Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

"Hallucinating" AIs sound creative, but let's not celebrate being wrong

thereader.mitpress.mit.edu

Post reply on HN
Print view View original

191–196 of 196 posts

Re: "Hallucinating" AIs sound creative, but let's not celebrate being wrong

#191

Post by nuancebydefault »

pdonis wrote: post #188
nuancebydefault wrote: post #184

Earlier quoted context omitted.

> Human beings can act on the world as well as perceive it Conceptually this is the same for LLMs. Input/perceive -> human/LLM -> output/act. You might say the acting part for LLMs is not autonomous but that is only dependent on how much agency is given to LLMs. LLMs outputs are massively used, hence their outputs lead to actions. > and humans have rich internal models of the world that get updated continously as a r…

> Conceptually this is the same for LLMs. No, it isn't. LLMs don't look at the results of their actions and update a world model. Humans do.

This is indeed a difference that I overlooked, there is limited feedback on LLMs output, other than a rater primitive survival model (models that aren't popular/correct/cheap/fast enough are not used anymore over time)

Re: "Hallucinating" AIs sound creative, but let's not celebrate being wrong

#192

Post by gaganyaan »

pdonis wrote: post #187
gaganyaan wrote: post #164

Earlier quoted context omitted.

> LLMs do not Prove it

I don't have to; the people who designed the LLMs have already done so by explicitly describing how they work. See, for example, my contrast of what an LLM does with what Wolfram Alpha does upthread.

Wolfram Alpha has a model of the world manually created by humans. LLMs generate a model of the world from training data. "Explicitly describing" how something works doesn't mean you wave your hands and say "therefore meat is magic".

I have no idea why you think Wolfram Alpha is relevant. LLMs can use tools like langchain to do the exact same thing Wolfram Alpha does. What proof do you have for your original claim?

Re: "Hallucinating" AIs sound creative, but let's not celebrate being wrong

#193

Post by Grimblewald »

Zigurd wrote: post #3

"Hallucinating" is really a euphemism for fluent bullshitting. It also doesn't fit the normal understanding of a hallucination. It's just wrong information expressed confidently. Partly it's how the conversational UI works, and partly how the model works. Without the nice generated text that creates in our minds an impression of competence, it would just be an error.

bullshitting implies internal knowledge of the deceit, if this were the case we could just train models to not bullshit and tell only the truth. A hallucination, when things are hallucinations, is when the model 'believes' what it is predicting to be equivalent to truth.

Re: "Hallucinating" AIs sound creative, but let's not celebrate being wrong

#194

Post by Grimblewald »

navjack27 wrote: post #78

LLMs aren't knowledge retrieval systems. Nor are they capable of creativity.

sure they are, they're a searchable archive made using a lossy compression algorithm, which is capable of synthesizing creative output from known elements.

Re: "Hallucinating" AIs sound creative, but let's not celebrate being wrong

#195

Post by timacles »

OrsonSmelles wrote: post #185
timacles wrote: post #181

Earlier quoted context omitted.

All you say is true but that is still not the point Ai does not and cannot create art, it’s as simple as that.

Just to be clear, even when a human writes and refines prompts, tweaks parameters, and iterates generation until they see something they imagined but lacked the traditional art skills to produce... we have not ended up with art? (If someone posts all 2000 iterates of their last prompt, that's bad social behavior, but it's hard for me to feel like it "un-arts" the results.) Or is there some other generation procedure…

With all due respect, I don’t think you really get what art is. Please just google the definition or something. No someone that doesn’t have art skills cannot create art precisely for the same reason, with or without ai .

The hypothetical situation doesn’t even make sense. For the exact same reason that no matter how many prompts you use, you can’t write a good book with chat gpt. Art “goodness” exists entirely beyond the realm of being able to communicate it through words. It’s not something you can quantify logically and refine it through a prompt. You’ll always end up with some mechanical, derivative crap

Re: "Hallucinating" AIs sound creative, but let's not celebrate being wrong

#196

Post by IIAOPSW »

invig wrote: post #144
IIAOPSW wrote: post #135

Earlier quoted context omitted.

In other words, LLM's are, in the technical sense defined by Harry Frankfurt, bullshitters. The essence of bullshit (as he defines it) is that it is neither truth nor lie but rather unconcerned with veracity altogether. A bullshitter wants to appear smart, or convince others to agree with them, or some other conversational objective. Bullshit statements may or may not ultimately turn out to be true, but the thing tha…

Isn’t the same thing true of any human who isn’t doing scientific research? Like how much work to get to veracity is actually going on out there?

“One of the most salient features of our culture is that there is so much bullshit.”

Literally the cold opening of "On Bullshit".

Post reply on HN