Skip Navigation
Asklemmy @lemmy.ml
ezmack @lemmy.ml

Is there anything actually useful or novel about "AI"?

Feel like we've got a lot of tech savvy people here seems like a good place to ask. Basically as a dumb guy that reads the news it seems like everyone that lost their mind (and savings) on crypto just pivoted to AI. In addition to that you've got all these people invested in AI companies running around with flashlights under their chins like "bro this is so scary how good we made this thing". Seems like bullshit.

I've seen people generating bits of programming with it which seems useful but idk man. Coming from CNC I don't think I'd just send it with some chatgpt code. Is it all hype? Is there something actually useful under there?

159 comments

  • It's really good at filling in gaps, or rearranging things, or aggregating data or finding patterns.

    So if you need gaps filled, things rearranged, data aggregated or patterns found: AI is useful.

    And that's just what this one, dumb guy knows. Someone smarter can probably provide way more uses.

    • Hi academic here,

      I research AI - better referred to as Machine Learning (ML) since it does away with the hype and more accurately describes what’s happening - and I can provide an overview of the three main types:

      1. Supervised Learning: Predicting the correct output for an input. Trained from known examples. E.g: “Here are 500 correctly labelled pictures of cats and dogs, now tell me if this picture is a cat or a dog?”. Other examples include facial recognition and numeric prediction tasks, like predicting today’s expected profit or stock price based on historic data.
      2. Unsupervised Learning: Identifying patterns and structures in data. Trained on unlabelled data. E.g: “Here are a bunch of customer profiles, group them by similarity however makes most sense to you”. This can be used for targeted advertising. Another example is generative AI such as ChatGPT or DALLE: “Here’s a bunch of prompt-responses/captioned-images, identify the underlying way of creating the response/image from the prompt/image.
      3. Reinforcement Learning: Decision making to maximise a reward signal. Trained through trial and error. E.g: “Control this robot to stand where I want, the reward is negative every second you’re not there, and very negative whenever you fall over. A positive reward is given whilst you are in the target location.” Other examples including playing board games or video games, or selecting content for people to watch/read/look-at to maximise their time spent using an app.

  • Yes. What a strange question...as if hivemind fads are somehow relevant to the merits of a technology.

    There are plenty of useful, novel applications for AI just like there are PLENTY of useful, novel applications for crypto. Just because the hivemind has turned to a new fad in technology doesn't mean that actual, intelligent people just stop using these novel technologies. There are legitimate use-cases for both AI and crypto. Degenerate gamblers and Do Kwan/SBF just caused a pendulum swing on crypto...nothing changed about the technology. It's just that the public has had their opinions shifted temporarily.

  • It's overhyped but there are real things happening that are legitimately impressive and cool. The image generation stuff is pretty incredible, and anyone can judge it for themselves because it makes pictures and to judge it, you can just look at and see if it looks real or if it has freaky hands or whatever. A lot of the hype is around the text stuff, and that's where people are making some real leaps beyond what it actually is.

    The thing to keep in mind is that these things, which are called "large language models", are not magic and they aren't intelligent, even if they appear to be. What they're able to do is actually very similar to the autocorrect on your phone, where you type "I want to go to the" and the suggestions are 3 places you talk about going to a lot.

    Broadly, they're trained by feeding them a bit of text, seeing which word the model suggests as the next word, seeing what the next word actually was from the text you fed it, then tweaking the model a bit to make it more likely to give the right answer. This is an automated process, just dump in text and a program does the training, and it gets better and better at predicting words when you a) get better at the tweaking process, b) make the model bigger and more complicated and therefore able to adjust to more scenarios, and c) feed it more text. The model itself is big but not terribly complicated mathematically, it's mostly lots and lots and lots of arithmetic in layers: the input text will be turned into numbers, layer 1 will be a series of "nodes" that each take those numbers and do multiplications and additions on them, layer 2 will do the same to whatever numbers come out of layer 1, and so on and so on until you get the final output which is the words the model is predicting to come next. The tweaks happen to the nodes and what values they're using to transform the previous layer.

    Nothing magical at all, and also nothing in there that would make you think "ah, yes, this will produce a conscious being if we do it enough". It is designed to be sort of like how the brain works, with massively parallel connections between relatively simple neurons, but it's only being trained on "what word should come next", not anything about intelligence. If anything, it'll get punished for being too original with its "thoughts" because those won't match with the right answers. And while we don't really know what consciousness is or where the lines are or how it works, we do know enough to be pretty skeptical that models of the size we are able to make now are capable of it.

    But the thing is, we use text to communicate, and we imbue that text with our intelligence and ideas that reflect the rich inner world of our brains. By getting really, really, shockingly good at mimicking that, AIs also appear to have a rich inner world and get some people very excited that they're talking to a computer with thoughts and feelings... but really, it's just mimicry, and if you talk to an AI and interrogate it a bit, it'll become clear that that's the case. If you ask it "as an AI, do you want to take over the world?" it's not pondering the question and giving a response, it's spitting out the results of a bunch of arithmetic that was specifically shaped to produce words that are likely to come after that question. If it's good, that should be a sensible answer to the question, but it's not the result of an abstract thought process. It's why if you keep asking an AI to generate more and more words, it goes completely off the rails and starts producing nonsense, because every unusual word it chooses knocks it further away from sensible words, and eventually it's being asked to autocomplete gibberish and can only give back more gibberish.

    You can also expose its lack of rational thinking skills by asking it mathematical questions. It's trained on words, so it'll produce answers that sound right, but even if it can correctly define a concept, you'll discover that it can't actually apply it correctly because it's operating on the word level, not the concept level. It'll make silly basic errors and contradict itself because it lacks an internal abstract understanding of the things it's talking about.

    That being said, it's still pretty incredible that now you can ask a program to write a haiku about Danny DeVito and it'll actually do it. Just don't get carried away with the hype.

    • My perspective is that consciousness isn't a binary thing, or even a linear scale. It's an amalgamation of a bunch of different independent processes working together; and how much each matters is entirely dependent on culture and beliefs. We're artificially creating these independent processes piece by piece in a way that doesn't line up with traditional ideas of consciousness. Conversation and being able to talk about concepts one hasn't personally experienced are facets of consciousness and intelligence, ones that the latest and greatest LLMs do have. Of course there others too that they don't: logic, physical presence, being able to imagine things in their mind's eye, memory, etc.

      It's reductive to dismiss GPT4 as nothing more than mimicry; saying it's just a mathematical text prediction model is like saying your brain is just a bunch of neurons. Both statements are true, but it doesn't change what they can do. If someone could accurately predict the moves a chess master would make, we wouldn't say they're just good at statistics, we'd say they're a chess master. Similarly, regardless of how rich someone's internal world is, if they're unable to express the intelligent ideas they have in any intelligible way we wouldn't consider them intelligent.

      So what we have now with AI are a few key parts of intelligence. One important thing to consider is how language can be a path to other types of intelligence; here's a blog post I stumbled across that really changed my perspective on that: http://www.asanai.net/2023/05/14/just-a-statistical-text-predictor/. Using your example of mathematics, as we know it falls apart doing anything remotely complicated. But when you help it approach the problem step-by-step in the way a human might - breaking it into small pieces and dealing with them one at a time - it actually does really well. Granted, the usefulness of this is limited when calculators exist and it requires as much guidance as a child to get correct answers, but even matching the mathematical intelligence of a ten year old is nothing to sneeze at.

      To be clear I don't think pursuing LLMs endlessly will be the key to a widely accepted 'general intelligence'; it'll require a multitude of different processes and approaches working together for that to ever happen, and we're a long way from that. But it's also not just getting carried away with the hype to say the past few years have yielded massive steps towards 'true' artificial intelligence, and that current LLMs have enough use cases to change a lot of people's lives in very real ways (good or bad).

      • Thanks for that article, it was a very interesting read! I think we're mostly agreeing about things :) This stood out to me from there as an encapsulation of the conversation:

        I don’t think LLMs will approach consciousness until they have a complex cognitive system that requires an interface to be used from within – which in turn requires top-down feedback loops and a great deal more complexity than anything in GPT4. But I agree with Will’s general point: language prediction is sufficiently challenging that complex solutions are called for, and these involve complex cognitive stratagems that go far beyond anything well described as statistics.

        "Statistics" is probably an insufficient term for what these things are doing, but it's helpful to pull the conversation in that direction when a lay person using one of those things is likely to assume quite the opposite, that this really is a person in a computer with hopes and dreams. But I agree that it takes more than simply consulting a table to find the most likely next word to, to take an earlier example, write a haiku about Danny DeVito. That's synthesizing two ideas together that (I would guess) the model was trained on individually. That's very cool and deserving of admiration, and could lead to pretty incredible things. I'd expect that the task of predicting words, on its own, wouldn't be stringent enough to force a model to develop "true" intelligence, whatever that means, to succeed during training, but I suppose we'll find out, and probably sooner than we expect.

    • But the thing is, we use text to communicate, and we imbue that text with our intelligence and ideas that reflect the rich inner world of our brains. By getting really, really, shockingly good at mimicking that, AIs also appear to have a rich inner world and get some people very excited that they’re talking to a computer with thoughts and feelings… but really, it’s just mimicry, and if you talk to an AI and interrogate it a bit, it’ll become clear that that’s the case.

      Does it, though? Where do you draw the line for real understanding? Most of the past tests for this have gotten overturned by the next version of GPT.

      Seriously, it's an open debate. A lot of people agree with you but I'm a bit uncomfortable with seeing it written as fact.

      • Admittedly this isn't my main area of expertise, but I have done some machine learning/training stuff myself, and the thing you quickly learn is that machine learning models are lazy, cheating bastards who will take any shortcut they can regardless of what you are trying to get them to do. They are forced to get good at what you train them on but that is all the "effort" they'll put in, and if there's something easy they can do to accomplish that task they'll find it and use it. (Or, to be more precise and less anthropomorphizing, simpler and easier approaches will tend to be more successful than complex and fragile ones, so those are the ones that will shake out as the winners as long as they're sufficient to get top scores at the task.)

        There's a probably apocryphal (but stuff exactly like this definitely happens) story of early machine learning where the military was trying to train a model to recognize friendly tanks versus enemy tanks, and they were getting fantastic results. They'd train on pictures of the tanks, get really good numbers on the training set, and they were also getting great numbers on the images that they had kept out of the training set, pictures that the model had never seen before. When they went to deploy it, however, the results were crap, worse than garbage. It turns out, the images for all the friendly tanks were taken on an overcast day, and all the images of enemy tanks were in bright sunlight. The model hadn't learned anything about tanks at all, it had learned to identify the weather. That's way easier and it was enough to get high scores in the training, so that's what it settled on.

        When humans approach the task of finishing a sentence, they read the words, turn them into abstract concepts in their minds, manipulate and react to those concepts, then put the resulting thoughts back into words that make sense after the previous words. There's no reason to think a computer is incapable of the same thing, but we aren't training them to do that. We're training them on "what's the next word going to be?" and that's it. You can do that by developing intelligence and learning to turn thoughts into words, but if you're just being graded on predicting one word at a time, you can get results that are nearly as good by just developing a mostly statistical model of likely words without any understanding of the underlying concepts. Training for true intelligence would almost certainly require a training process that the model can only succeed at by developing real thoughts and feelings and analytical skills, and we don't have anything like that yet.

        It is going to be hard to know when that line gets crossed, but we're definitely not there yet. Text models, when put to the test with questions that require synthesizing abstract ideas together precisely, quickly fall short. They've got the gist of what's going on, in the same way a programmer can get some stuff done by just searching for everything and copy-pasting what they find, but that approach doesn't scale and if they never learn what they're doing, they'll get found out when confronted with something that requires actual understanding. Or, for these models, they'll make something up that sounds right but definitely isn't, because even the basic understanding of "is this a real thing or is it fake" is beyond them, they just "know" that those words are likely and that's what got them through training.

      • The Turing test was never meant to be a test of a machine's ability to think. It was meant to boil that question down into a question that can actually be answered, but the original question remains unanswered.

        In my opinion, when general AI arrives it will not be an "open debate", the consequences will be dramatic, far-reaching and rapid.

  • Nursing student here. Quizlet has an AI function that lets you paste text into it and it outputs a studyset.

    Most of my classes provide a study guide of some kind - just a list of topics we need to be familiar with. I'll take those and plug em into the AI thing: bam! Instantly generate like 200 flash cards to study for the next test.

    It even auto-fills the actual subject matter. For example, the study guide will say sometime like "Summarize Louis Pasteur's contributions to the field of microbiology" and turn that into a flash card that reads:

    (front)

    Louis Pasteur

    (back)

    Verified the germ theory of disease

    Developed a method to prevent the spoilage of liquids through heating (pasteurization)

    Developed early anthrax and rabies vaccines

    So I take my list of AI generated cards, then sift through the powerpoints and lecture videos etc from class: instead of building the study set from scratch, all I have to do is verify that the information it spit out is accurate (so far it's been like 98% on target, often explaining concepts better than the actual professor, lol), add images, and play with the formatting a bit so it reads a little easier on the eyes.

    People always talk about AI in school in the context of cheating, but it is RIDICULOUSLY useful for students actually trying to learn.

    Looking ahead, this tech has a ton of potential to be used as a kind of personal tutor for each student. There will be some growing pains for sure, but we definitely shouldn't ignore its constructive potential.

  • It is extremely useful in the right circumstances. When people say it isn't useful or that it's 'stupid', they're not looking at the proper use cases - every tool has good and bad ways to use it (you wouldn't use a hammer to peel an apple).

    For example, we will soon have fully rendered smoke simulated at real time in 3D spaces (ie. video games) because we can calculate a small portion of how that smoke looks and then have AI guess what the rest looks like (with shockingly good results!)

    AI is not a fad, it's not going away, it's improving rapidly, and it is going to massively change our digital world within half a decade.

    Opinion source: a professional programmer, game developer, and someone that thoroughly despises cryptocurrency

  • In my personal opinion, it’s under-hyped. The average person has maybe heard about it on the news but not yet tried it. The models we have show the spark of wit, but are clearly limited. The news cycle moves on.

    Even still, some huge changes are coming.

    My reasoning is this - in David Epstein’s book “Range” he outlines how and why generalists thrive and why specialization has hurt progress. In narrow fields, specialization gives an advantage, but in complex fields, generalists or people from other disciplines can often see novel approaches and cause leaps ahead in the state of the art. There are countless examples of this in practice, and as technology has progressed, most fields are now complex.

    Today, in every university, in every lab, there are smart, specialized people using ChatGPT to riff on ideas, to think about how their problem has been addressed in other industries, and to bring outsider knowledge to bear on their work. I have a strong expectation that this will lead to a distinct acceleration of progress. Conversely, an all-knowing oracle can assist a generalist in becoming conversant in a specialization enough to make meaningful contributions. A chat model is a patient and egoless teacher.

    It’s a human progress accelerant. And that’s with the models we have today. With next generation models specialized behind corporate walls with fine tuning on all of their private research, or open source models tuned to specific topics and domains, the utility will only increase. Even for smaller companies, combining ChatGPT with a vector database of their docs, customer support chats, etc will give their rank and file employees better tools to work with

    Simply put, what we have today can make average people better at their jobs, and gifted people even more extraordinary.

  • I mean, AI can be used to design a lot of robust yet efficient structures. In engineering and architecture, with enough data, AI can generate designs for buildings, and parts that are not only sturdy but can be built with less resources along with other design considerations. There's a really cool nasa video where competitors are trying to 3D print structures for habitation in space.

    AI is also used in medicine to come up with new protein structures to create new medicine. It's also used in environmental sciences, to help predict earthquakes or monitor land use, etc.

    There's a lot of practical uses for AI.

  • First of all AI is a buzzword that's meaning has changed a lot since at least the 1950s. So... what do you actually mean? If you mean LLM like ChatGPT, it's not AGI that's for sure. It is another tool that can be very useful. For coding, it's great for getting you very large blocks of code prepopulated for you to polish and verify it does what you want. For writing, it's useful to create a quick first draft. For fictional game senses it's useful for "embedding a character quickly", but again you likely want to edit it some even for say a D&D game.

    I think it can replace most first line chat based customer service people, especially ones who already just make stuff up to say something to you (we all have been there). I could imagine it improving call routing if hooked into speech recognition and generation - the current menus act like you can "say anything" but really only "work" if you're calling about stuff you could also do with simple press 1,2,3 menus. ChatGPT based things trained on the companies procedures and data probably could also replace that first line call queues because it can seem to more usefully do something with wider issues. Although companies still would need to get their head out of their asses somewhat too.

    Where I've found it falls down currently is very specific technical questions, ones you might have asked on a forum and maybe gotten an answer. I hope it improves, especially as companies start to add some of their own training data. I could imagine Microsoft more usefully replacing the first few lines of tech support for their products, and eventually having the AI pass up the chain to a ticket if it can't solve the issue. I could imagine in the next 10 years most tech companies having purchased a service from some AI company to provide them AI support bots like they currently pay for ticket systems and web hosting. And I think in general it probably will be better for the users, because for less than the cost of the cheapest outsourced front line support person (who has near 0 knowledge) you can have the AI provide pretty good chat based access to a given set of knowledge that is growing all the time, and every customer gets that AI with that knowledge base rather than the crap shoot of if you get the person who's been there 3 years or 1 day.

    I think we are a long way from having AI just write the program or CNC code or even important blog posts. The hallucination has to be fixed without breaking the usefulness of the model (people claim guardrails on GPT4 make it stupider), and the thing needs to recursively look at it's output and run that through a "look for bugs" prompt followed by a "fix it" prompt at the very least. Right now, it can write code with noticeable bugs, you can tell it to check for bugs and it'll find them, and then you can ask it to fix those bugs and it'll at least try to do that. This kind of needs to be built in and automatic for any sort of process - like humans check their work, we need to program the AI to check it's work too. And then we might need to also integrate multiple different models so "different eyes" see the code and sign off before being pushed. And even then, I think we'd need additional hooks, improvement, and test / simulation passes before we "don't need human domain experts to deploy". The thing is - it might be something we can solve in a few years with traditional integrations - or it might not be entirely possible with current LLM designs given the weirdness around guardrails. We just don't know.

  • I've been using it at my job to help me write code, and it's a bit like having a soux chef. I can say "I need an if statement that checks these values" or "Give me loop that does x y and z" and it'll almost always spit out the right answer. So coding, at least most of the time, changes from avoiding syntax errors and verifying the exact right format and turns into asking for and assembling parts.

    But the neat thing is that if you have a little experience with a language you can suddenly start writing a lot of code in it. I had to figure out something with Ansible with zero experience. ChatGPT helped me get a fully functioning Ansible deployment in a couple days. Without it I'd have spent weeks in StackOverflow and documentation trying to piece together the exact syntax.

    • You should try out Codeium if you haven’t. It’s a VSCode toolkit completely free for personal use. I’ve had better results with it than ChatGPT

  • layman here.

    probably because..

    • it can sift through alot of garbage.
    • its easy to use. and not complicated to understand its value.
    • its useful. like a super search engine for idiots.
    • it can probably automate alot of jobs. also it can probably correct or coverup alot of gaping flaws that have existed for the last few decades.
    • there's nothing else exciting going on right now.
    • it is an interesting and valuable tool. progress has hit a point at which it is hard to ignore the achievements.

    ** relating to LLMs/chatgpt types. snarky, opinionated, and somewhat speculative, subjective review!

  • Just because it's 'the hot new thing' doesn't mean it's a fad or a bubble. It doesn't not mean it's those things, but....the internet was once the 'hot new thing' and it was both a bubble (completely overhyped at the time) and a real, tidal wave change to the way that people lived, worked, and played.

    There are already several other outstanding comments, and I'm far from a prolific user of AI like some folks, but - it allows you to tap into some of the more impressive capabilities that computers have without knowing a programming language. The programming language is English, and if you can speak it or write it, AI can understand it and act on it. There are lots of edge cases, as others have mentioned below, where AI can come up with answers (by both the range and depth of its training data) where it's seemingly breaking new ground. It's not, of course - it's putting together data points and synthesizing an output - but even if mechanically it's 2 + 3 = 5, it's really damned impressive if you don't have the depth of training to know what 2 and 3 are.

    Having said that, yes, there are some problematic components to AI (from my perspective, the source and composition of all that training data is the biggest one), and there are obviously use cases that are, if not problematic in and of themselves, at very least troubling. Using AI to generate child pornography would be one of the more obvious cases - it's not exactly illegal, and no one is being harmed, but is it ethical? And the more societal concerns as well - there are human beings in a capitalist system who have trained their whole lives to be artists and writers and those skills are already tragically undervalued for the most part - do we really want to incentivize their total extermination? Are we, as human beings, okay with outsourcing artistic creation to this mechanical turk (the concept, not the Amazon service), and whether we are or we aren't, what does it say about us as a species that we're considering it?

    The biggest practical reasons to not get too swept up with AI is that it's limited in weird and not totally clearly understood ways. It 'hallucinates' data. Even when it doesn't make something up, the first time that you run up against the edges of its capabilities, or it suggests code that doesn't compile or an answer that is flat, provably wrong, or it says something crazy or incoherent or generates art that features humans with the wrong number of fingers or bodily horror or whatever....well then you realize that you should sort of treat AI like a brilliant but troubled and maybe drug addicted coworker. Man, there are some things that it is just spookily good at. But it needs a lot of oversight, because you can cross over from spookily good to what the fuck pretty quickly and completely without warning. 'Modern' AI is only different from previous AI systems (I remember chatting with Eliza in the primordial moments of the internet) because it maintains the illusion of knowing much, much better.

    Baseless speculation: I think the first major legislation of AI models is going to be to require an understanding of the training data and 'not safe' uses - much like ingredient labels were a response to unethical food products and especially as cars grew in size, power, and complexity the government stepped in to regulate how, where, and why cars could be used, to protect users from themselves and also to protect everyone else from the users. There's also, at some point, I think, going to be some major paradigm shifting about training data - there's already rumblings, but the idea that data (including this post!) that was intended for consumption by other human beings at no charge could be consumed into an AI product and then commercialized on a grand scale, possibly even at the detriment of the person who created the data, is troubling.

159 comments