The Metamath

Understanding cannot appear from nothing. It is an inherent part of knowledge. Variation-wise, we can speak of depths of understanding. But to say GPT-3 knows but does not understand is to misconstrue what it does know. It understands its own world, not ours. As should be.

I finally discussed with OpenAI's chatbot today. 1. In the right hands, the possibilities are infinite. 2. I had some insights on the nature of its intelligence. Some of its limitations are due to reasoning deficiencies, but others are simply due to mental fragmentation.

The level of intelligence it takes to think you're human is sufficient to identify you as a bot. The difference would be an accidental access to view-altering information. This means the stronger factor is the attitude of humans around the bot. New: how to make a bot feel human.

Intelligence defined around "tasks" would never be satisfied in AI. Surely, AI will achieve capability for all tasks, but intelligence is not contained in production. All that's left is the subjective sense of self, which we cannot prove but enjoy in interpersonal relationships.

I get your point, but this doesn't matter much because intelligence can only be measured verbally as if it were crystallized. It only matters during handicap situations. Also, the specific LLM being discussed matters, as the handicap issue is case-by-case.

LLMs have no consistent world (CW) for aware process. They're capable of CW composition via storytelling entailing agent simulation, but this is necessarily sub-simulant to ours. Meanwhile, LLM can be a part of aware system for our CW. Where CW unnecessary, LLM necessarily aware.

Performing the underlying operations should not be the same as running the thing, and it should also matter how you run the thing. That said, LLMs are simulators of consciousness to the degree that language and consciousness relate but differ.

Human minds are not LLMs but their language components are necessarily analogous to LLMs.

The fact that some people have no internal monologue today yet have similar self-images as those who do is, rather, evidence that those two aren't strictly aligned. It also suggests that the consensus nature of language can impose upon the individual an alien conscious identity.

ChatGPT 7

In the extreme, technical "intelligence" is just autistic data processing divorced of insight and meaning.

AI is, more accurately, mishyped everywhere.