[retiredguy123]

Originally Posted by Bill14564

AI absolutely does make stuff up, that is precisely what AI is used for. Yes, it derives the information it provides from source documents but then it provides a narrative to answer the question that was posed. In creating (making up) that answer mistakes can happen. AI has been notoriously bad at simple math too though hopefully it will get better.

I guess we can agree to disagree on that point.

Sometimes, AI will summarize information compiled from various sources. In my opinion, that is not the same as "making stuff up".

Some posters on this website will make a false statement, and they cannot provide any source for the information. That is making stuff up.

[Velvet]

Originally Posted by Bill14564

AI absolutely does make stuff up, that is precisely what AI is used for. Yes, it derives the information it provides from source documents but then it provides a narrative to answer the question that was posed. In creating (making up) that answer mistakes can happen. AI has been notoriously bad at simple math too though hopefully it will get better.

You’re right that AI can make mistakes and can sometimes confidently give information that isn’t true — that’s usually called a hallucination. But “making stuff up” isn’t precisely what AI is used for. It generates responses by predicting language from patterns learned during training, and it doesn’t necessarily derive each answer from source documents. When it has access to search or other tools, it can check sources, which makes verification much easier. And yes, traditional language models have been surprisingly bad at arithmetic because they generate language rather than calculate — although newer systems can use calculators and other tools for that. So skepticism is healthy; I just wouldn’t describe fabrication as AI’s purpose.

[Bill14564]

Originally Posted by retiredguy123

I guess we can agree to disagree on that point.

Sometimes, AI will summarize information compiled from various sources. In my opinion, that is not the same as "making stuff up".

Some posters on this website will make a false statement, and they cannot provide any source for the information. That is making stuff up.

Originally Posted by Velvet

You’re right that AI can make mistakes and can sometimes confidently give information that isn’t true — that’s usually called a hallucination. But “making stuff up” isn’t precisely what AI is used for. It generates responses by predicting language from patterns learned during training, and it doesn’t necessarily derive each answer from source documents. When it has access to search or other tools, it can check sources, which makes verification much easier. And yes, traditional language models have been surprisingly bad at arithmetic because they generate language rather than calculate — although newer systems can use calculators and other tools for that. So skepticism is healthy; I just wouldn’t describe fabrication as AI’s purpose.

I don’t think we are that far apart. I wasn’t even thinking about hallucinations but that would truly be a case of AI making things up. A good example might be legal filings that cite cases that do not exist.

Yes, AI synthesizes more than it fabricates. But to me, synthesis that is guided to make good sentences rather than to make good summaries is only a little better than a random guess.

I appreciate the AI summaries that I see at the top of my “google” searches. I use them to get an idea of what I should expect to find and then I begin to look at the pages that were returned to get the real data. AI can be good for pointing in the right direction but the user still needs to dig out the real data or facts.