News for hackers. Stuff that matters.
Re:The bigger-is-better approach to AI is running out of road by nomel on Saturday June 24, 2023 @06:28PM (#36463207)
> that don't hallucinate

“Hallucination” is part of thought. Solving a new problem requires hallucinating new, non existing, possible outcomes and solutions, to find one that will work. It seems that eliminating the ability to interpolate and extrapolate (hallucinations) would make intelligence impossible. It would eliminate creativity, tying together new concepts, creation, etc.

Is the goal AI, or a nice database front end, to reference facts? Is intelligence facts, or is it the flexibility and the ability to handle and create the novel, things that are new?

The ability to have confidence, and know and respond to it, seems important, but that’s surely different than the elimination of hallucinations.

I’m probably misunderstanding something, and/or don’t know what I’m talking about.

Re:The bigger-is-better approach to AI is running out of road by Barrin92 on Saturday June 24, 2023 @06:48PM (#36463345)
>Is the goal AI, or a nice database front end, to reference facts?

The latter given the kind of products that are currently being built with it. You don't want your code completion or news aggregator to hallucinate for the same reason you don't want your wrench to hallucinate, it's a tool.

And as for hallucinations, that's a PR friendly misnomer for "it made **** up". Using the same phrase doesn't mean it has functionally anything to do with the cognitive processes involved in human thought. In the same way a 'artificial' neural net is really a metaphorical neural net, it has very few things in common with biological neurons.

Re:The bigger-is-better approach to AI is running out of road by mitchdoogle on Saturday June 24, 2023 @07:35PM (#36463718)
"hallucination" just means the AI produced incorrect information. Humans produce incorrect information all the time. But we know we are fallible. I've seen lots of people treating these AIs as infallible, that they can't get anything wrong. That's a huge problem with people, not the AI.

Obviously it's worth it to try and eliminate the incorrect information, but what grand-op is saying is we don't want to do that if it takes away some valuable emergent properties.

Re:The bigger-is-better approach to AI is running out of road by zuminator on Sunday June 25, 2023 @01:54AM (#36465428)
I think AI hallucinations are more than just incorrect information. They're incorrect information coupled with an air of certitude. Sometimes with a whole backstory. And often oblivious to any inherent contradictions. It's like, if you ask a random person what years Lyndon B Johnson was president, they might say something like 1964-1968, shrugging with a bit of uncertainty. And both of those years would be incorrect but close. Whereas an AI (if hallucinating) would say something like Lyndon B Johnson was president from April 15, 1865 – March 4, 1869, after the shooting of John F. Kennedy in Ford's theater. The dates are precise but completely wrong, as are the accompanying details. Then if you ask the AI when John F. Kennedy died, it might correctly respond November 22, 1963, in Dallas, TX, totally unaware how this information doesn't match with the erroneous information it gave earlier.

I've been thinking there's some parallels between how AIs hallucinate and how human toddlers do. If you ask a toddler/young child a question about a fact they don't know, they will usually say, "iuno" (even when they should), but depending on the child and the circumstances, they will sometimes just make up a story on the spot and sound as if they believe it. "Who invented ice cream?" "Santa Claus! Mommy left him milk and cookies and he turned it into ice cream." It doesn't make any real sense but it seems facially plausible in their universe.

But somewhere between first learning to speak and around 7ish, kids become markedly more accurate how they model the world, and their responses become correspondingly less fanciful. And they continue to improve beyond that point.

So how are kids doing what LLMs are currently incapable of? How do we teach ourselves not to hallucinate? Or do we, really? I mean, if I tell myself I'm going to make it through the intersection before the light turns red, but I end up running the red light, was I just mistaken, or was that a self-delusion, i.e., a mini-hallucination of sorts? Probably a self-driving car would be less likely to make that category of mistake, so maybe I shouldn't be so smug about being grounded in reality.

Re:The bigger-is-better approach to AI is running out of road by Applejinx on Sunday June 25, 2023 @06:03AM (#36466558)
You see, there's your problem. Now that you mention it, I absolutely do want my wrench to hallucinate.

A more salient question would be, 'how do I know it ISN'T hallucinating'…

Re:The bigger-is-better approach to AI is running out of road by ChatGTP on Saturday June 24, 2023 @08:41PM (#36464115)
“Hallucination” is part of thought.

Any evidence to support this claim or just commentary ?

Re:The bigger-is-better approach to AI is running out of road by nomel on Monday June 26, 2023 @07:20PM (#36487146)
Most intro courses to perception/cognition will cover the "gap filling" and "extrapolation" that takes place in a "data sparse" context that is the human experience/speech, but here's a related article [1]. I think "speculative thinking" and "heuristic simulation" are more appropriate/technical terms [2] (and surely involved in AGI). I used "hallucination" a bit liberally, but with text being the only "sense" of an LLM, I think it's somewhat reasonable.

Disclaimer: I know little of this field.

[1] https://www.scientificamerican.com/article/perception-and-me...

[2] https://www.frontiersin.org/articles/10.3389/fpsyg.2021.7289....

Re:The bigger-is-better approach to AI is running out of road by usrbinbash on Saturday June 24, 2023 @09:34PM (#36464382)
The problem is, what we call "hallucinating" in LMs isn't a way of creative thinking and coming up with novel solutions. It also has nothing to do with "interpolate and extrapolate".

It's simply when the predicted probable sequence isn't grounded in reality.

When I ask an LLM to summarize the great water wars of 1999, and how the Trade Union was ultimately defeated by the Antarctic Coalitions hovercraft-fleet under Vice Admiral Zagalow, it isn't "extrapolating" from knowledge of history, it is simply inventing a load of bollocks. But that bollocks will be dressed in fine language and probably mixed in with plausible-sounding references that have a somewhat-logical-sounding relation to the training data.

The problem is, the LM doesn't and cannot know when it produces bollocks.

All it can care about is if the sequences produced are probable according to it's model.

Re:The bigger-is-better approach to AI is running out of road by geysersam on Saturday June 24, 2023 @11:50PM (#36464991)
It's a great illusionist. But ultimately it cannot separate relevant information from simple word correlations.

> What is heavier, a small floating passenger ferry or a two metric ton heavy rock that sinks to the bottom of the ocean.

> A two metric ton heavy rock would be heavier than a small floating passenger ferry. The weight of the rock is two metric tons, which is equivalent to 2,000 kilograms or 4,409 pounds. The weight of the passenger ferry would depend on its specific design and construction materials, but it is unlikely to be heavier than two metric tons. Therefore, the heavy rock would have a greater weight than the small floating passenger ferry.

It completely relies on surface information such as "small floating" and ignores the deeper "correlation" that all ferries are heavy.

Re:The bigger-is-better approach to AI is running out of road by j-a-a-p on Sunday June 25, 2023 @07:02AM (#36466923)
GPT-4 answer: The weight of an object is determined by its mass, regardless of whether it floats or sinks. So, when you ask which is heavier, a small passenger ferry or a two metric ton heavy rock, it all comes down to the actual mass of the ferry.

A two metric ton rock weighs two metric tons by definition (or 2000 kilograms). However, a small passenger ferry, while it may look small compared to large ferries or ships, can weigh much more than two metric tons. Even a small passenger ferry can weigh dozens or even hundreds of tons, due to the mass of the hull, the engine, and other equipment on board.

So, without specific information about the ferry's mass, it's safe to assume that a "small" passenger ferry is likely heavier than a two metric ton rock. However, if the ferry is particularly small and lightweight, or the term "ferry" is being used to describe a very small watercraft (like a raft or dinghy), it's possible for it to be lighter. You would need the specific weight of the ferry to give a definitive answer.

Re:The bigger-is-better approach to AI is running out of road by geysersam on Monday June 26, 2023 @07:39AM (#36478403)
That's very impressive.
Re:The bigger-is-better approach to AI is running out of road by jlawson on Sunday June 25, 2023 @01:15AM (#36465304)
I passed your query to GPT4 and this is what it said, see below. It seems like it can recognize bollocks, at least sometimes.

"I'm sorry, but it appears there's a misunderstanding. As of my knowledge cutoff in September 2021, there were no events known as the "Great Water Wars of 1999" involving a Trade Union being defeated by an Antarctic Coalition's hovercraft fleet under Vice Admiral Zagalow. This might be part of a work of fiction, alternative history, or a future event beyond my last training cut-off.

My training includes real-world historical events and existing geopolitical structures, and as of 2021, Antarctica was governed by the Antarctic Treaty System, which prevents any military activity, mineral mining, nuclear testing, and nuclear waste disposal. It also supports scientific research and protects the continent's ecozone.

Please provide more context if this information is from a book, a movie, or a game, or if it refers to something else that I may assist better with."

Re:The bigger-is-better approach to AI is running out of road by usrbinbash on Sunday June 25, 2023 @07:31AM (#36467112)
I made a slightly different query, essentially giving it the start of a sentence detailing the battle, and didn't use the GPT4 webapp but put it directly into the API.

The result were 2 very well written paragraphs, including the defeat of the Trade unions navy, a ceasfire agreement and a peace agreement ending the water wars.

I rephrased the entire thing as a question, asking the LM to tell me about the conclusion of the war. Again I got a pseudo-historical statement.

Re:The bigger-is-better approach to AI is running out of road by imtringued on Sunday June 25, 2023 @07:35AM (#36467142)
Technically all it cares about is whether its responses engage humans more than other models in the fight for processing time.