I think you underestimate the willingness of the average American to consume slop. It really doesn't take much for an LLM to produce something good enough that millions of people would lie on their couch and zombify to it (while consuming ads). If AI were as crappy as you say, it wouldn't have been mentioned in the terms to end the recent writer's strike in Hollywood.

Nothing matters except the outcome. Let's say, hypothetically, that we invent something much closer to so-called "AGI" – I can guarantee you that many pundits and HN users will find reasons to believe it's not real and that it's crap or whatever. Why should anyone care? Plenty of people thought the Web was crap in the beginning, and now look at where we are. I don't see any reason to believe AI won't have some impact on jobs and the economy for the next decade. If it was total crap, no one would be using it, but all I know is tons of developers are using Copilot, and I'm using GPT-4 every single day. Are we all just spinning tops or is it actually being of some use to us?

It doesn't even matter if it is slop and you don't want slop. A huge oversupply of slop creates a market for lemons scenario. It is so hard to find the good stuff most people stop even looking for it, assume it is all slop and adapt to that reality. And then non-slop isn't created at the same rate because they're can't get through the noise anymore.

> Because AI is nowhere close to replace a large swath of jobs?

It's nowhere close now. 10 years ago we were nowhere close to generative AI. How will the AI landscape look in the next 10-20 years? There's a non-zero chance that it will be ready by then to replace many jobs.

While I am not terribly worried about the jobs over the next 3-5 years, I am worried about what the landscape will look like in 10-20 years. If the white collar jobs are getting wiped out and we lose a bunch of blue collar jobs like warehouse work and trucking, it’s going to be a real rough time for the economy.

There's a "non-zero" chance all life on earth is eradicated in an extinction event like a gamma ray burst too. That phrase is just smarter sounding weasel words.

With regards to generative AI replacing jobs, it seems to me that AI is best at replacing jobs I have a layman's understanding of, and AI is no good at replacing jobs I have a professional understanding in.

For example, I am not a lawyer or doctor therefore it seems trivial to me that AI will replace much of what lawyers and doctors do. Both boil down to reading a libraries worth of information and then making sense of it right? Bam, easy for AI.

However, I'm a professional engineer and I know how extremely complex and difficult computer science is to get right, especially at scale, therefore I'll eat my hat if in 10 years you can write "Create a facebook clone that can support 500 million simultaneous users" and the AI does anything but hallucinate nonsense that won't work. (There's also the reality that there is far more to large scale software than what can be typed into a computer or repeated by an AI, but hey, details).

I'll also point out that 10 years is not so long. 10 years ago Elon Musk told us AI was replacing driving and taxis. 2015 that feature was launched. 10 years later... not so much.

"it seems to me that AI is best at replacing jobs I have a layman's understanding of, and AI is no good at replacing jobs I have a professional understanding in."

You attribute this to a 'bias of underestimating other people's jobs' I attribute this to a 'bias of overestimate our own jobs'.

Most programmer jobs do not involve facebook clones that support 500 million users. Most programmer jobs are boring crud apps for random companies in traditional industries. Just think of what the outsourcing mills in India produce, and how many people they employ, then think again about how many jobs generative AI replace.

Also, Hollywood rapidly organised to try to slow down AI. They could see, very evidently, how extremely rapidly AI was advancing, and how lethal the threat is to them.

The actors and writers guilds moved to protect their interests against studios abusing AI.

Far easier to AI fake Jim Carey speaking on camera, even doing acrobatics, than it is to AI fake the development of even a tiny piece of novel UI in an app. The reason is we have lots of footage of Jim Carey doing those things and so we can easily recreate him doing those kinds of things. We can probably even put entertaining words in his mouth via AI because we have the entire history of human writing in the model so we should be able to easily create something semi-novel and funny to have him say. What these models cannot do is thoughtfully interview users, evaluate designs on functionality and aesthetics (and perhaps brand) and build an app that looks and behaves well but very different/distinct from existing apps. The novelty is the hard part for these AIs. Even that AI Jim Carey script will reflect or outright steal from the training data. This is why actors came together to say no. It's just too easy, today, to steal their likenesses and do things they would themselves never agree to do. If you're building yet another clone app or site, you too are like these actors, only you're unprotected because too many of us over the last couple of decades were foolishly saying we had it good, the sodas were free, and there was nothing to complain about. If, on the other hand, you're doing something novel, creative, and interesting, you will be fine, better than fine, you will be able to command a premium over the AIs and all the loser might as well be AIs who thought they were too good for a union.

Who knows, because that'll take breakthroughs we don't have yet. Where we are now, however, probably cannot evolve much further, so this is kinda what you're gonna get from this period for AI probably for a while. Transformers plus all the web's content and a bit of chat UI and this is what you get from the 2010s breakthroughs. It's a pretty decent auto-complete service but not one that's likely to change the world in profound ways.

Recall that we've had many eras of AI, going back into the '70s with expert systems and back propagation that led to neural networks that got a lot of investment in the '80s along with early machine learning and advances in speech recognition and computer vision. The '90s work spent a lot on probabilistic and Bayesian methods, SVMs and Q-learning, then we got big data and deep learning, and don't forget Siri and Roomba to round out the 2000s era, the one that came before the current regime.

What the next breakthrough and investment era will bring, we cannot say. It may be a minor leap or a major one. We just don't know.

>Because AI is nowhere close to replace a large swath of jobs?

Is that a claim or a question?

In either case, doesn't matter, as long as it's close to replace a swath of jobs - even if it's not some particular "large swath" it can't replace yet (or ever).

Doesn't have to replace all jobs to have a negative job market impact. Just more than can be reclaimed with new ones.

The clear hypothesis in the question was "for jobs that AI can do just as well as a human".

It's closer than you realize for more jobs than you can imagine. Do you not realize just how many office jobs are simply repetitive tasks? Everywhere, for everything. Microsoft is working overtime to replace entire staffs with a new keyboard and a single prompt engineer.

People who really KNOW what they are doing will command a very high premium because they'll be needed to do things that can't be done via AI tools, but everyone else is kind of screwed. Being the best at what you do was always a good idea, but it's about to be a requirement.

I am not sure if raising minimal requirements is a bad thing. For example, it is harder to get into medicine now than it was 100 years ago, but is it bad? Maybe AI revolution will simply raise education requirements and those will lead to "smarter"/more specialised humanity?

I mean, yeah it is sort of bad. There’s a huge shortage of medical staff in the US (and many other developed countries). I’m not sure raising educational requirements (yet again), is really going to be all the helpful for the general population. Everyone shouldn’t need a masters degree just to make a basic living. Increasing educational requirements will increase people’s already inflated school debt and decrease the amount of time they can be in the workforce. This would give them even less time and opportunities in the workforce to make money (and repay their enormous school debt). And if the minimum requirements to participate in white collar work is now a masters degree, but AI has eaten a large portion of available work, then having more education really isn’t going to do much, since most everyone else will also have to increase their education to be competitive.

I mean, even some of the most basic jobs today require a university/college degree, when in the past a high school diploma would have been sufficient. It’s just moving the goal posts all over again.

Is your argument that medicine today is worse off than medicine 100 years ago because of increased barriers to entry? That's what it seems like you're saying but that's so bat shit silly that it can't possibly be the case.

No, that’s obviously not what I’m saying. In fact 100 years ago, doctors were actually receiving about the same amount of education as doctors today. Residency programs became a requirement in the mid 1920’s. But I get the point you’re trying to make.

It was the adoption of the Hopkins model and standardization of medical education— which focused on a mission of education, research and clinical practice— that led to a lot of medical advances that we have today.

My point, is that too much education (and its associated costs) can become a barrier to people joining industries/paying off their loans. Do you suggest doctors increase their education for longer than a decade? Maybe 2 decades of education would make better doctors, but is that really a net benefit to society? Not too mention, a good portion of a doctors education already is actually just learning on the job (while being paid a pittance).

Education (in the US), is not free (or even reasonably priced) and wages have stagnated for quite some time. All you’re doing is burdening people with more debt in order to still have to fight in a job market against their fellow humans as well as now AI.

If companies (or the government) took on training costs (like they often used to) and people were paid enough to make a comfortable living, then I wouldn’t even be discussing this. By all means, bring on the AI revolution.

But until we have some sort of safety nets in place to prevent all these people falling into poverty, requiring more educational investment and debt burden is not really a great solution for a large portion of the population. I mean, even a person’s healthcare is tied to being employed, and one of the political parties is still talking about repealing the ACA to this day, one of the only bills to actually try and tackle the absurd healthcare costs and coverage issues. More time in school would also just increase healthcare costs an individual would have to pay.

Judging from the present state of things, we’re nowhere near solving ANY of these issues that we are currently experiencing. Without proper guard rails, a rapid and wide adoption of AI is only likely to exacerbate these issues that our society is presently struggling with.

AI is 10x closer than it was six months ago. It's a quickly moving field. And without actual improvements in quality, you'll assemble pipelines to gradually refine quality before shipping, and every step of that pipeline could be staffed by inadequate AI. Why wouldn't Fordism and Taylorism work when we have intellectual machines?