Hallucination is a problem we'll have to live with for a long time
1–10 of 12 posts
Re: Hallucination is a problem we'll have to live with for a long time
#2I know I'm a bit of a broken record here, but "sometimes it hallucinates instead of being factual" is a bit like "sometimes the Ouija board fails to reach the afterlife spirits, instead of channeling them."
Both falsely imply that there's a solvable mechanical difference going on between results people like versus results people dislike.
Re: Hallucination is a problem we'll have to live with for a long time
#3I know I'm a bit of a broken record here, but "sometimes it hallucinates instead of being factual" is a bit like "sometimes the Ouija board fails to reach the afterlife spirits, instead of channeling them." Both falsely imply that there's a solvable mechanical difference going on between results people like versus results people dislike.
I hate neologisms like hallucinate. Terms of art are fine for cognoscenti, but are misunderstood by naive readers as implying aspects of intelligence. It's just statistics ma'am.
Re: Hallucination is a problem we'll have to live with for a long time
#4I know I'm a bit of a broken record here, but "sometimes it hallucinates instead of being factual" is a bit like "sometimes the Ouija board fails to reach the afterlife spirits, instead of channeling them." Both falsely imply that there's a solvable mechanical difference going on between results people like versus results people dislike.
I hate neologisms like hallucinate. Terms of art are fine for cognoscenti, but are misunderstood by naive readers as implying aspects of intelligence. It's just statistics ma'am.
Perhaps true. But if you are warning against the dangers of so-called "AI" - "hallucinate" is also suggestive of "dangerous", "insane", and some other strong negatives.
Re: Hallucination is a problem we'll have to live with for a long time
#5Earlier quoted context omitted.
I hate neologisms like hallucinate. Terms of art are fine for cognoscenti, but are misunderstood by naive readers as implying aspects of intelligence. It's just statistics ma'am.
Perhaps true. But if you are warning against the dangers of so-called "AI" - "hallucinate" is also suggestive of "dangerous", "insane", and some other strong negatives.
That's reaching for a positive in the negative to me, but undeniable. So .. I dunno you win!
I continue to respect Hinton, he uses and chooses his words carefully I think.
Re: Hallucination is a problem we'll have to live with for a long time
#6People... employees... friends... lovers... "hallucinate" too.
What rational agent is infallible?
Re: Hallucination is a problem we'll have to live with for a long time
#7Earlier quoted context omitted.
Perhaps true. But if you are warning against the dangers of so-called "AI" - "hallucinate" is also suggestive of "dangerous", "insane", and some other strong negatives.
That's reaching for a positive in the negative to me, but undeniable. So .. I dunno you win! I continue to respect Hinton, he uses and chooses his words carefully I think.
“It’s just completely obvious that within five years deep learning is going to do better than radiologists.… It might be 10 years, but we’ve got plenty of radiologists already.” - Geoffrey Hinton 2016 [1]
[1] https://newrepublic.com/article/187203/ai-radiology-geoffrey...
Re: Hallucination is a problem we'll have to live with for a long time
#8People... employees... friends... lovers... "hallucinate" too. What rational agent is infallible?
No one is claiming that any agent is or should be considered infallible, but for every other form of technology humans create - including software - there is a minimal standard of predictability and efficiency that is considered acceptable for that software to be useful.
For some reason, LLMs are the exception. It doesn't matter how much they hallucinate, confabulate, what have you, someone will always, almost reflexively, dismiss any criticism as irrelevant because "humans do the same thing." Even though human beings who hallucinate as often as LLMs do would be committed to asylums.
In general terms, the more mission critical a technology is, the more reliable it needs to be. Given that we appear to intend to integrate LLMs into every aspect of human society as aggressively as possible, I don't believe it's unreasonable to expect it to be more reliable than a sociopathic dementia patient with Munchausen's syndrome.
But that's just me. I don't look forward to the future when my prescriptions are written by software agents that tend to make up illnesses and symptoms and filled out by software agents that cant' do basic math, and it's all considered ok because the premise that humans would always be as bad or worse and shouldn't be trusted with even basic autonomy has become so normalized we just accept the abuse of the unstable technologies rolled out to deprecate us from society as inevitable. Apparently that just makes me a Luddite. IDK.
Re: Hallucination is a problem we'll have to live with for a long time
#9I agree, lots of the body of human knowledge is wrapped up in natural reactions, moving visuals, textures, smells, and sounds.
The current batch is trained on just text afaik.
Re: Hallucination is a problem we'll have to live with for a long time
#10I know I'm a bit of a broken record here, but "sometimes it hallucinates instead of being factual" is a bit like "sometimes the Ouija board fails to reach the afterlife spirits, instead of channeling them." Both falsely imply that there's a solvable mechanical difference going on between results people like versus results people dislike.
I hate neologisms like hallucinate. Terms of art are fine for cognoscenti, but are misunderstood by naive readers as implying aspects of intelligence. It's just statistics ma'am.
I find it's a perfectly fine word to describe the result. Humans do the same as our visual system samples a low amount of data points to construct the view we see. In effect we're always hallucinating with the difference being that we maintain high context to filter it to the correct hallucinations. This shows up in dreams where we don't maintain such context, or when context is largely changed. When I got back from a vacation where there were many small and larger lizards, I swear I saw one upon returning, but it turned out to be a similarly colored leaf moving in a scurrying motion.
Edit: "mis-remembering" is another good term as it's using its vast training to output tokens and sometimes it maps them wrong.