I have a whole host of concerns with LLMs, and I'm mostly of the same mind as this person.
-
RE: https://hachyderm.io/@inthehands/116619120828325229
I have a whole host of concerns with LLMs, and I'm mostly of the same mind as this person. Resource-wise, its a massive waste. Ethically, the companies producing the datasets said that intellectual property theft was fine, so long as they were the ones doing it.
but my biggest concern is that, as a cybersecurity professional, for a lot of the "vibe coding" models, they are pulling in data from github, reddit, and stack exchange over the past decade or so, maybe even longer than that.
While all of those sources over the years have been somewhat nice (if a little bit elitist) for learning programming concepts, there are A LOT of incomplete and incorrect answers out there. Or things that work on paper, but we then discovered through YEARS of CVEs that they aren't the most secure or best method for doing the thing(tm).
LLMs make no distinction about that. The training set, is the training set. There's no context that says this is "technically correct, but if you do this in prod, that is really fucking dumb idea." Even so, companies fucking RAN WITH IT. Like it was THE golden calf. Sacrificed their workforces, sacrificed lives, bought out ALL of the RAM, and ALL of the SSDs, only for the models to be wholly inadequate, or in the case of mythos, and the other new models, a nothingburger.
I am waiting with bated breath for the rug pull when these companies stop subsidizing the cost of compute, so that all of the assholes who dropped their workforce will have to eat massive piles of shit.
I've said in the past that we missed our chance for the tech industry to unionize in the 90s. The second best time would be when the shit hits the fan on LLM pricing.
-
RE: https://hachyderm.io/@inthehands/116619120828325229
I have a whole host of concerns with LLMs, and I'm mostly of the same mind as this person. Resource-wise, its a massive waste. Ethically, the companies producing the datasets said that intellectual property theft was fine, so long as they were the ones doing it.
but my biggest concern is that, as a cybersecurity professional, for a lot of the "vibe coding" models, they are pulling in data from github, reddit, and stack exchange over the past decade or so, maybe even longer than that.
While all of those sources over the years have been somewhat nice (if a little bit elitist) for learning programming concepts, there are A LOT of incomplete and incorrect answers out there. Or things that work on paper, but we then discovered through YEARS of CVEs that they aren't the most secure or best method for doing the thing(tm).
LLMs make no distinction about that. The training set, is the training set. There's no context that says this is "technically correct, but if you do this in prod, that is really fucking dumb idea." Even so, companies fucking RAN WITH IT. Like it was THE golden calf. Sacrificed their workforces, sacrificed lives, bought out ALL of the RAM, and ALL of the SSDs, only for the models to be wholly inadequate, or in the case of mythos, and the other new models, a nothingburger.
I am waiting with bated breath for the rug pull when these companies stop subsidizing the cost of compute, so that all of the assholes who dropped their workforce will have to eat massive piles of shit.
I've said in the past that we missed our chance for the tech industry to unionize in the 90s. The second best time would be when the shit hits the fan on LLM pricing.
@da_667 i cant wait to dance on some ashes
-
@da_667 i cant wait to dance on some ashes
@0x00string amen, brother.
-
RE: https://hachyderm.io/@inthehands/116619120828325229
I have a whole host of concerns with LLMs, and I'm mostly of the same mind as this person. Resource-wise, its a massive waste. Ethically, the companies producing the datasets said that intellectual property theft was fine, so long as they were the ones doing it.
but my biggest concern is that, as a cybersecurity professional, for a lot of the "vibe coding" models, they are pulling in data from github, reddit, and stack exchange over the past decade or so, maybe even longer than that.
While all of those sources over the years have been somewhat nice (if a little bit elitist) for learning programming concepts, there are A LOT of incomplete and incorrect answers out there. Or things that work on paper, but we then discovered through YEARS of CVEs that they aren't the most secure or best method for doing the thing(tm).
LLMs make no distinction about that. The training set, is the training set. There's no context that says this is "technically correct, but if you do this in prod, that is really fucking dumb idea." Even so, companies fucking RAN WITH IT. Like it was THE golden calf. Sacrificed their workforces, sacrificed lives, bought out ALL of the RAM, and ALL of the SSDs, only for the models to be wholly inadequate, or in the case of mythos, and the other new models, a nothingburger.
I am waiting with bated breath for the rug pull when these companies stop subsidizing the cost of compute, so that all of the assholes who dropped their workforce will have to eat massive piles of shit.
I've said in the past that we missed our chance for the tech industry to unionize in the 90s. The second best time would be when the shit hits the fan on LLM pricing.
The models aren't going to get any better. They're running out of data to feed to the machine.
Everyone is fighting the installation of datacenters in their backyard and rightly so because they are a MASSIVE problem, with very little reutrn.
The infra noise is harmful, the community ends up having to foot the massive power bill, the water supply ends up noticeably worse, and in many cases, they supplement with gas turbines causing the air quality to suffer as well.
The leaders of these companies are all shitheads who are willfully sticking their heads into the sand, because its in interest of their financial survival to do so, and to continue with the smoke and mirrors.
Fuck AI, and fuck those who support it.
-
R relay@relay.infosec.exchange shared this topic
-
RE: https://hachyderm.io/@inthehands/116619120828325229
I have a whole host of concerns with LLMs, and I'm mostly of the same mind as this person. Resource-wise, its a massive waste. Ethically, the companies producing the datasets said that intellectual property theft was fine, so long as they were the ones doing it.
but my biggest concern is that, as a cybersecurity professional, for a lot of the "vibe coding" models, they are pulling in data from github, reddit, and stack exchange over the past decade or so, maybe even longer than that.
While all of those sources over the years have been somewhat nice (if a little bit elitist) for learning programming concepts, there are A LOT of incomplete and incorrect answers out there. Or things that work on paper, but we then discovered through YEARS of CVEs that they aren't the most secure or best method for doing the thing(tm).
LLMs make no distinction about that. The training set, is the training set. There's no context that says this is "technically correct, but if you do this in prod, that is really fucking dumb idea." Even so, companies fucking RAN WITH IT. Like it was THE golden calf. Sacrificed their workforces, sacrificed lives, bought out ALL of the RAM, and ALL of the SSDs, only for the models to be wholly inadequate, or in the case of mythos, and the other new models, a nothingburger.
I am waiting with bated breath for the rug pull when these companies stop subsidizing the cost of compute, so that all of the assholes who dropped their workforce will have to eat massive piles of shit.
I've said in the past that we missed our chance for the tech industry to unionize in the 90s. The second best time would be when the shit hits the fan on LLM pricing.
@da_667 As usual, us proles will be the ones who will get fucked over.
-
RE: https://hachyderm.io/@inthehands/116619120828325229
I have a whole host of concerns with LLMs, and I'm mostly of the same mind as this person. Resource-wise, its a massive waste. Ethically, the companies producing the datasets said that intellectual property theft was fine, so long as they were the ones doing it.
but my biggest concern is that, as a cybersecurity professional, for a lot of the "vibe coding" models, they are pulling in data from github, reddit, and stack exchange over the past decade or so, maybe even longer than that.
While all of those sources over the years have been somewhat nice (if a little bit elitist) for learning programming concepts, there are A LOT of incomplete and incorrect answers out there. Or things that work on paper, but we then discovered through YEARS of CVEs that they aren't the most secure or best method for doing the thing(tm).
LLMs make no distinction about that. The training set, is the training set. There's no context that says this is "technically correct, but if you do this in prod, that is really fucking dumb idea." Even so, companies fucking RAN WITH IT. Like it was THE golden calf. Sacrificed their workforces, sacrificed lives, bought out ALL of the RAM, and ALL of the SSDs, only for the models to be wholly inadequate, or in the case of mythos, and the other new models, a nothingburger.
I am waiting with bated breath for the rug pull when these companies stop subsidizing the cost of compute, so that all of the assholes who dropped their workforce will have to eat massive piles of shit.
I've said in the past that we missed our chance for the tech industry to unionize in the 90s. The second best time would be when the shit hits the fan on LLM pricing.
@da_667 I've seen posts suggesting it's started, with copilot pricing, I think by a factor of 10x. I haven't chased down sources to confirm because I'm exhausted by the whole thing. But I metaphorically have my popcorn ready
-
The models aren't going to get any better. They're running out of data to feed to the machine.
Everyone is fighting the installation of datacenters in their backyard and rightly so because they are a MASSIVE problem, with very little reutrn.
The infra noise is harmful, the community ends up having to foot the massive power bill, the water supply ends up noticeably worse, and in many cases, they supplement with gas turbines causing the air quality to suffer as well.
The leaders of these companies are all shitheads who are willfully sticking their heads into the sand, because its in interest of their financial survival to do so, and to continue with the smoke and mirrors.
Fuck AI, and fuck those who support it.
also, do not @ about "lol, you use services that are hosted in a datacenter. Hypocrite." You sound exactly like the "we should improve sociecy somewhat." comic, trying to catch me in some kind of hypocrisy. No, Miss me with that shit.
My e-mail serves a purpose. Hosting services serve a purpose. Social media services serve a purpose. E-commerce infra, as god awful as it is, serve a purpose. The only purpose AI serves is to consolidate the workforce, produce slop that at first glance looks good enough, and is accepted before anyone realizes its a gigantic pile of shit, and to make the rich richer at the expense of literally everything else.
-
@da_667 I've seen posts suggesting it's started, with copilot pricing, I think by a factor of 10x. I haven't chased down sources to confirm because I'm exhausted by the whole thing. But I metaphorically have my popcorn ready
@da_667 the real meaning of 10x engineer
-
RE: https://hachyderm.io/@inthehands/116619120828325229
I have a whole host of concerns with LLMs, and I'm mostly of the same mind as this person. Resource-wise, its a massive waste. Ethically, the companies producing the datasets said that intellectual property theft was fine, so long as they were the ones doing it.
but my biggest concern is that, as a cybersecurity professional, for a lot of the "vibe coding" models, they are pulling in data from github, reddit, and stack exchange over the past decade or so, maybe even longer than that.
While all of those sources over the years have been somewhat nice (if a little bit elitist) for learning programming concepts, there are A LOT of incomplete and incorrect answers out there. Or things that work on paper, but we then discovered through YEARS of CVEs that they aren't the most secure or best method for doing the thing(tm).
LLMs make no distinction about that. The training set, is the training set. There's no context that says this is "technically correct, but if you do this in prod, that is really fucking dumb idea." Even so, companies fucking RAN WITH IT. Like it was THE golden calf. Sacrificed their workforces, sacrificed lives, bought out ALL of the RAM, and ALL of the SSDs, only for the models to be wholly inadequate, or in the case of mythos, and the other new models, a nothingburger.
I am waiting with bated breath for the rug pull when these companies stop subsidizing the cost of compute, so that all of the assholes who dropped their workforce will have to eat massive piles of shit.
I've said in the past that we missed our chance for the tech industry to unionize in the 90s. The second best time would be when the shit hits the fan on LLM pricing.
@da_667 personally, I want workers to demand to get retroactively paid for the time their position got canned for AIslop - WITH INTEREST!
Otherwise corporations won't learn, but painful costs!
#Sarcasm #Commentary #AIslop #AIbubble #Costs #Corporations #WorkersRights
-
also, do not @ about "lol, you use services that are hosted in a datacenter. Hypocrite." You sound exactly like the "we should improve sociecy somewhat." comic, trying to catch me in some kind of hypocrisy. No, Miss me with that shit.
My e-mail serves a purpose. Hosting services serve a purpose. Social media services serve a purpose. E-commerce infra, as god awful as it is, serve a purpose. The only purpose AI serves is to consolidate the workforce, produce slop that at first glance looks good enough, and is accepted before anyone realizes its a gigantic pile of shit, and to make the rich richer at the expense of literally everything else.
The graduating classes are booing all the assholes telling us that AI is the next revolution because they all know its bullshit. AI can't replace human creativity, it can't replace your ingenuity. It can't replace your perspective. By its very nature its retrospective and derivative. I know this isn't exactly comforting considering how shit the job market is right now, and I'm sorry for that.
-
The graduating classes are booing all the assholes telling us that AI is the next revolution because they all know its bullshit. AI can't replace human creativity, it can't replace your ingenuity. It can't replace your perspective. By its very nature its retrospective and derivative. I know this isn't exactly comforting considering how shit the job market is right now, and I'm sorry for that.
@da_667 there's no such thing as a clean revolution.
