Hallucinations And Other Things

2 views

Skip to first unread message

Buddy Brannan

unread,

Jul 10, 2026, 10:55:26 AM (6 days ago) Jul 10

to echovisio...@agiga.ai

Hi y’all,

I’ve had something on my mind for a couple days, and I’d like to bring it up for discussion, if y’all will indulge a mere tech support monkey. That’s the concept of “hallucinations.” Not the tripping on acid kind, but the AI kind.

I think that the term “hallucination” is getting tossed about more broadly than it probably should be. Not every error is a hallucination.

A hallucination in AI, very like the acid tripping kind, is something the AI just makes up. That’s a technical term. It’s something that isn’t there at all. For example, if you take a picture of your living room and the AI says that there’s a purple rhinoceros standing in the back corner, that would be a hallucination. If the AI tells you there’s a shadowy figure of a skeletally thin man in a black robe standing behind you looking over your shoulder, that would be a hallucination. If you take a picture of a Marie Callender pumpkin pie and you’re Sharon and believe the instructions the AI gives you to set your oven to the hell setting, (if you don’t know this reference, I’ll fill you in later), that’s also a hallucination.

OK, so what isn’t a hallucination? Mis-identifying an object isn’t necessarily a hallucination. COntrary to what sighted people would have us believe, and other blind people too, sight is not infallible, and camera sight is even less infallible. One thing may well look like a different thing. Mixing up left and right is generally nothing more than a matter of perspective, where the AI is trying to interpret a picture and gets it wrong from the perspective of a person looking at it straight on. Well, the number of humans who also do this is pretty staggering, is it not?

Yeah, all right. There’s some nuance here. Putting an element into a scene that simply isn’t there and can’t be mistaken for something else is a hallucination. Extrapolating text from a package and filling it in with a similar image that’s 100% wrong? Definitely a hallucination. Misreading a 1 as a capital I? Not a hallucination, merely an OCR error.

I guess my point is, where technology is concerned, even more so than where humans are concerned, you still have to use some discernment. Never take what the AI tells you as gospel. Trust but verify. And if something sounds wrong, question it.

Very like GPS when it first started rolling out to the masses, people will believe that the tech will solve our problems. With GPS, we were promised, in some quarters anyway, that it would solve our travel and orientation problems. Never mind that GPS is at best 30 feet inaccurate, potentially more, by design. If you’re blind and looking for the front door, 30 feet may as well be a mile. In the same way, AI vision can only take us so far. And in both cases, it is important, even imperative, that we recognize the limitations as much as we recognize the potential and the capabilities. I really can’t stress this enough, especially as a longtime technology enthusiast. Especially as I make my living in technology. I’m going to be the first guy to tell you that technology is great, except when it isn’t.

Buddy Brannan

Customer And Technical Support Monkey

Agiga

Email: bud...@agiga.ai

Jenine Stanley

unread,

Jul 10, 2026, 11:40:09 AM (6 days ago) Jul 10

to echovisio...@agiga.ai

Buddy gets a banana for this post.

It’s also why at the present time, using AI for navigation is, or should be, limited, imho. If you want to know if there’s anything 10 to 20 feet in front of you, great. If you want AI to warn you about that tree limb you’re about to hit from 2 feet away, not so much.

Recently I reviewed some AI generated descriptions of photos and found that the model I used often transposed them. I knew that when looking at a certain scene, the chair was on the left. The AI description put the chair on the right. One photo was described as being upside down. The camera was in the exact same position as the photo before it.

--
You're receiving this message because you're subscribed to the EchoVision Community Forum.

To view the archive or catch up on discussions:
https://groups.google.com/a/agiga.ai/g/echovision-discuss

To unsubscribe, send an email to:
echovision-disc...@agiga.ai
To unsubscribe from this group and stop receiving emails from it, send an email to echovision-disc...@agiga.ai.