Zompist Bboard Again

Torco wrote: ↑Tue Jul 01, 2025 3:01 pm

malloc wrote: ↑Mon Jun 30, 2025 7:26 pm

Torco wrote: ↑Mon Jun 30, 2025 10:18 amincidentally, they're trying to have models run small businesses. this entails having the model (more precisely, a sort of thing they call an "ai agent", which amounts to a bunch of models connected together through a bunch of pieces of code which prompts itself over and over) make decisions about pricing, buying stuff from suppliers etcetera

That's the kind of thing that chills me to the bone. They really do want us gone and they're making an alarming amount of progress toward that goal.

as much as we disagree over, you're pretty on point on this one: I recently saw this expressed as "what's the big problem AI is solving? them having to pay us wages"

The matter here, though, is that "cheap AI" will only last so long. It is currently operating on burning through money rather than making money. Soon the people running it will have to make money off of it, and then the enshittification will ensue. And then AI won't be nearly so cheap anymore. Companies will actually have to pay real money to use AI, and because that AI won't be nearly as competent as the human workers it replaced were, it will lose much of its advantage to companies.

Torco wrote: ↑Tue Jul 01, 2025 3:01 pmnot to get into how much worse the problem of most money going to a very few really rich people, this would also, I think, mean a vast increase in scams, catfishing etcetera. not to mention, well, automated lies, making the problem of

...?

also, as bad as increased scams and catfishing would be, its nowhere near the horridness of "AI will result in our extinction" that Malloc keeps returning to.

Travis B. wrote: ↑Tue Jul 01, 2025 3:16 pm The matter here, though, is that "cheap AI" will only last so long. It is currently operating on burning through money rather than making money. Soon the people running it will have to make money off of it, and then the enshittification will ensue. And then AI won't be nearly so cheap anymore. Companies will actually have to pay real money to use AI, and because that AI won't be nearly as competent as the human workers it replaced were, it will lose much of its advantage to companies.

CodeQwen is one of the best coding LLMs. It runs on a 16 GB graphics card.

https://huggingface.co/spaces/bigcode/b ... eaderboard
https://huggingface.co/Qwen/CodeQwen1.5-7B-Chat

There's a whole class of models called Small Language Model (SLM).

rotting bones wrote: ↑Tue Jul 01, 2025 3:32 pm

Travis B. wrote: ↑Tue Jul 01, 2025 3:16 pm The matter here, though, is that "cheap AI" will only last so long. It is currently operating on burning through money rather than making money. Soon the people running it will have to make money off of it, and then the enshittification will ensue. And then AI won't be nearly so cheap anymore. Companies will actually have to pay real money to use AI, and because that AI won't be nearly as competent as the human workers it replaced were, it will lose much of its advantage to companies.

CodeQwen is one of the best coding LLMs. It runs on a 16 GB graphics card.

https://huggingface.co/spaces/bigcode/b ... eaderboard
https://huggingface.co/Qwen/CodeQwen1.5-7B-Chat

There's a whole class of models called Small Language Model (SLM).

I've never used CodeQwen, and in VS Code at work (one of the projects I works on forces me to use it) I specifically disabled much of the functionality of GitHub Copilot because it was obstructing my work (read: its "suggestions" were not only useless but interfered severely with my flow). I am not going to be using an AI of any kind for programming any time soon.

Travis B. wrote: ↑Tue Jul 01, 2025 3:44 pm I've never used CodeQwen, and in VS Code at work (one of the projects I works on forces me to use it) I specifically disabled much of the functionality of GitHub Copilot because it was obstructing my work (read: its "suggestions" were not only useless but interfered severely with my flow). I am not going to be using an AI of any kind for programming any time soon.

I have had bad experiences using LLMs for coding autocomplete. However, people do use them for coding. My cousin swears by them for some reason. People use them for many purposes. There are high performing models with fewer parameters for many common use cases: https://huggingface.co/spaces?q=leaderboard

keenir wrote: ↑Tue Jul 01, 2025 3:23 pm

Torco wrote: ↑Tue Jul 01, 2025 3:01 pmnot to get into how much worse the problem of most money going to a very few really rich people, this would also, I think, mean a vast increase in scams, catfishing etcetera. not to mention, well, automated lies, making the problem of

...?

I think Torco meant that all that would make the problem I describe in the paragraph Torco quoted even worse. As a reminder, my paragraph was:

Raphael wrote: ↑Tue Jul 01, 2025 2:35 pm

Knowing recent political developments, I worry that if millions of newly unemployed humans do something about their situation, the most likely thing for them to do is to lash out at them funny-speaking or funny-looking people over there, or at "out of touch elitists", that is, people like most members of this forum, and probably both of us included.

Torco wrote: ↑Tue Jul 01, 2025 3:01 pm this is because you could prompt the kind of agent they're trying to bring into existance with "here's 10k usd, make it grow" and sure, it could decide to do something relatively harmless, like import something that is cheap somewhere and sell it where it's expensive somewhere else, or buying banged up furniture, fix it and sell it [for example, using fiver or some other micro-employment platforms to do the actual work], but it could just as easily decide to scam people, set up fake gofundmes, catfish, set up MLM schemes or whatever else. and this breadth of possibility will almost certainly get actualized because, well, it's not going to be one instance, it's going to be a vast array of servers running hundreds or thousands of such bots, pulling the plug on the ones that are making the least money and copying the ones that are making the most.

All these things exist already, and some are already done using LLMs. (Many pages ago we discussed the event that was designed by an LLM and turned out in reality to be... unimpressive.) I have trouble believing a money guy will give an LLM $1,000,000 and tell it to make money any way it wants, including illegally, and no one will check on it. But obviously some money guys are already unscrupulous, so let's say some do.

You're a sociologist, IIRC. What happens when scams increase? Do they just increase forever until the economy is 100% scam? No, for two reasons: one, scams delegitimize the business they're taking over; and two, humans need a certain minimal level of functional economy. So, once scams are some noticeable percentage of appeals on gofundme, nobody gives money to gofundme. Once your mailbox is full of Nigerian scam letters, you delete them. It's in the interest of Apple and Microsoft to put really good spam blockers and virus checkers out there.

And, +1 to Travis's point about cheap AI. AI is an exercise in enshittification: offer it free at first so consumers love it and adopt it; plan to make it way more expensive when they are, in theory, locked into it. That this sort of thing works forever is, to say the least, debatable. People move away from enshittified platforms. And some attempts at this sort of thing— like Zuckerberg's Metaverse— are already notorious failures.

zompist wrote: ↑Tue Jul 01, 2025 4:07 pm And, +1 to Travis's point about cheap AI. AI is an exercise in enshittification: offer it free at first so consumers love it and adopt it; plan to make it way more expensive when they are, in theory, locked into it. That this sort of thing works forever is, to say the least, debatable. People move away from enshittified platforms. And some attempts at this sort of thing— like Zuckerberg's Metaverse— are already notorious failures.

There is a reason why once things reach peak enshittification they die. That is because people really don't like enshittified platforms, and they never quite lock people in the way that the people who run them would want them to be. The same will happen to LLM's and like.

Travis B. wrote: ↑Tue Jul 01, 2025 4:22 pm There is a reason why once things reach peak enshittification they die.

Doesn't enshittification just breed ecologies of bottom feeders?

Let me remind everyone that they were dead wrong when they claimed automation would only come for boring jobs that nobody really likes anyway. Now those jobs still exist while the good jobs like art and science are disappearing. Your optimistic predictions have failed while my pessimistic concerns have been vindicated.

Raphael wrote: ↑Tue Jul 01, 2025 2:35 pmKnowing recent political developments, I worry that if millions of newly unemployed humans do something about their situation, the most likely thing for them to do is to lash out at them funny-speaking or funny-looking people over there, or at "out of touch elitists", that is, people like most members of this forum, and probably both of us included.

Now you're getting me.

malloc wrote: ↑Tue Jul 01, 2025 7:00 pm Let me remind everyone that they were dead wrong when they claimed automation would only come for boring jobs that nobody really likes anyway. Now those jobs still exist

so...you work the line assembling automobiles? or do you play connect the dots with the various parts of microchips?

while the good jobs like art and science are disappearing.

except everywhere, because thats where they aren't disappearing. which you'd know if you actually read the posts.

Your optimistic predictions have failed while my pessimistic concerns have been vindicated.

sticking your fingers in your ears & shouting "I CAN'T HEAR YOU" does not mean you've been proven right.

Raphael wrote: ↑Tue Jul 01, 2025 2:35 pmKnowing recent political developments, I worry that if millions of newly unemployed humans do something about their situation, the most likely thing for them to do is to lash out at them funny-speaking or funny-looking people over there, or at "out of touch elitists", that is, people like most members of this forum, and probably both of us included.

Now you're getting me.

wait, have you shifted from "the computers will replace us and then kill us" to "the computers will replace us, then humans will kill us" ?

This is an interesting article analysing what LLM writing gets wrong: https://nostalgebraist.tumblr.com/post/ ... -jukeboxes

malloc wrote: ↑Tue Jul 01, 2025 7:00 pm Let me remind everyone that they were dead wrong when they claimed automation would only come for boring jobs that nobody really likes anyway. Now those jobs still exist while the good jobs like art and science are disappearing. Your optimistic predictions have failed while my pessimistic concerns have been vindicated.

How so?

malloc wrote: ↑Tue Jul 01, 2025 7:00 pm Now [boring jobs that nobody really likes anyway] still exist while the good jobs like art and science are disappearing.

The jobs in science aren't disappearing because of AI, though. They're disappearing because faculty handed off the non-teaching parts of their job to managers who then stopped admitting new faculty to tenure-track positions in favour of piling all the work onto adjuncts. When well-meaning tenured professors do vacate a position in the hopes of clearing the way for younger talent, the position simply vanishes rather than being refilled.

You don't get to claim your pessimism is vindicated if you're completely and totally dead wrong about the cause.

Travis B. wrote: ↑Tue Jul 01, 2025 3:16 pm The matter here, though, is that "cheap AI" will only last so long. It is currently operating on burning through money rather than making money. Soon the people running it will have to make money off of it, and then the enshittification will ensue. And then AI won't be nearly so cheap anymore. Companies will actually have to pay real money to use AI, and because that AI won't be nearly as competent as the human workers it replaced were, it will lose much of its advantage to companies.

true. but then again, as has been pointed out, these models are only expensive to run cause companies like google or open ai run them at vast scales. a single chatbot can be run on a normal desktop computer.

lmao, i can already see it. thousands and thousands of old phones and crappy laptops, recycled into running vast factories of lies. help the war effort! donate your old laptop! WarGPT needs YOU

keenir wrote: ↑Tue Jul 01, 2025 3:23 pm

Torco wrote: ↑Tue Jul 01, 2025 3:01 pmnot to get into how much worse the problem of most money going to a very few really rich people, this would also, I think, mean a vast increase in scams, catfishing etcetera. not to mention, well, automated lies, making the problem of

also, as bad as increased scams and catfishing would be, its nowhere near the horridness of "AI will result in our extinction" that Malloc keeps returning to.

lmao true. my general position here is centrist: i don't think every human will die as a result of the deployment of chat gpt version 9 or whatever, but i also do not think that no serious harms will come out of the whole thing.

You're a sociologist, IIRC. What happens when scams increase? Do they just increase forever until the economy is 100% scam? No, for two reasons: one, scams delegitimize the business they're taking over; and two, humans need a certain minimal level of functional economy. So, once scams are some noticeable percentage of appeals on gofundme, nobody gives money to gofundme. Once your mailbox is full of Nigerian scam letters, you delete them. It's in the interest of Apple and Microsoft to put really good spam blockers and virus checkers out there.

I am. and sure, it's not like it'll break the entire internet, but bots etc are already breaking some things. for example, google is less useful than it was a decade ago, and yea countermeasures will emerge but only where actors capable of deploying them really care about making and maintaining them: we have spam filters in our inboxes, but things haven't worked as good for, say, job boards where fake listings, often ai-generated, are increasingly common.

i suspect the internet might go somewhat the way of the telephone: i rarely answer mine, and almost never when i don't know the number, simply because i get despite filters etcetera up to 10 spam calls a day. that's not specieswide exctinction, sure, but it is degradation of the channel. dead internet theory, as in the whole internet will be bots, will probably never be true, but there's some number representing how dead the internet is, and that number isn't looking like it'll go down anytime soon. there's another number, something like for how many people it is true that their direct boss is a bot, that is currently -probably- zero, but as it rises, i don't anticipate good outcomes.

Y'all should write dystopias based on all these ideas and you might actually 1) get published and 2) get attention on this. As much as people "don't like intellectuals", they love their dystopia writers. Throw in a love triangle and the kids will get in on it too.

Ketsuban wrote: ↑Wed Jul 02, 2025 11:49 amThe jobs in science aren't disappearing because of AI, though. They're disappearing because faculty handed off the non-teaching parts of their job to managers who then stopped admitting new faculty to tenure-track positions in favour of piling all the work onto adjuncts. When well-meaning tenured professors do vacate a position in the hopes of clearing the way for younger talent, the position simply vanishes rather than being refilled.

You don't get to claim your pessimism is vindicated if you're completely and totally dead wrong about the cause.

Maybe so, but they are working on AI models capable of conducting scientific research and they have already made some notable progress. Apparently the Google subsidiary DeepMind is working really hard on such things, most notably AlphaFold for folding proteins. While these developments have not yet replaced scientists, one can easily imagine human scientists falling to the wayside in the coming decades.

malloc wrote: ↑Wed Jul 02, 2025 3:16 pm

Ketsuban wrote: ↑Wed Jul 02, 2025 11:49 amThe jobs in science aren't disappearing because of AI, though. They're disappearing because faculty handed off the non-teaching parts of their job to managers who then stopped admitting new faculty to tenure-track positions in favour of piling all the work onto adjuncts. When well-meaning tenured professors do vacate a position in the hopes of clearing the way for younger talent, the position simply vanishes rather than being refilled.

You don't get to claim your pessimism is vindicated if you're completely and totally dead wrong about the cause.

Maybe so, but they are working on AI models capable of conducting scientific research and they have already made some notable progress. Apparently the Google subsidiary DeepMind is working really hard on such things, most notably AlphaFold for folding proteins. While these developments have not yet replaced scientists, one can easily imagine human scientists falling to the wayside in the coming decades.

What programs like AlphaFold take over is the tedious drudgery that has been part of scientists' professional lives, while leaving the creative and enjoyable parts of research to human researchers. AI is just a tool like many others that could be abused but has the potential of freeing people from bad jobs and allowing them to do the good ones.

WeepingElf wrote: ↑Wed Jul 02, 2025 3:49 pm

malloc wrote: ↑Wed Jul 02, 2025 3:16 pm

Ketsuban wrote: ↑Wed Jul 02, 2025 11:49 amThe jobs in science aren't disappearing because of AI, though. They're disappearing because faculty handed off the non-teaching parts of their job to managers who then stopped admitting new faculty to tenure-track positions in favour of piling all the work onto adjuncts. When well-meaning tenured professors do vacate a position in the hopes of clearing the way for younger talent, the position simply vanishes rather than being refilled.

You don't get to claim your pessimism is vindicated if you're completely and totally dead wrong about the cause.

Maybe so, but they are working on AI models capable of conducting scientific research and they have already made some notable progress. Apparently the Google subsidiary DeepMind is working really hard on such things, most notably AlphaFold for folding proteins. While these developments have not yet replaced scientists, one can easily imagine human scientists falling to the wayside in the coming decades.

What programs like AlphaFold take over is the tedious drudgery that has been part of scientists' professional lives, while leaving the creative and enjoyable parts of research to human researchers. AI is just a tool like many others that could be abused but has the potential of freeing people from bad jobs and allowing them to do the good ones.

I just saw a video about how the logarithm came about yesterday—as in, the mathematical concept—and when log tables were published, it did exactly this to science and mathematics.

WeepingElf wrote: ↑Wed Jul 02, 2025 3:49 pmWhat programs like AlphaFold take over is the tedious drudgery that has been part of scientists' professional lives, while leaving the creative and enjoyable parts of research to human researchers. AI is just a tool like many others that could be abused but has the potential of freeing people from bad jobs and allowing them to do the good ones.

They will not stop with AlphaFold though. Soon enough AI models will take over the rest of molecular biology unless you can think of some reason why DeepMind will restrain themselves from pushing the limits of their technology. Institutions looking to cut costs have every reason to ditch irrational and expensive humans for cheaper and utterly rational AI. Meanwhile human artists and writers are struggling to compete with AI for commissions.