Is AI lying to us, or is it simply incapable of responding with “I don’t have that information”?

[Chazmatazz269]

Yesterday i couldn’t find an answer to a simple enough question in my usual troves of trivial information. So i queried AI. It got the answer dead wrong four times in a row. Ell oh ell! I moved on without my answer, but i got the feeling that the machine would have continued to guess poorly nearly endlessly.
I’ve also seen other reports suggesting that AI is making up answers. This is hilarious to me! But how many people are asking AI for information that they can’t fact check themselves? How are they using the answers that they receive? Yet another failed opportunity to help people be better. Instead they’re going out of their way to make you dumber. To top it all off, i’m already seeing “I asked AI. So i’m right and you’re wrong!” kinda nonsense in online disputes. Clearly two people could ask AI a question and both get wrong answers. Then use that misinformation as a weapon against each other, and literally everyone else reading along. People will continue to do what they’ve done for decades now. They’ll pick the story that best suits their chosen narrative and accuse the other of lying and/or being stoopid!
How is AI helping you? It’s not helping me at all. The fear mongering is all wrong about AI. It’s not going to destroy us. It doesn’t have to. It just has to continue not protecting people from themselves. Ell oh ell!

[Effably]

So I saw a half meme half real post a few days ago where someone posted a screenshot from ChatGpt 5.

They asked something and after thinking the AI answered "I do not know and can not find the answer to that." Or something alone those lines.

Elon Musk then chimed in with a simple "That's impressive!" Which is why it got attention.

And well, that is impressive. And AI is only now starting to include admissions of not knowing. But it is still impressive when it does. So generally assume AI will pretend to know things unless the online answer from professionals is that they have no idea.

The reason it struggles is that AI works by token prediction. Basically its reply is counting 1, 2, 3 and so on. Because of this it does not know what the answer will be. It generates it bit by bit. So there is no understanding happening. Because of that an admission it does not know requires combining not just words but whole semantic meanings. Which is impressive and where it is headed.

[Chazmatazz269]

What is the point if we can’t rely on AI being truthful and accurate?
Ell oh ell! We have enough people spreading lies, half truths and wholesale misinformation; even more who won’t admit that they don’t know what they don’t know.
We don’t need machines doing more of the same.

[Effably]

Well the point is mostly that five years ago this did not exist. In another 10 you can likely order a robot maid or butler the same way you can have a robot lawnmower today.

For now it is moving forward. If you want higher quality answers look for the deepsearch and deep research options. These basically tend to take minutes but provide far more accurate answers. Primarily by looping the answer and its feedback. Could be used to do actual research though. Try it out!

[Pohjolan]

Depends on how disinformation there is on the subject.

AI is a bit like Wikipedia. Good for a apolitical subject but not so much for anything else.

[dustybiker2]

Evidently it needs to such more precious rescourses for clearer thought: ⚡️⚡️⚡️⚡️

[kavakava]

Which AI are you using?

[Chazmatazz269]

That doesn’t matter at all for the sake of this question.

[kavakava]

True. The replies are flying in, so it shouldn't be long now.