a larger holistic understanding of the problem. The models are weak, brittle reasoners, because they have an indirect and contradictory understanding of the wold. We're several breakthroughs away and several hardware generations from having models that are robust reasoners for grounded, non-kind problems.