[gdcbe]

Why do people talk about hallucinations? Pretty deceptive word of you ask me.

Not an expert though, but isn’t that behaviour inherent to how it works? Bit of a misnomer and giving people the wrong idea of what is going on here.

[mariopt]

Hallucinations reduce the success rate of AI workflows, which must be taken seriously. Imagine a workflow with 8 steps where each step/agent has a 95% success rate, the success rate of this workflow is only (1-0.05)^8 = 0.66 ~= 66%. Not bad but not enough to replace humans yet (unless 66% makes you profitable).

The hallucinations/errors compound and can misguide decisions if you rely too much on AI.

[realusername]

Not enough to replace humans in most critical tasks, but enough to replace Google, that's for sure. My own success rate to find information on Google these days is around 50% by query at best.

[jldugger]

Because "fabrication" seems worse, if more accurate.

[firejake308]

I prefer "confabulation," which describes the analogous human behavior where you have no idea what the objective truth actually is, so you just make up something that sounds right

[ein0p]

Fabrication implies malicious intent or at least intentional deception. LLMs don’t have any “intent”.

[jedbrown]

Their developers have intent. That intent is to give the perception of understanding/facts/logic without designing representations of such a thing, and with full knowledge that as a result, it will be routinely wrong in ways that would convey malicious intent if a human did it. I would say they are trained to deceive because if being correct was important, the developers would have taken an entirely different approach.

[wumbo]

generating information without regard to the truth is bullshitting, not necessarily malicious intent.

for example, this is bullshit because it’s words with no real thought behind it: “if being correct was important, the developers would have taken an entirely different approach”

[jedbrown]

If you are asking a professional high-stakes questions about their expertise in a work context and they are just bullshitting you, it's fair to impugn their motives. Similarly if someone is using their considerable talent to place bullshit artists in positions of liability-free high-stakes decisions.

Your second comment is more flippant than mine, as even AI boosters like Chollet and LeCun have come around to LLMs being tangential to delivering on their dreams, and that's before engaging with formal methods, V&V, and other approaches used in systems that actually value reliability.

[nicce]

In reality, it should sound worse so people don’t trust it so much.

But those who sell AI products don’t want that.

[kristiandupont]

It's obviously an analogy, but it seems pretty fitting to me? What would you call it?

[TillE]

Making errors, generating nonsense, being wrong. It's a catchy term but it's not accurate in any meaningful way.

[seadan83]

Hallucinating has the implication of being wrong. The word further adds the context of being elaborately wrong. That feels pretty accurate to describe an AI going into detail when it is wrong.

[phaedryx]

What would you call it when AI doesn't have the answer so it makes stuff up (sometimes in a dangerous way)?

[SAI_Peregrinus]

Confabulating if you want a non-"vulgar" word. Bulshitting if you don't care.

[jamil7]

Bullshitting

[elevaet]

That's called bullshit.

[wumbo]

Isn’t hallucinating inherent to biological brains too?

It’s normal in small degrees even for mentally healthy individuals.

[__loam]

Stop anthropomorphizing the token generator please.

[wumbo]

lmao, not what i said. fix your attention head.

metaphors grounded in reality are fine. otherwise don’t call it a “generator”

[x3n0ph3n3]

That's the accepted word to describe it making up bullshit instead of regurgitating existing information.