Generative AI isn't going anywhere, and will replace a number of jobs.
2mon 18d ago by sh.itjust.works/u/PlzGivHugs in changemyview@sh.itjust.works
Generative AI has a number of uses that are already widespread, and I don't see going anywhere. Things like clipart and stock art, and initial contact customer support. AI automates these jobs, making it far, far cheaper than hiring a human to do the same job. It only takes one higher-end PC to do a job that a human would have had to be paid for. The economic incentive is already there.
Furthermore, Generative AI is a genie thats been let out of the bottle, and I don't see ever being put back in. These models are just files, which have already been replicated and become widespread. Sure, progress may slow as the "We're making a general purpose AI." bubble bursts, but if these tools work, they'll continue to be developed, and people will continue to get better at manipulating or augmenting them. I don't see any reason that would stop generative AI from continuing to exist from this point forward.
Generative AI isn't going anywhere, and will replace a number of jobs.
Change my view.
This is a hard one to debate because "a number of jobs" isn't really a statement you can counter. Yeah it'll probably replace somebody, the question is how significant will the impact be and will those people be able to get other jobs/will any jobs be created by it. Currently it seems to be creating far more jobs than it's costing, mostly just building places to put servers.
What you're probably looking for: Generative models are just an algorithm, arguably even a fairly simple one. We're already seeing the limits of how far we can push the current paradigm of model structures so without a big change in how they work things may stagnate. The problem with the current state of the art is that the average person can, if not always specifically call out generative model output, definitely note the mediocre quality. The most promising use case is writing certain types of code because it's repetitive and predictable in structure, but even there they reliably barf on moderately complex tasks where you can't easily let them run forever closed loop (as in, they can't self-test the program because it generates output that isn't text). Most other use cases fare far worse so if we're not seeing mass software layoffs I'm not convinced we'll see much damage elsewhere.
The only other case where one could argue it's truly all-powerful is scams because the scams deliberately target the bottom of the bell curve already and the models just make it easier to automate. So, does putting scammer foot soldiers out of a job count?
My perspective is more in response to the seemingly widespread view that generative AI will just go away if we complain enough, and/or the AI bubble pops.
I'm actually not too concerned about AI in the case of programming. Generally, the sort of programming work simple enough for an LLM to do well is stuff handed off to IT or small contractors to figure out, rather than dedicated programmers. My thinking was more for """low-level""" arts jobs and customer service. For example, most stock art corporations use is just meant to be fast, minimal, and cheap. Previously, this would have meant buying a bunch of stock assets, or hiring an artist or two to create graphics packs or templates to be used in stuff like publications. Similarly, for something like game art. For projects targetting a more discerning playerbase, this won't work, but for casual games, its already common. The art is just something to distract players while they're on the toilet, so cheaper is better. Customer service jobs on the other hand, AI doesn't replace quite as well, but the cost difference is so great, I'd be shocked if they don't stick around. This isn't going to upend the economy like AI bros claim, but will replace a significant number of entry level jobs.
LLMs and generative AI will, and already has, replace "a number" of jobs.
LLMs have revealed something that shocked lots of job sectors: Quite often, having absolutely shit quality is good enough. 15 years ago, you could get two types of translation: Excellent, and None. So if you didn't want "none", you had to pay for Excellent, and as a result there was a very large sector providing Excellent translations of all sorts of materials. LLMs have revealed that many, maaaaany people don't want Excellent translations, they just want not-None. Suddenly, the market for Excellent translation plumets, and the market of not-None skyrockets. This has happened for quite a few different sectors. Especially the ones that work with text, like content writing, translations, etc. Then it happened for images. People don't generally want amazing artworks, they want not-a-blank-space.
Yes, LLMs generate slop, and will always generate slop, but we've recently been confronted by the reality that there's an absolutely immense market for slop.
But there's a different matter: LLMs are expensive, and nobody wants to pay for them. You're not paying what it costs to run the LLM you're using. If you're a private user, you're not paying what it costs to answer questions, generate cake recipes or make an easter-bunny picture. If you're a huge company paying millions for tokens, you're not paying what it costs to run the LLM. Tokens are hugely subsidized, but even they weren't, LLM companies couldn't afford to run. They're going to collapse 0.01 seconds after the venture capital stops coming.
Will the major collapse of large LLM companies end LLMs? Nah, of course not. Smaller, local LLMs will work just fine for many purposes. But almost every large scale LLM implementation is currently running on an industry that might as well be shoveling cash into a steamboiler.
So yes, "a number" of jobs will be replaced. It won't be a huge number, but contest writers, translators, basic artists and people in similar industries will be replaced, at least temporarily.
What will happen is that as the post-covid job bloat is reducing, the economy is slowing down and hitting its next crisis thanks to the US adventuring in the middle-east, there will be LOTS of people getting fired. And many of those companies will not say "We're firing people because the company isn't making money and we need to reduce our output" they will all be proclaiming "We're firing people because of AI, please don't think we're doing poorly!"