6 ms·
“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
by doodpants 8mo ago
“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.
- simianwords 8mo agothis is incorrect. it has the concept of truth and facts.
- usefulcat 8mo agoHow 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 8mo agohow would you prove that a human has it?
- usefulcat 8mo agoWhataboutism 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 8mo agoits 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 8mo agoI don't know. What is your answer to my question?
- simianwords 8mo agoKnowing which word is likely to come after the other is trivially the concept of knowing truth for me. Why not? We have optimised for truth and we are predicting the best words that ensure this optimal value.
- imtringued 8mo agoHumans 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 8mo agowhat about a person with short term memory?
- idopmstuff 8mo agoI 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.