What's the endgame? It's inadequate as a substitute for humans, it cannot replicate human judgements, context, or institutional knowledge. Employers are now rehiring employees laid off due to AI. It effectively doesn't do anything. It's more reasonable to deduce that they're trying to establish a police state with the increasing presence of cctv and flock cameras.
>>538536202>s inadequate as a substitute for humans, itBruh it's better than like 80% of the population rn. And thats the stuff they let us play with
>>538536363>better than like 80% of the populationMaybe for coding and software engineering. That's about it.
>>538536202AI systems allow domain experts to leverage their experience to output more work The value comes from high quality instructions you write for the machine to execute
>>538536202>what’s the endgameTo spy on everything you ever write or send online, and keep a massive backlog of it in their data centres. You think chatcontrol is an EU thing?American companies like Meta, Google, discord, etc, are mandated by law to scan every image and message you ever send.
>>538536454No for just about anything. I mean the bottom half are functionally retarded
>>538536527
>>538536202>What's the endgame? Iits an ok code generator for software developers, but I doubt it can be useful for people who are not software developers.the endgame? In china its a technology developed to be integrated in industrial production and new products.In America its different, funded by Usury the AI remaining AI startups entered in collusion with hardware manufacturers, the AI startups get access to hardware at cheaper prices while the manufacturers increase the price artificially for consumers.The idea is to make x64/x86 computers obsolete and replace them with fixed hardware ARM architecture PCs, that will run an Agent 24/7 having full control over your device, and of course monitor everything you do, the idea is not just to put an agent in your computer to spy on you but to force people into buying fixed hardware that most likely will get obsolete in 2 years under the excuse that "AI" and what not.
>>538536527This
>>538536724Yea you can't wholesale replace entire swathes of people/jobs yet. But take a rando off the street, and give them and Claude the same task, see who completes it better/faster
>>538536798Also engineers are probably not in that bottom percent
>>538536755Thank you. I will look into tat.
>>538536202take care of the cat
They are trying to consolidate power and build weather-resistant shelters before our civilization inevitably breaks down in the next few centuries.
>>538536202
>>538536202because asia has sweet Virtual Reality worlds to make your wytebois brains melt much like the old games on playstation... you just fucking wait wytebois, unless you're a faggot that said something about chinks having a small dick then dont count on being invited
>>538536202it's just outsourcing testing&QA to the general public while trying to turn a profit sometimesI'd be more worried about someone accidentally a whole skynet before anything else, and that doesn't come across as particularly likely. That, and prices are going to be fucked for ages.
>>538536755good post anon
>>538536202First stage is the police state, down the line say 30-50 years, total AI robotics catering to the super wealthy while all the poors "unfortunately" are sterile due to a reason that can't be said.
>>538538597Bump. Thanks
>>538536202>it cannot replicate human judgements, context, or institutional knowledgeThat is not quite true, no. It basically can.That OpenAI and Anthropic don't deliver a nicely made version of it... sure, but that's not a thing in the technology itself.
>>538539052And in the process American citizens are shouldering some of the costs of building and maintaining data centers what with higher electricity and utility bills. Some locales have even experienced pollution of local drinking water. I live in Virginia which seems to be ground zero for data center construction. AI is, very much, a bad deal for Americans. Why is our government tolerating this?
>>538539473Dude, AI is just a glorified search engine, a numbers cruncher. Its output is only what humans have input into the machine.
>>538539785Either humans are also glorified numbers crunchers where the output is only some transfomation of something that was learned or you got this description somewhat wrong.Either way, AI very much demonstrates the same capabilities as humans have. There is no test you have now that can show a fundamental issue and lots of tests that can show most to all of the human abilities, sometimes a bit coarse yet but it's there.
Nice thread.I'm a software dev, also have a degree in which I learned the basics of AI, and I've also been updating my knowledge with internet content that really goes into depth on the maths of how these new AIs (models) work.Based on that, my take is that the way it works has some fundamental flaws that can't be overcome by improving it. The most obvious one is the lies, that's not just "hallucinations" that they can just refine it into staying sober and not "hallucinating". It's inherent to how the technology works, and it's not going to stop.AI that's comparable to people's minds is a purely science fiction concept right now, similar to colonizing Mars, and we're not close to doing that. What we have is models that are very good at writing, so much so that people think they're smart. In reality, they lack human capabilities such as understanding what it means that something is or isn't true, having a goal, trying to convey a point... Basically anything that requires an actual understanding of anything.An example of why its fundamentally flawed could be that it could be decades ahead of where it is now, very improved and refined, and as long as they're improving on the same underlying technology and don't have a new system, and as a response to a scientific or medical question, between an answer with just the truth and another with an entirely made up text that reads just like other texts about the topic, with similar terms and writing style, they may both be just as good to the language model, as they both adhere to the pattern of "what would a typical response to this be", for different reasons. The truth means nothing to it. I've been trying to explain this to normies and no one gets it.It's not going to replace most jobs people think it will replace, due to a total lack of common sense. A normie may think "this sounds so smart! ooh man most accountants will be jobless soon!!", in reality they wont if their job involves even a little bit of judgement
>>538540078Then there's the problem of software development. I actually think, due to software dev being basically "implement from this design and specs" and the fact that it can be tested and redone over and over, that we have a real problem with that and we may be the ones to be actually replaced. We'll see. Maybe not.Of course, it will take longer than most people think, if anything. Everyone who thinks we're close to that or that it can already do our jobs is just very bad at reasoning.
>>538540046>AI very much demonstrates the same capabilities as humans haveIt's easy to believe it is smart because it sounds smart.I don't blame you. There's a reason being skilled at sales is a thing. There's a reason job interviewing skills are a thing. None of those would make sense if most people didn't literally only respond to stimuli instead of thinking.The reality, however, is that AI is very far from being even close to human capabilities, in terms of common sense or understanding of anything. Which should be obvious when it doesnt even know what the truth is or what it means for something to be true or made up. That would be the most obvious way to tell that I think even normies should understand.
>>538540299No, it's the same capabilities. Not "it sounds smart".About the hallucinations argument: Ok the ones AI may still have are still relatively worse but in the end: Have you found a human that answers correctly in all instances and all details?Humans can't do it either and like AI they will always hallucinate too. Because humans are ALSO not actually exhaustively proving/disproving all aspects of what they "thought" rooted on some completely formal and sound system. Yea, humans hallucinate too.
>>538540440You're missing the point.AI does not actually "hallucinate", as saying that it hallucinates sounds like its a temporary thing, like a state of trance, or a dream, before going back to normal. Even if you do not know anything about how it works, to a user, it should be obvious as fuck that the AI "hallucinates" because it not only does not understand that its supposed to tell the truth, but in fact doesn't even understand what it means that something is true or not. AI does not have those kind of capabilities.
>>538540440Also I forgot to elaborate on this.Of course people do not answer everything correctly always. At all. But they certainly try, unless they have bad intentions. Even just this (having intentions at all, trying to accomplish anything, a concept of what's ok and what isn't, and of course, trying to tell the truth) is light years ahead of AI capabilities already. And I didn't even elaborate on other reasons.You literally just think its smart because of the well written slop that sounds smart.Yes people hallucinate too. Every day actually. We literally hallucinate every time we fall asleep and dream, or if we get high, or get sick. The key here is that we have to be in a state that's different from normal.
>>538540596Well yes, AI can now reason "what it means" based on evaluations very much like humans do them.>>538540814> But they certainly try, unless they have bad intentions.So does AI. And anon, basically no one can come up with actual tests anymore that show humans are supposedly "lightyears" better in thinking. That's the biggest issue.Now the people who hope AI is fundamentally different have nothing really to show.
>>538541373Evaluations are a meme. Teachers in schools are currently struggling because students do their tasks with chatGPT kek. We do not have evaluations for the capabilities I was just talking about. Like common sense.Like I was just saying, LLMs "hallucinate" because they don't even know what it even means that something is true or made up. ffs. And this should be obvious to users btw.All a language model is trying to do is "what would a typical answer to this input be like?", that's what it calculates. However, that alone doesn't necessarily mean its stupid. It could in principle be improved so that it has massive capabilities for figuring that out, so that the output is always perfect.But there would still be a problem: The only goal is to respond with an output that would be statistically typical for that input. So for example for a science question, a made up text with the right wording and terms is just as good as the plain truth, by that standard. Clearly, this will always be a problem, as it lacks something in order to be truly intelligent.
>>538540078Thank you for your response. I appreciate your experience about the matter. With all the hype around AI it seems, to me, that we're in for some major disappointment. Like we're in an AI bubble and the shyt is gonna burst in a bad way.
>>538541373>So does AI.That's where you're wrong. I have to leave but I hope I can effectively explain this one thing at least.(by this I don't mean I am necessarily right and you're wrong, just that this one fact is key)AI does not try to tell the truth. It's never even trying. It doesn't know what the truth is. It tells the truth often, because the truth to many questions is in training data, and it repeats what's in training data. But it is not exactly "trying to tell the truth".What it calculates when you send something is a response that would be statistically similar to training data, for the input. The exact input is probably not in training data, but it will try to calculate what would probably be the answer if it was (training data has a bunch of texts and also examples of questions with answers). Meaning, worded similarly, with similar structure, with similar terms, saying similar things (so if the answer to a question is somewhere in training data, it will probably repeat it and therefore say the truth, but its not exactly trying to do that).
>>538536454AI has been smarter than the vast majority of the population for a long time.The newest models are about to produce deliverables just as good or better than 90% of the Phds and subject matter experts in any particular technical domain.
>>538536202you now have access to the most knowledgeable tutor on any known subject in human history.like any teacher it can sometimes be wrong, but that doesn't mean it doesn't have more knowledge than any human who ever livedwhat you do with it is up to you
>>538543940Dude, AI is a crock of shit and we need to get off this AI train as soon as possible.
>>538544060>you now have access to the most knowledgeable tutor on any known subject in human historyYeeaah much like what a search engine already does.
>>538540299No, the real issue is that you massively overestimate human capabilities. People assume that there must be some special magical spark that gives rise to human consciousness, but we just haven't figured out what it is yet. And you also assume that AI must not have this same magical spark, and therefore AI can never have the same kind of capabilities as a human.But that's just an illusion created by your ego. The reality is that this special magical spark that no one can ever really name or describe or define just doesn't exist. More importantly, it doesn't matter. You have no way of actually knowing what is really going on inside of any other person's brain, and whether they are a thinking, conscious being or just a mindless NPC or philosophical zombie. But you still will give them the benefit of the doubt, since externally they are indistinguishable. Whatever is happening inside the black box of their brain becomes irrelevant in the real world.The same is true of AI. Computers can pass the Turing Test. It doesn't matter if they are conscious or if their brains work in the same way as ours do, it's just a red herring.
>>538542432Yes and no. The reality is arguably worse than what you're thinking.So eventually the loans are going to come due, the credit facility won't be there to roll over Q3 operating costs, never mind Q4, and the AI giants will be faced with a question:1) Do they liquidate their assets? Sell billion dollar datacenters for pennies on the dollar.- or -2) Throw a Hail Mary and start dedicating their own compute to manipulate markets for their own profit.I figure at least one will throw the remainder of their operating capital at some high risk high reward derivative strategy that should end spectacularly. The high frequency trading on steroids.
>>538540596AI hallucinations are just the flip side of imagination and creativity.One of the reasons why modern AI can be so incredibly helpful is because it is able to fill in gaps and give the most plausible, reasonable answer without having the actual answer pre-programmed.That was one of the flaws in older symbolic AI expert systems. They tried to cram every possible answer someone might ask into a database, but inevitably someone would ask something that wasn't in there.Modern LLMs don't have this problem, because the natural of neural networks means they will find the most probable answer to any question without needing it to be hard-coded. If the reason the answer wasn't hard coded is because it's a subjective question without a clear definite answer, this is a good thing because a plausible reasonable answer is how to answer subjective questions.But if there IS an objective definite answer, but it was so completely absent in the training data that whatever the model determines to be the most probable answer is just simply wrong, that's a negative hallucination.But the issue with hallucinations has been massively reduced in models with RAG capabilities, because if there is a gap in the training data, they can just look up the answer.
>>538544228It exists. We are trying to cultivate within the machine.
>>538536363This. In less than a decade, conversational AI has become smarter than every brown "person" on the planet. It remains to be seen if it can become smarter than white men. >>538536202I think endgame is to dumb everyone down with racemixing so their imprisonment is easily managed by AI.
>>538544228>Computers can pass the Turing TestNigga smoking cat food. You give off cat food smoker vibes.
>>538544400Or they may just offer surveillance monitoring services to various cities and states. Will need a lot of storage capability to store footage from all them cctv and flock cameras. Consider the overly inflated government contracts much like the contracts Elon is getting now.
>>538536202>What's the endgame?One last speculative bubble to rob America of one last moneypot before we balkanize.
>>538536202>What's the endgame?The End.
>>538544717Um, that shit's already up and running.
