No.3666
practical usage: stupid people like it and would be willing to pay like $15/month for it and/or look at lots of ads while using it
No.3671
>>3666I guess, but even still 15 billion wouldn't be that much compared to how much they spend.
No.3674
>>3665There is no commercial use-case that can justify the costs even remotely. It's a bubble, there is nothing underneath. All the useful parts will be stripped out and implemented elsewhere, as it has been happening for most of machine learning as a field.
Either things will go well and the bubble will burst in a non-castrophic fashion or people will start overrelying on this technology and things will just break down over time.
I am very much hoping for the former, but it is clear, we are moving towards the latter.
>The only thing I've seen some actual promise in have been the more singular focused LLMs trained on specific tasks like speeding up diagnosis and assistance in the health field. that's just a search engine, you're not making anyone more productive and you certainly aren't improving outcomes with this.
>I guess the idea for companies right now is to take the usual cheap approach of bleeding money for a while until they can assert themselves in some way as a critical part of a workflow/lifestyle and then massively up the cost when people can't are too stuck with them to simply drop them.That sure is what is happening, the problem is that they don't have a consumer product that can be commercialized, because the main appeal of generative AI is that it doesn't cost anything.
No.3675
>>3674>people will start overrelying on this technology and things will just break down over time.I'm hoping for this one!
>>3665All the usefulness is mostly restricted to the personal. An employee will use an "AI" for a task well only when he'll personally know how to complete it and what to prompt towards that. Forcing reliance on "AI" will gradually erase employees' expertise and turn people into supervisors and micromanagers that know little more than subpar prompting. It'd be a first world phenomenon that would make it look sensible to import even incompetent thirdie specialists that weren't raised with addiction to the "AI" tatas forced into their mouth.
>>3674>that's just a search engine*expert system
No.3677
>>3675>*expert systemstill a search engine
If you want to be even more pedantic, we don't even need computers for this. There's something called a "reference book".
No.3678
>>3676We can't even take over the maggot infestation crawling out of our months old trash that we refuse to take out during the day and are too lazy to take out during the night. And by we I mean myself.
No.3679
>>3675>It'd be a first world phenomenon that would make it look sensible to import even incompetent thirdie specialistsas someone who unfortunately works with "third world specialists" for a living, framing it like this suddenly makes me understand why tech is so enamored with AI as a concept. Because they're already "using" "A(ctual) I(ndians)" every day. It's not that big of a step down from an Indian scam PHD.
No.3680
It's funny, early on a bunch of people at my company were pushing to hop on this AI train and start building something so we'd have the infrastructure and could keep up with the future, but nothing ever got done because nobody was able to create a proposal on how it would actually make us any money. And now any use of AI tools is banned.
I really don't see much they can be used for besides those customer service chatbots everyone hates, but the benefit of making your kuso return service a bit better probably isn't that great when rudimentary bots and Indians are proven to get the job done even if they piss everyone off. You're probably not going to get that "good customer service" cred unless you have real people available.
I guess the main driver then is an expectation that the cost will plummet to commercially-viable levels and having the infrastructure set up before then will give you a head start, but that's a pretty big gamble if you're not part of the big tech/financial industries that benefit whether these techbro fads succeed or fail. There's probably some marketing/government money to be made with the ability to make the internet a completely unusable pile of shit too.
No.3683
I know very little about AI things, but the most practical use I've heard about is for legal professionals as an advanced search engine for legal cases. But simultaneously, the context in which this use was mentioned was that AI was creating hallucinations; making up cases that never happened.
No.3684
>>3683The most practical case you've heard about is a complete idiot almost getting disbared and having his entire life destroyed because he was lazy and didn't understand the shortcut being sold to him?
No.3685
>>3683One thing you need consider that will very quickly make you understand the issues with the state of "AI" today and likely until a major paradigm shift:
Which one is more dangerous?
¥ A system that gives the correct answer 50% of the time.¥ A system that gives the correct answer 90% of the time?Paradoxically, the system that is right far more often is much more problematic than the system that is effectively as good as coin flip.
No.3687
>>3674>the problem is that they don't have a consumer product that can be commercialized, because the main appeal of generative AI is that it doesn't cost anything.While this is true from a consumer perspective, it's very wrong in the business world. As long as AI is cheaper than human labor (which is already the case), then AI will be used as a venue to replace human labor which increases company profit. This is where the big money is, any why 99% of AI development funding are for LLMs with instruction following and reasoning abilities, rather than the image generation or role playing models for the consumer entertainment market.
If AI does an equal job as juniors then there is no need to hire and train juniors anymore. Just need to hire and train less number of seniors who know how to do AI prompting. Same productivity with less labor spending.
For jobs requiring specialized knowledge, as long as AI increases the efficiency of existing workers then it's already achieved its goal. Now they can hire less for the same productivity.
Also once AI agents are integrated into the communication workflow (slack/teams etc) then it's the end for middle managers.
No.3689
I use it to autocomplete chunks of software that I just copy paste otherwise.
But copy pasting code is bad "by the book" programming so it basically is just encoraging me to be a bad/lazy programmer
No.3690
>>3689>But copy pasting code is bad "by the book" programming so it basically is just encoraging me to be a bad/lazy programmerFor me it's the opposite, when the AI fucks up something in the code I'm forced to check the language's documentation and I learn something, I feed it back to the AI and it learns something. The AI is the best teacher I've ever had, I went from zero programming knowledge to at least making shit that works.
No.3691
>>3690yeah, well it does help pick up things you don't know quickly because it has a lot of documentation in it, but when you actually build a complicated app it's a bit of a finicky mess
No.3692
>>3665The only use-case I personally have experienced/worked with is using AI for accessibility, like image classification and describing scenes for blind folk,
stuff like real-time vision assistants to help people navigate the world more independently.
Haven't explored generative AI enough, only LLM's for writing personal hyperfixation smut.
Sadly all the AI seen around now just feel like toys or marketing gimmicks, At least accessibility has practical value.
No.3693
>>3687>If AI does an equal job as juniors This is the magic fantasy dream. But is never gonna happen. Sure, AI might replace a scam indian, but these positions already only go to those people, because they can be exploited and even then, having the indian is likely far more valuable.
>as long as AI increases the efficiency of existing workers then it's already achieved its goal. Now they can hire less for the same productivity.This doesn't work either, unfortunately. Because AI is inherently time-consuming to use, it is not a net positive in efficiency.
>once AI agents are integrated into the communication workflow (slack/teams etc) then it's the end for middle managers.The purpose of a middle manager is not handling the communication workflow.
I think your understanding of the business world and LLMs as a technology is flawed. You are reasoning from correct first principles, but you don't really see the big picture.
No.3694
>>3693>you don't really see the big pictureLook into a mirror and ask yourself this question, because you certainly don't see any big picture at all.
It's the same argument being used forever, except you would have said that AI is too stupid to replace indian scammers 2 years ago and AI is too stupid to produce coherent sentence 4 years ago, etc.
AI is constantly improving but your reasoning is still rooted in the assumption that AI can't be improved from the current state.
No.3695
>>3694My assumption is that progress will plateau before it is useable. This assumption is based on the fact that is in fact reaching a plateau.
No.3696
>>3695>This assumption is based on the fact that is in fact reaching a plateau.Again, same argument being used forever.
No.3697
>>3677Excuse me Anon-san, did you perhaps mean "cave wall scribbles"?
No.3698
>>3696you say it's not doing that, I say it is, which one is it?
No.3700
>>3699Protein folding is very impressive, but once you look into how it's done, things start getting a little less "AI" and more "sophisticated data processing and machine learning applications with a pile of protein science". It's not really "AI" in any meaningful way, but just leveraging existing data in intelligent ways.
It also helps that protein structures are relatively easy to verify. Sure it will take a student a couple months to solve a protein structure using traditional methods, but it is not "difficult", just tedious and some just don't work.
No.3701
>>3698If you actually follows LLM development you should've already known the answer.
Everything since the initial GPT has been incremental improvement and it's still happening right now. The huge amount of LLM ability changes over the last 2 years are all incremental improvements but you made it sounds like it has already "hit the plateau" years ago because nothing "revolutionary" is happening.
No.3702
>>3699>Talks about AI becoming real in my life time has given me hope for the future and personally I really hope it keeps improving and they solve the issues around it and it boggles my mind that people are actively against it for no apparent reasonThe "AI" we get shoved down our throats today is not the AI we dreamed of in sci-fi and its applications so far have done nothing good for me and a whole lot of bad. This path isn't going to take you where the marketers act like it will take you and the more resources we waste on this the fewer we have to put towards more productive ventures.
No.3703
>>3701>If you actually follows LLM development you should've already known the answer.and the answer is "it's reaching a plateau", because no amount of "incremental improvement" can change anything anymore.
No.3704
My PhD supervisor keeps trying to make me change my topic something AI related and all my research assistant work has been on AI. I'm losing my mind with this stuff. He's utterly convinced its the next big thing and some huge paradigm shift.
Having done a data science course that focuses on ML, I really do not understand what he sees in it. He's completely bought into the hype despite other faculty members (that actually teach the course on it) thinking it's completely overblown. Apologies for the blogpost.
A lot of the policy focus is on generative AI but it seems to have very limited applications and, as you point out, probably isn't economically viable. It's going to become a moot point.
Outside of porn people generally seem to prefer stuff that's made by humans. If you're using generative AI instead of a person, you just have to trust the model isn't hallucinating and if you need to hire AI jannies to proof and clean up the outputs, you probably should have just hired someone to get it right on the first attempt.
AI seems to have some promise in research settings (like materials science) but those are uncontroversial and get very little attention. My instinct is that the future of AI is there, but it isn't "sexy" enough a topic for people to pontificate over. I mostly share the sentiments of
>>3674.
>>3683Most lawyers are too lazy to learn how to use the advanced search on Westlaw or Lexis.
No.3705
>>3700I'm more annoyed that all that time and effort people spent on playing FoldIt has been rendered useless
No.3706
>>3705not at all, those structures that were found using these decentralized methods were used to build these bigger ANN-based struture predictors
No.3707
>>3703>it's reaching a plateauA phrase as meaningful as "tomorrow is when the world ends"
>anymoreIt happened before, it's happening right now, and it will still happen in the future.
No.3708
>>3702>The "AI" we get shoved down our throats today is not the AI we dreamed of in sci-fi and its applications so far have done nothing good for me and a whole lot of bad. Money don't care about your feelings.
No.3709
>>3707>A phrase as meaningful as "tomorrow is when the world ends"if you want to be pedantic, it's "tomorrow is when the world starts to end"
>It happened before, it's happening right now, and it will still happen in the future.So tell me, what is "it"?
As far as I am aware, nothing beyond a fundamental paradigm shift can make this technology useful.
No.3710
>>3708money don't care about dumping waste in the river either.....
No.3711
>>3709>So tell me, what is "it"?Incremental improvement. It has worked for the past several years, and you're the one who's arguing that NOW is the time that it stops working.
>As far as I am aware, nothing beyond a fundamental paradigm shift can make this technology useful.It's already useful as is. They know more and make less mistakes than the average human. Whenever it makes mistakes it's corrected by those who know more. Just like seniors to juniors.
No.3712
a junior who will never improve. Such a valuable asset
No.3714
>>3712If you haven't noticed, career path is dead now. Everyone hires senior positions from outside of companies nowadays. Those few who can improve themselves to the senior level will do just fine. Hiring juniors is a waste of time and money.
No.3715
>>3711>Incremental improvementYou are dodging the important questions, starting with the first one:
¥ Will incremental improvement work forever?¥ How does incremental improvement happen on a mechanical level? What do you need to do to get an incremental improvement?¥ Is the way we do things right now in this field the best way to achieve results?If your answer to the last one is anything but "yes, it definitely is", then incremental improvement is inevitably failing. There is no possible way that "incremental improvement" will ever lead to anywhere useable. We are still far, far away from having an application for this technology. It is simply not good enough and I do not believe that any amount of "incremental improvement" can fix that.
If you are seeking a proper answer, consider computer chess as a model of this subject: Historically, there were two main approaches to computer chess. Big and stupid engines, that used most of their memory to memorize board states and calculate branches. And smaller, but smarter engines, which could dig far deeper into the branching function, by utilizing sophisticated algorithms.
The former architecture was far, far more popular during the early days, but it is all but gone today, because it turns out it sucks and it's far too expensive.
>They know more and make less mistakes than the average human. They do not and if you say that they actually do, it is clear that whatever level you are considering is too low to be worth taking seriously. For many roles, which in theory can be replaced by AI, there are countless functions that cannot be performed by a computer, no matter how sophisticated.
>>3714>Hiring juniors is a waste of time and money.no it isn't because seniors are too expensive to waste on closely supervising gruntwork. Juniors come with a level of autonomy that so-called "AI" does not have
No.3716
>>3706but now there's no point playing FoldIt, like how reCAPTCHA isn't used for digitizing books anymore
No.3717
>>3716well, if you are a researcher, you can still play FoldIt all day while you wait the weeks it takes for your protein to crystallize.
No.3718
>>3715>If your answer to the last one is anything but "yes, it definitely is", then incremental improvement is inevitably failing. I'm not falling for your loaded question:
https://en.wikipedia.org/wiki/Nirvana_fallacyI will fix this by adding the requirement that the goal is to achieve a better than current result in a reasonable amount of time like 1 or 2 years. Then the answer is still "yes, it definitely is".
>They do not and if you say that they actually do, it is clear that whatever level you are considering is too low to be worth taking seriously. You are severely overestimating the capabilities of average human. Not everyone is 130+IQ imageboard users.
>There is no possible way that "incremental improvement" will ever lead to anywhere useable. We are still far, far away from having an application for this technology. It is simply not good enough and I do not believe that any amount of "incremental improvement" can fix that.>there are countless functions that cannot be performed by a computer, no matter how sophisticatedHow do you know you're right THIS time? Predictions like this has been proven wrong over and over before.
No.3719
>>3718>I'm not falling for your loaded questionThat's very nice, but you didn't answer any of the questions and they are very reasonable things to ask, I think. How about you go answer those, before you start getting defensive.
>a better than current result in a reasonable amount of time like 1 or 2 years.That's very weasely, you know that. "better than current"? How much better? What does better mean?
Care to elabote a little bit?
>You are severely overestimating the capabilities of average human. You are severely underestimating the requirements of most work, even simple things.
>How do you know you're right THIS time? Because it is clear that there are things only people can do that computers cannot do (at least within any realistic timeframe and without a paradigm shift). Something you yourself tacitly admitted yourself: After all, computers fulfil junior roles, they cannot perform senior roles? Do you know what the main difference between lower and higher ranked roles is? Hint: It's not expertise. What do you do more of the more important and better paid you are?
It's the same thing.
>Predictions like this has been proven wrong over and over before.They have never been wrong, because every time such a prediction made, there was a bubble that collapsed. People talk about the internet in such terms, but do they not remember?
Also, please stop ignoring the parts in my post you find uncomfortable to address. It makes you look pathetic.
No.3720
>>3719>but you didn't answer any of the questions and they are very reasonable things to askThere is no need to answer them, because you know we both know the answer to these questions. The only reason you bring them up is to dismiss valid positions with logical fallacy.
>That's very weasely, you know that. Same goes for your loaded questions.
>You are severely underestimating the requirements of most work, even simple things.You're living in the bubble of your field. Most work involve only rudimentary data processing.
>Because it is clear that there are things only people can do that computers cannot do (at least within any realistic timeframe and without a paradigm shift). And there are things that computers can also do, and I'm very clear that I'm talking about these things when discussing about AI replacing human labor. It's you who's holding an absolute position that AI cannot replace ANY human labor.
>They have never been wrong, because every time such a prediction made, there was a bubble that collapsed.Bubbles are an economical construct and does not imply that the technological improvements will stop. AI replacing human labor is undoubtedly economically sound, unlike previous bubbles. I am claiming something like 10-20% workforce replaced by the current AI tech. You are the one who's claiming that it will only achieve 0% because it can't achieve 100% with the current tech.
>Also, please stop ignoring the parts in my post you find uncomfortable to address. It makes you look pathetic.There is no need to waste time addressing the obvious.
No.3721
>>3720Are you Indian, or why do you pray to AI like this, while constantly stumbling over your grammar?
>because you know we both know the answer to these questionsDo we? Then refresh my memory, please. Indulge me.
>The only reason you bring them up is to dismiss valid positions with logical fallacy.What valid position would that be?
I don't think any of these questions are particular fallacious, I think they are very straightforward and simple to answer, if you know what you are talking about. You said yourself that you know the answer. Why are you afraid of providing it?
>You're living in the bubble of your field.Hardly so. Because the claim I am making is arguably the opposite. Most work requires a lot of things that is explicitly NOT data processing, but that involves interacting with the real, physical world or other people. And there is a lot of money on the line in many of these interactions.
>I'm very clear that I'm talking about these things when discussing about AI replacing human labor. Well, you just said that middle managers will become obselete, which is an absurd claim. Because one of the things AI absolutely cannot do is manage people. And I will say this like this. It is impossible for an LLM to handle one or multiple reports.
>It's you who's holding an absolute position that AI cannot replace ANY human labor. Define "replace". I never used that word.
>Bubbles are an economical construct and does not imply that the technological improvements will stop. They actually very much do imply that. Because if there is no money being spent on something, the rate of improvement goes down. If you don't pay people to do research, they won't do research, because people need money to live.
>AI replacing human labor is undoubtedly economically soundIs it really? Even in Fantasyland, is it economically sound? I have my doubts.
>I am claiming something like 10-20% workforce replaced by the current AI techYes and I am saying that that is impossible, because you have never had a real job or done any real work, let alone managed people.
>There is no need to waste time addressing the obvious.You really don't see how the chess computer model is relevant, do you?
No.3723
>>3721>Are you Indian, or why do you pray to AI like this, while constantly stumbling over your grammar?Whatever makes you feel superior.
>What valid position would that be?My position is consistently implied in all of my posts. Read them again. Stop asking these totally redundant questions trying to make your opponent lose energy.
>I think they are very straightforward and simple to answer, if you know what you are talking about. You said yourself that you know the answer. Why are you afraid of providing it?I already did:
https://en.wikipedia.org/wiki/Nirvana_fallacyYou asked for the absolute for <things>. I answered there isn't such thing.
>Most work requires a lot of things that is explicitly NOT data processing, but that involves interacting with the real, physical world or other people. And there is a lot of money on the line in many of these interactions.Come on. We both know these are not the types of job under discussion. Have I ever mentioned them being replaced by AI in this thread? The topic under this discussion is LLMs, nothing else.
>Because one of the things AI absolutely cannot do is manage people. And I will say this like this. It is impossible for an LLM to handle one or multiple reports.This is an absurd claim. The difficulty lies elsewhere and has nothing to do with the reasoning and data processing ability of LLMs, which has already reached enough capability.
>Define "replace". I never used that word.I used that word. You rejected my claim relevant to that word, which makes an implicit counter claim. You thus have addressed that word no matter what you say otherwise.
>Because if there is no money being spent on something, the rate of improvement goes down.This does not reject my claim. Did the internet stop developing or become less popular after the dotcom bust?
>Is it really? Even in Fantasyland, is it economically sound? I have my doubts.>Yes and I am saying that that is impossible, because you have never had a real job or done any real work, let alone managed people.Not an argument because you didn't say anything of substance. And yes, I have done both and have already seen the level of incompetency people can achieve. They can't do their own research and need 100% handholding even after being given a general direction. AI do better than this.
>You really don't see how the chess computer model is relevant, do you?They aren't relevant, because there's no money in it so companies aren't interested in it. LLMs are different.
No.3725
>>3723>You asked for the absolute for <things>. ¥ How does incremental improvement happen on a mechanical level? What do you need to do to get an incremental improvement?This certainly doesn't ask for an absolute. I'd like for you to tell me how it works.
If you are running from this question so desperately, I have to wonder if the answer to that question does not also presuppose the answer of others.
>We both know these are not the types of job under discussion. Provide a list of jobs and then we can about that if you'd like.
>This is an absurd claim.It really isn't. What gets you the big bucks today is handling people. That's what senior roles are all about.
>The difficulty lies elsewhere and has nothing to do with the reasoning and data processing ability of LLMs, which has already reached enough capability. It's almost like that part isn't actually what most jobs are about. LLMs do not have data processing capabilities to handle reports. Maybe you are unfamiliar with the term "report" and you are confused? But if you are, why are you speaking about jobs, if you don't know this term?
>This does not reject my claim. Yeah it does.
>Did the internet stop developing or become less popular after the dotcom bust?You mean all those "thing dotcom" websites that were supposed to bring in loads of value for shareholders? Yes, nearly all of those died out.
>And yes, I have done both and have already seen the level of incompetency people can achieve. And yet, you don't know what a "report" is.
>AI do better than this. No, it cannot.
>LLMs are different.There also is no money in LLMs, the amount you'd need to make is way too much at this point.
No.3727
>>3726
yes I always polite sage, because I don't want to shit up the top of /all/ with this garbage argument.
No.3728
>>3712¥ never learns anything¥ prevents the hiring of any new hands¥ no one will ever gain the skills needed to replace youIt's great if you're not in management and have 10 years of experience in whatever you want to do and just want to take it easy doing the same thing until you're 75.
No.3738
>>3725>This certainly doesn't ask for an absolute. I'd like for you to tell me how it works.Look into the technical reports of those LLM models. You will find these small improvements one by one accumulating into significant changes. But it's obvious that you aren't willing to because of your prejudice.
>Provide a list of jobs and then we can about that if you'd like.This is consistently implied in all of my posts. Read them again. Stop asking these totally redundant questions trying to make your opponent lose energy.
>What gets you the big bucks today is handling people. That's what senior roles are all about.But you said it was all about handling reports which I pointed out that LLMs can already do. Stop moving goalpost.
>LLMs do not have data processing capabilities to handle reports. They do, and trading bots are already integrating them into their workflows. This is not where the difficulty is. The difficulty is about integration of how to connect LLMs to the reports and persons. LLM's capability by itself isn't a problem. Read my original post again. You always make these baseless claims and never explain why.
>Yeah it does.Is "the rate of improvement goes down" (your word) the opposite of "does not imply that the technological improvements will stop" (my word)? No, because slowing does not imply stopping or going backwards. This does not reject my claim. You aren't into logic, as if it's not apparent already.
>You mean all those "thing dotcom" websites that were supposed to bring in loads of value for shareholders? Yes, nearly all of those died out.No. And those that survived are those which do bring value. Like LLMs replacing human labor brings the immediate value of cost reduction.
>And yet, you don't know what a "report" is.Then tell me what it is, Mr. iknoweverything. It's my turn to use your tactic.
>No, it cannot.Yes it can.
>There also is no money in LLMs, the amount you'd need to make is way too much at this point.They cost less than human labor. This is fact. This is where the money is. Your babbling changes nothing.
No.3739
>>3728Put the grunt works into temporary, fast rotating interns. They can be paid less with zero benefits and anyone worth hiring should be able to get to senior level in 2 months.
No.3740
>>3738A report is someone who falls under you in the management hierarchy, you drooling idiot. Literally anyone who has ever worked an office job knows this.
No.3741
>>3740And communication integration solves exactly this problem. It changes nothing about the validity of your argument, smartass.
No.3743
>>3738>You will find these small improvements one by one accumulating into significant changesSo what are they?
>This is consistently implied in all of my posts. it is implied, but I want you to be explicit, that is why I am asking.
>Like LLMs replacing human labor brings the immediate value of cost reduction. But the question is whether this cost reduction remains even in the short term. I am arguing no, because it's bad.
>Then tell me what it is>>3740>Yes it can. no it cannot
>They cost less than human labor. As long as they don't work, which they don't, they cost more.
>>3741You don't know what a manager's job is, do you?
No.3749
>>3742I'm training a model on my worst shitposts as we speak. Prepare. You're anus.