It's kind of funny how AI has the exact same problems some humans have.
I always thought AI wouldn't have that kind of problems, because they would be carefully fed accurate information.
Instead they are taught from things like Facebook and the thing formerly known as Twitter.
What an idiotic timeline we are in. LOL
this post was submitted on 12 Jun 2024
1 points (100.0% liked)
Technology
59672 readers
2908 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related content.
- Be excellent to each another!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, to ask if your bot can be added please contact us.
- Check for duplicates before posting, duplicates may be removed
Approved Bots
founded 1 year ago
MODERATORS
I thought the main issue was that AI don't really know how to say I don't know or second guess themselves, as it would take a lot more robust architecture with multiple feedback loops. Like a brain.
Anyway, LLM's aren't the only AI that do this. So them being trained on Facebook data certainly isn't the whole issue.
Yeah it's the old garbage in, garbage out problem, the AI algorithms don't really understand what they are outputting.
I think at this point voice recognition and text generation AI would be more useful as something like a phone assistant. You could tell it complex things like "Mute my phone for the next 2 hours" or "Notify me if I receive an email from John Smith." Those sort of things could be easily done by AI algorithms that A) Understand your voice and B) Are programmed to know all the features of the OS. Hopefully with a known dataset like a phone OS there shouldn't be hallucination problems, the AI could just act as an OS concierge.
The narrow purpose models seem to be the most successful, so this would support the idea that a general AI isn't going to happen from LLMs alone. It's interesting that hallucinations are seen as a problem yet are probably part of why LLMs can be creative (much like humans). We shouldn't want to stop them, but just control when they happen and be aware of when the AI is off the tracks. A group of different models working together and checking each other might work (and probably has already been tried, it's hard to keep up).
load more comments
(1 replies)
Instead they are taught from things like Facebook and the thing formerly known as Twitter.
Imagine they would teach in our schools to inform yourself about all the important things, and therefore you should read as many toilet walls as newspapers...
There's also the fact that they can't tell reality apart from fiction in general, because they don't understand anything in the first place.
LLMs have no way of differentiating a fantasy RPG elements from IRL things. So they can lose the plot on what is being discussed suddenly, and for seemingly no reason.
LLMs don't just "learn" facts from their training data. They learn how to pretend to be thinking, they can mimic but not really comprehend. If there were facts in the training data, it can regurgitate them, but it doesn't actually know which facts apply to which subjects, or when to not make some up.
They learn how to pretend
True, and they are so darn good at it, that it can be somewhat confusing at times.
But the current AIs are not the ones we read about in SciFi.
I'd argue that referring to it as "AI" is a stretch since it's all A and no I.
load more comments
(1 replies)
The problem with AI hallucinations is not that the AI was fed inaccurate information, it's that it's coming up with information that it wasn't fed in the first place.
As you say, this is a problem that humans have. But I'm not terribly surprised these AIs have it because they're being built in mimicry of how aspects of the human mind works. And in some cases it's desirable behaviour, for example when you're using an AI as a creative assistant. You want it to come up with new stuff in those situations.
It's just something you need to keep in mind when coming up with applications.
Not in the case of the google search AI. It quotes directly from unreliable sources.
load more comments
(1 replies)
What weirds me out is that the things it has issues with when generating images/video are basically a list of things lucid dreamers check on to see if they're awake or dreaming.
-
Hands. Are your hands... Hands? Do they make sense?
-
Written language. Does it look like normal written language?
(3. Turn the lights off/4. Pinch your nose and breath through it) - these two not so much
-
How did I get here? Where was I before this? Does the transition make sense?
-
Mirrors. Are they accurate?
-
Displays on digital devices. Do they look normal?
-
Clocks. Digital and analog... Do they look like they're telling time? Even if they do, look away and check again.
(9. Physics, try to do something physically impossible, like poking your finger through your palm. 10. Do you recognize people/do they recognize you) - on two more that aren't relevant.
But still... It's kinda remarkable.
Also, Nvidia launched their earth 2 earth simulator recently. So, simulation theory confirmed, I guess.
Also, check your cell phone. Despite how ubiquitous they are in our daily lives, I don't think I've seen a single cell phone in my dreams. Or any other phone, for that matter.
And now that I think about it, I've definitely had a dream of being in my living room where there's a TV, but I don't remember the TV actually being in the dream.
Weird.
It's not the exact same problems humans have. It's completely different. Marketers and hucksters just use anthropomorphic terminology to hype their dysfunctional programs.
load more comments
(1 replies)
That's what it comes by not really understanding what you're doing. Most of the AI models I work with are the state of the art just because they happen to work.
In my case, when I solve a PDE using finite difference schemes, there are precise mathematical conditions that guarantee you if the method is going to be stable or not. When I do the same using AI, I can't tell if my method is going to work or not unless I run it. Moreover, I've had it sometimes fail and sometimes succeed.
It's just the way it is for now.
I mean companies world wide just jumped in the AI bandwagon like a lot of people did with the NFT one. Mostly because AI actually has solid use cases and can make a big difference in broad situations.
Just since people are just slapping AI in everything it's gonna end up being another fad to raise stock prices, like firing people last year.
Let's just hope when all of the hype blows over and the general public thinks of AI as the marketing buzzword that never works quite right we'll keep AI in the things it's actually useful for
AI interest has come and gone. Some decades ago, people would slap the AI label to expert systems. If we go further back, one would call AI to solving problems in blocks world. It's eventually going to fade away, just like all the previous waves did.
He's only being honest for the sake of the shareholders.
This is the best summary I could come up with:
Even Apple CEO Tim Cook isn’t sure the company can fully stop AI hallucinations.
In an interview with The Washington Post, Cook said he would “never claim” that its new Apple Intelligence system won’t generate false or misleading information with 100 percent confidence.
These features will let you generate email responses, create custom emoji, summarize text, and more.
Recent examples of how AI can get things wrong include last month’s incident with Google’s Gemini-powered AI overviews telling us to use glue to put cheese on pizza or a recent ChatGPT bug that caused it to spit out nonsensical answers.
The voice assistant will turn to ChatGPT when it receives a question better suited for the chatbot, but it will ask for your permission before doing so.
In the demo of the feature shown during WWDC, you can see a disclaimer at the bottom of the answer that reads, “Check important info for mistakes.”
The original article contains 334 words, the summary contains 153 words. Saved 54%. I'm a bot and I'm open source!
I doubt anyone can for as long as "AI" is synonymous with LLMs. LLMs are just inherently unreliable because of how they work.
I'm not exaggerating when I say there's only like a dozen true experts for generative AI on the planet and even they're not completely sure what's going on in that blackbox. And as far as I'm aware Tim Cook isn't even one of them. How would he know?
I would expect that Apple has hired some of those experts and they told him.
These programs are averaging massive amounts of data into a massive averaging function. There's no way that a human could ever understand what's going on inside that kind function. Humans can't hold millions of weights/etc in their head and comprehend what it means. Otherwise, if humans could do this, there would be no point in doing this kind of statistics with computers.
I only trust moguls and political figures that are 100% sure of everything. I really like the confidence and it makes me feel like they deserve big paychecks and special rights because they must be so smart to have have no room for the doubt like the rest of us spineless imps. This guy is displaying weakness and should be shamed!
I bet Tim Apple is going to fire his ass.
I'm 100% sure they can't because what they call AI isn't intelligence.
Even people hallucinate. Under your definition intelligence doesn't exist
Wow whoosh. The point is that "AI" isn't actually "intelligent" like a human and thus can't "hallucinate" like an intelligent human.
All of this anthropomorphic terminology is just misleading marketing bullshit.
Who said anything about human intelligence? AIs have a different kind of intelligence, an artificial kind. I'm tired of pretending they don't
Ever heard of the Turing test? Ever since AIs could pass it it became not a thing. Before that, playing Go was the mark of AI.
Any time an AI achieves a new thing people move goalposts. So I ask you: what does AI need to achieve to have intelligence?
The Turing Test says that any person could have any conversation with a machine and there's no chance you could tell it's a machine. It does not say that one person could have one conversation with a machine and not be able to tell.
Current text generation models out themselves all the damn time. It can't actually understand the underlying concepts of words. It just predicts what bit of text would be most convincing to a human based on previous text.
Playing Go was never the mark of AI, it was the mark of improving game-playing machines. It doesn't represent "intelligence", only an ability to predict what should happen next based on a set of training data.
It's worth noting that after Lee Se Dol lost to Alphago, researchers found a fairly trivial Go strategy that could reliably beat the machine. It was simply such an easy strategy to counter that none of the games in the training data had included anyone attempting that strategy, so the algorithm didn't account for how to counter it. Because the computer doesn't know Go theory, it only knows how to predict what to do next based on the training data.
load more comments
(5 replies)
The same thing actually passing a turing test would require. You've obviously read the words "Turing test" somewhere and thought you understood what it meant, but no robot we've ever produced as a species has passed the turing test. It EXPLICITLY requires that intelligence equal to (or indistinguishable from) HUMAN intelligence is shown. Without a liar reading responses, no AI we'll produce for decades will pass the turing test.
No large language model has intelligence. They're just complicated call and response mechanisms that guess what answer we want based on a weighted response system (we tell it directly or tell another machine how to help it "weigh" words in a response). Obviously with anything that requires massive amounts of input or nuance, like language, it'll only be right about what it was guided on, which is limited to areas it is trained in.
We don't have any novel interactions with AI. They are regurgitation engines, bringing forward sentences that aren't theirs piecemeal. Given ten messages, I'm confident no major LLM would pass a Turing test.
load more comments
(4 replies)
load more comments
(13 replies)
No, really, if you understood how the language models work, you would understand it's not really intelligence. We just tend to humanize it because that's what our brains do.
There's a lot of great articles that summarize how we got to this stage and it's pretty interesting. I'll try to update this post with a link later.
I think LLMs are useful (and fun) and have a place, but intelligence they are not.
I'm still waiting for the definition of intelligence that won't have the same moving of goalposts the Turing Test did
I'm happy with the Oxford definition: "the ability to acquire and apply knowledge and skills".
LLMs don't have knowledge as they don't actually understand anything. They are algorithmic response generators that apply scores to tokens, and spit out the highest scoring token considering all previous tokens.
If asked to answer 10*5, they can't reason through the math. They can only recognize 10, * and 5 as tokens in the training data that is usually followed by the 50 token. Thus, 50 is the highest scoring token, and is the answer it will choose. Things get more interesting when you ask questions that aren't in the training data. If it has nothing more direct to copy from, it will regurgitate a sequence of tokens that sounds as close as possible to something in the training data: thus a hallucination.
load more comments
(12 replies)
load more comments
(1 replies)
load more comments
(9 replies)
load more comments
(3 replies)
They could make Siri change its voice and Genmoji based on the degree of certainty of the response:
- Trust me: Arnold as Terminator 😎
- Eehhhh, could be bullshit: shrugging old man meme 🤷🏻♂️
- Just kiddin' here: whacky Jerry Lewis 🤪
They could sell different voice packages. Revive the ringtone market.
The AI is confidently wrong, that's the whole problem. If there was an easy way to know if it could be wrong we wouldn't have this discussion
While it can’t “know” its own confidence level, it can distinguish between general knowledge (12” in 1’) and specialized knowledge that requires supporting sources.
At one point, I had a chatGPT memory designed for it to automatically provide sources for specialized knowledge and it did a pretty good job.
load more comments
(1 replies)
They can't. AI has hallucinations. Google has shown that AI can't even rely on external sources, either.
load more comments
(1 replies)
As others are saying it's 100% not possible because LLMs are (as Google optimistically describes) "creative writing aids", or more accurately, predictive word engines. They run on mathematical probability models. They have zero concept of what the words actually mean, what humans are, or even what they themselves are. There's no "intelligence" present except for filters that have been hand-coded in (which of course is human intelligence, not AI).
"Hallucinations" is a total misnomer because the text generation isn't tied to reality in the first place, it's just mathematically "what next word is most likely".
all we know about ourselves is what's in our memories. the way normal writing or talking works is just picking what words sound best in order
That's not the whole story. "The dog swam across the ocean." is a grammatically valid sentence with correct word order. But you probably wouldn't write it because you have a concept of what a dog actually is and know its physiological limitations make the sentence ridiculous.
The LLMs don't have those kind of smarts. They just blindly mirror what we do. Since humans generally don't put those specific words together, the LLMs avoid it too, based solely on probability. If lots of people started making bold claims about oceanfaring canids (e.g. as a joke), then the LLMs would absolutely jump onboard with no critical thinking of their own.
load more comments
(1 replies)
An LLM once explained to me that it didn't know, it simulated an answer. I found that descriptive.
I was wondering, are people working on networks that train to create a modular model of the world, in order to understand it / predict events in the world?
I imagine that that is basically what our brains do.
load more comments
(4 replies)
They do have internal concepts though: https://www.lesswrong.com/posts/yzGDwpRBx6TEcdeA5/a-chess-gpt-linear-emergent-world-representation
Probably not of what a human is, but thought process is needed for better text generarion and is therefore emergent in their neural net
load more comments
(3 replies)
load more comments
(1 replies)
Tim Apple
If Apple can stop AI hallucination, any other AI company can also stop AI hallucination. Which is something they could have already done instead of making AI seem a joke on purpose. AI hallucinations are a sort of phenomena that nobody has control over. Why would Tim Cook have unique control over it?
Unless Apple became the first to figure out how, then they suddenly have a huge leg up on the rest. Which is kinda how Apple has been making their bread for most of their successes in my lifetime
eh. I don't think Apple's gonna be a pioneer in AI. If anybody can do it, it would be openai figuring it out first. Happy to be proven wrong tho.
Oh I’m not suggesting the will or are able to, I’m coming from a strategic standpoint
Yeah. When Apple says it’s coming into a market, they mean they have already perfected it.
(Or let other companies polish up a feature/concept for a few years, slap a coat of Space Gray on it, and release it as a revolutionary "new" feature for apple)
load more comments
(1 replies)
view more: next ›