[doodpants]

“our AI has messed up completely.”

No, it worked as designed. Generative AI simply creates content of the type that you specify, but has no concept of truth or facts.

[idopmstuff]

I find takes like this very strange. Whether or not it gives the correct information, it's clearly not designed to give false information to factual queries.

The design of it is based on the intention of the people creating it, not the actual outcome, and it's pretty clear from all available information, plus a general understanding of incentives, that it's designed to be as accurate as possible, even if it does make errors.

[simianwords]

this is incorrect. it has the concept of truth and facts.

[usefulcat]

How is knowing what word is most likely to come next in a series of words remotely the same as having "the concept of truth and facts"?

[simianwords]

how would you prove that a human has it?

[imtringued]

Humans update their model of the world as they receive new information.

LLMs have static weights, therefore they cannot not have a concept of truth. If the world changes, they insist on the information that was in their training data. There is nothing that forces an LLM to follow reality.

[simianwords]

what about a person with short term memory?

[usefulcat]

Whataboutism is almost never a compelling argument, and this case is no exception.

ETA:

To elaborate a bit: based on your response, it seems like you don't think my question is a valid one.

If you don't think it's a valid question, I'm curious to know why not.

If you do think it's a valid question, I'm curious to know your answer.

[simianwords]

its not whataboutism, i'm simply asking how you would perform the same test for a human. then we can see if it applies or not to chatgpt?

[usefulcat]

I don't know. What is your answer to my question?