Raw LLM Responses

Unfortunately, half of NLP researchers don't think even fine-tuned LLMs can ever understand NLP (https://arxiv.org/abs/2208.12852), so unless all those experts are wrong and someone does make a breakthrough I don't see how things can get drastically better. Hence why when you use AI to do anything you're immediately impressed because it broadly came up with a good answer, but once you start drilling down into specifics and increase the complexity of the ask it struggles. That's an NLP complexity problem not a context or memory problem. Realistically speaking, I think they're going to have to walk back the idea of LLMs having fully autonomy, and instead go down a modified supervised learning path where LLMs are fine-tuned based on manual corrections made by humans.