> you can train the network to act as if it knows its knowledge limits Humans need more training to do the same. And this does not answer the question of whet

> you can train the network to act as if it knows its knowledge limits Humans need more training to do the same. And this does not answer the question of whether there's self-reflection going on. Practical LLMs available today are perfectly capable of telling you about their own reasoning process. And much like a human, an LLM is incapable of fully explaining it's reasoning process from first principles to a human. Why is that? Probably because humans are too limited to grok large scale complex processes like the human brain or an LLM. Finally, there is no difference between acting as if you know your training limits and knowing your training limits, if your acting is good enough. This goes for both humans and LLMs.