I think all the AI stuff is oversold. I don’t think it’s very good, but I think it’s going to be hugely successful because big companies will happily replace people with half broken garbage. They don’t care if nothing works because no one will have working systems and we’re forced to buy a lot of their junk no matter what due to oligopolies and monopolies, especially for critical services like communications, electricity, etc..
Perhaps more to the point, executives will replace people with half broken garbage and then ride the wave of cost savings and innovation woo just long enough to cash out a fat bonus and bounce to another organization like a swarm of locusts.
So much of the discussion seems to be along the lines of “yeah it’s broken now, but it’s just the start, it’ll get better”
I can list so many things that were hyped up the same way 10 years ago that still aren’t revolutionising the world today.
Maybe it will get super good really fast. Or maybe it’s the next self driving trucks or 3D tv where solving the last of the problems is much harder than the first 90%.
> I can list so many things that were hyped up the same way 10 years ago that still aren’t revolutionising the world today.
The problem with it is that it is being marketed as the do-all productivity enhancer. It is not. There are some fields in which it excels, but are all rather very specific (think summarizing a long document).
I feel like even in that specific task, it's only really good when you care more about processing huge amounts of data over accuracy. I wouldn't trust it to summarize a single PDF research paper, but it would be fine as a search engine looking over millions of documents, or trying to summarize the common points in thousands of user feedbacks.
Something like "Find me documents that talk about achieving x using y process". If it finds a few irrelevant ones or misses some, its ok and still provided value by searching far more things than you ever could manually.
I've a few times given an analogy on her about how it's like these companies are selling really delicious chocolate and then just add shit on top of it saying its a bonus. I don't get the hype, especially when there's something quite useful in there. Why is everyone overselling everything? Not by a little, but by a lot? I feel like it's good for short term rewards but terrible for long term.
Battery though, we have seen the price (USD/KWh) go down every year, and the energy density (KWh/kg) increase every year.
Research articles are usually things that take years to trickle down into the mass production. And by then the hopes or lab results no longer translate, or they've evolved into improvements and savings that are harder to link back to their original source.
I don't think the AI stuff is oversold. If anything, it's going to hit us sooner than most people expect. Once robots and their AI software are good enough, once self-driving is good enough, it's game over for most manual jobs. And many software-based jobs will be automated even faster (accountants, lawyers, data analysis, etc).
> Once robots and their AI software are good enough, once self-driving is good enough, it's game over for most manual jobs
Why would those getting good enough happen sooner than people expected? I'd assume those will get good enough at roughly the rate pessimists expected, at least if we look historically that seems like a good bet, these things move really really slowly once you try to apply it to the real world in practice.
> I don't think the AI stuff is oversold. If anything, it's going to hit us sooner than most people expect. Once robots and their AI software are good enough, once self-driving is good enough, it's game over for most manual jobs.
Would you call an AI bot to fix your heater? I would not.
There's no substitute for artisanship.
If company can replace people with AI, emphasize on 'can', then it is not oversold by any means.
AI is different than previous generation of disruptions, because its primary target will be software industry itself. ChatGPT has replaced stackoverflow and Google for me.
I don't think the market will allow that. AI doesn't do autonomy well. It can work ok for a few steps, maybe surpassing average humans, but in long time horizon tasks it doesn't carry out. Companies need their people to extract the AI juice, they need humans-in-the-loop. Especially in critical tasks.
Even if AI can't carry out serious tasks, it can still have better form than humans, it would be an improvement for customer support compared to what some companies are doing today (almost nothing). We still need humans to escalate to, when AI is useless.
I hope instead of leading to layoffs AI will lead to competition on quality, diversity, price or customizability. Raising the bar for everyone, instead of replacing us. Is there any reason a company would prefer lowering costs to competing for market share with AI? There's more upside to increasing profits than reducing costs.
I think the market already allowed that. AI shitifying everything is just another step (a big one) in the same direction we've been heading anyway.
Think about offshoring support representatives. Instead of someone actually trying to help you get someone bound to a script and following the motions. This infected the entire service culture. And the market accepted.it,because all big companies did the same.
AI in support roles is an obvious next step, now that everyone is used to scripted interaction, a half baked AI solution is not much worse.
This is exactly what I see happening. It doesn't matter if it's a terrible consumer product as long as the entire market moves at the same time.
The other area that really worries me is government. I think they're the most likely to get sold on AI as a cost cutting measure, we'll all suffer as a result, and there won't be any recourse.
I can't believe how many people are willing to hand the world over to a bunch of non-deterministic, opaque boxes. Being an outlier is going to suck in the future.
Yes but now it is easier to break the rank and offer better customer service using AI, especially if AI gets better every year. There will be no excuse for offering 2023 level quality in 2025.
I believe it is a coping mechanism for many people.
Software engineering is a field in which many revel in its perceived complexity, the 'status' that comes with doing a job that brings so much to the table, yet is not understood by the average person.
Now, all of a sudden, there is a tool that is starting to bite into that and it is very easy to dismiss it using phrases such as "half broken garbage".
To be fair, regardless of the status and what have you I've seen a lot of people in my time complain that they're making unoptimized garbage with half broken tooling and they weren't shy about saying it.
For now AI in this context is often just another piece of such tooling and people using it may experience for example github copilot or the like spit out faulty blocks of code most of the time when it's not doing something cookiecutter.
When you then know full well it needs to improve till it's almost correct 100% of the time and do a whole lot more then yeah people are going to be dismissive for a good while longer.
Self-checkout machines in supermarkets are famously shit and generally require customers to perform extra work, and yet they are preferred by shops because they can pay one employee to manage 10 machines instead of having 10 employees
A lot of stuff companies are doing is overly complicated because of their legacy. Awful lot of people are doing work beneath their qualifications because this. The AI can remove a lot of complications and do a lot of mundane jobs IMHO.
Not the AIs we have today, I don't even think the initial AGI will be smart enough to significantly simplify current processes, it would need to be much smarter than the humans it replaces to do that.
I feel like AI will replace a lot of the "pointless"/bottom of the barrel jobs.
If you program entirely by copy and pasting from stack overflow, sure you should be worried. But despite that being a meme, most people do more than that.
No one really knows how AI will develop and what sort of capabilities the next iterations will have. These kind of predictions are pointless.
1 year ago if you’d told me that commercial photographers would be toast, I wouldn’t have believed you. But after Midjourney V6, I have zero doubts that commercial photography is absolutely doomed.
I'm not so sure, at all. I think folks are woefully underestimating most people's desires for at least some level of "authenticity". For example, in the past 6-9 months or so, it feels like nearly every random blog post or corporate announcement I read feels the need to use an AI-generated header image. I understand why - "stock imagery" has always been used to paint a scene for an article. The thing is, though, is that, at least for me, it just cheapens the use of imagery in the first place. Every time I see one of these now I roll my eyes, I think they look cheesy and stupid. Even ones that are photo-realistic quality, you can tell by the composition that it's just generic AI.
If you make something ridiculously cheap to produce, it also has the effect of discounting the value people take from it.
MJ 6 is only a few weeks old or so though (I think…internet time is weird, but not 6 months for sure!) and is quickly becoming nearly indistinguishable from real photos. And that’s not even mentioning the obvious future v7.
We’re crossing the uncanny valley and soon enough you won’t know if an image is real or AI. The tell-tale signs will fade (already are) and the “authenticity” you speak of will actually be there. So to speak.
I can already take a photorealistic photo with my camera, no AI needed. We don't hire photographers just because they can create photorealistic images, we hire them because they can take the right photorealistic images.
Thanks very much for putting the point I was trying to make so well and succinctly.
It's similar to how people are worried about AI killing all programming jobs. Yes, I've also been blown away by the ability of LLMs to write some code, but my value as a developer comes more from being detail oriented, understanding when requirements are underspecified, thinking through edge-cases, and then finally being able to turn this into something a machine can execute. Compilers didn't put earlier assembly language programmers out of business, it just required most programmers to program differently.
While I do worry about the sheer number of jobs (because I think more junior, entry-level jobs will get supplanted, which makes it harder to move up the career ladder in the first place), I think people will still want the expertise of good photographers.
I don't see why we need AI for things that already exist in stock photo portfolios, but for very specific things, especially where they are clearly imaginary, it makes sense.
Stock imagery is annoying, hopefully AI can kill it.
Maybe we’ll get ubiquitous local stable diffusion, companies can just send the prompts and then we can just skim the text to get the idea if we don’t feel like rendering them.
Why do you think commercial photography is toast? (asking as someone who paid obscene amount to few photographers last year ((wedding, budoir, maternity photos hoot))
If anything I feel like ads and product listings are a case that definitely won't move to AI. Surely the ad needs to be a photo of the actual thing, rather than an approximation? I realise ads are often heavily edited, but correctness is still a very important concern in advertising in general and well supported with legal precedent in much of the world. There are industries that have moved almost entirely to digital renders (car adverts, Ikea catalogues, etc), but in these cases I think there's still a legal necessity to use highly accurate modelling.
I think the area that will move will be stock photos. They're only really used when the photo doesn't actually matter. I'm already seeing many AI generated images replacing stock photos on random blog posts.
Maybe it's a hard thing to search for, but I mostly just found FTC guidelines which state "Your Ads Must Be Truthful and Accurate" but I can't find any more details. I think there's stuff like you can't misrepresent the dimensions of something or the color but I doubt there would be laws that don't effect photoshop but would apply to AI.
Depends. I already see low quality news sites and stuff doing this, but the images are always a bit mangled.
It’s fine for filler images that don’t serve any purpose. But say you want a photo of a specific street for your story, having a street that looks a bit similar but mostly randomised significantly detracts from the value.
Mass production hasn’t wiped out hand-made furniture, but it’s damn fucking hard to get decent furniture these days on a reasonable budget. There used to be a kind of comfortable mid-range where you could spend extra money and get decent furniture, without going into luxury or bespoke furniture. Now, it seems like everything is either similar to Ikea in terms of build quality, or some kind of expensive / luxury good.
>Quality issues. Mass production haven’t wiped out hand made furniture.
No, but it has turned anything but the very highest end furniture into absolute garbage. It's nearly impossible to find furniture made of solid wood (whether a couch frame or a dresser drawer) unless you're spending darn near 5 figures on a set.
Yes it has. Just because _some_ woodworkers still exist doesn't mean the profession as a whole hasn't massively declined with the advent of the likes of IKEA.
Honestly, I could see AI also disrupting wedding photography. For example, instead of paying several thousands of dollars for a photographer, you could instead pay $1,000 to rent a set of 4 tripod-mounted 360 degree cameras. These would monitor everything that's happening during the ceremony from different angles. Then you have each guest do an "optimization photo" when they sign your guest book. This would be like an abbreviated version of setting up FaceID on your iPhone — getting a couple different angles.
The software would then create a bunch of photos based on the positioning and expressions captured by the tripod-mounted panoramic cameras, but enhanced with the facial photos taken at check-in. The number of photos you could create would be infinite, of course (and would likely be priced based on the number of photos you review and download).
Photographers do provide value when they compose photos, but this aspect would be less important if you can edit photos after-the-fact using AI. Hell, I might like to touch up some of the photos from my wedding, to get rid of unsightly background elements that the (expensive) photographer failed to account for.
While I can see that being interesting, almost like a photo booth that's a common addition to weddings, at the same time I also think it wouldn't replace an actual photographer for pretty much anyone, since they would capture what actually happened during that time period with proper composition.
If people wanted just any old pictures at their wedding, they'd put some disposable cameras on the tables and call it a day (and some do exactly that, to be fair, and save a lot of money. But most people don't).
A.I. is not likely going to provide authentic in the moment photos anytime soon, if ever, and if it ever does then it's basically just an motorized mechanical photographer anyway, that would need to be able to go over various terrain effectively or be a drone, probably. At that point it'd likely be about as expensive as just hiring a person, and a person would probably be more friendly and get better photos out of people.
I'm envisioning the ability to actually create photos with proper composition, using a combination of the tripod-mounted cameras. In particular, I'm thinking of how the NFL uses multiple cameras to create imagery that pivots through space, even where there are no cameras. The quality isn't super great, but that is created in near-realtime. With these tripod cameras, you could create shots from anywhere, after the fact. And you wouldn't have someone crouching in the aisle, clicking away. I'm sure the technology isn't ready just yet, but I imagine that within a few years it will be possible to create these sorts of shots with hardware that is competitively priced compared to a professional photographer.
Maybe you're right, I don't know. I still think it will remain a niche novelty for quite some time, though, as it won't be considered authentic enough.
NFL has plenty of authentic camera work mixed in with a bit of flair like that (I didn't know they did that now, that's interesting), but I imagine for a wedding they care even more about making sure it's fully accurate and not generated. Maybe I'm wrong, guess we'll see what happens in the next ten years.
I do know in my case that we could have gone cheaper on photographers and we didn't, though (we saved money by not getting video or paying for a physical album afterwards, not the people), and I wouldn't be interested in A.I. generating the pictures (even assuming it would be better quality than today) in any way instead of snapping what actually happened.
And I'm not someone who's against people using generative A.I. for the most part, so that doesn't have anything to do with my preference.
From a few crappy photos of your face, you can generate photos of you in nearly every setting: so I think all the "studio" part of a commercial photographer's job is toast.
> Why do you need the actual baby when you can have a fake baby with mangled hears and 7 fingers per hand?
Comments like these are fine on Twitter and Facebook, but if you're on HackerNews, commenting on an article about AI, it makes you look completely out of touch. You've clearly not kept pace with the technology and are just repeating verbatim what you saw 6 months ago.
HN, as a whole, has been so behind the curve that it's not even funny.
You can with finetuning, but I think parent is being sarcastic :-) I think the point is that no one (ok, not no one, but few people) will view fake photos in the same way as real ones when it comes to sentimentality.
We don't need to predict what the next iteration will have to calculate the damage, just understanding what "this iteration can do" is almost enough to cover that 40% job losses due to increased productivity.
It’s really hard to say how many unsolved issues with AI are going to be fixed tomorrow vs 20+ years from now.
Midjourney v6 still has massive issues with lining up stuff that goes behind something else and I have no idea if that’s fixable or an inherent limitation to their approach. Hands have noticeably improved, but are still very hit and miss etc.
What’s really frustrating is I often can’t tell if I’m bad at prompts or the underlying system is has issues.
If we get to a point where AI can generate realistic 3D models it’s game over for a lot of things. If you have a 3D model of a character it completely eliminates continuity issues. At that point entire games and movies should be possible without a single actor/actress or any support crew.
It's not just NVIDIA, Ubisoft also has some very impressive research in this area. I believe Ubisoft has already implemented some of the new tech in their recent games - it makes sense, they have hundreds and hundreds of people work on their games and they require huge amounts of animation, any savings there are going to be huge for them.
If anything it’s easier. It gives a nice abstraction to capture and transform movement rather than figuring out each frame how to draw everything from scratch
Why do people pay premium for "hand made", "artisanal" etc?
If AI is cheap, then by using AI generated content to cut costs, you will cheapen your brand by association. There are examples of brands doing this to themselves already.
Some people pay a premium for handmade, artisanal stuff. In the past, that was the only option. Today, most people buy mass produced goods. You can still buy handmade furniture, but most don’t. You can still ride a horse-drawn carriage, most don’t.
And who says you can't use AI to produce cheaper "handmade" and "artisanal" stuff?
I am not saying this in gist or to as a gotcha, the reality is that many --I dare say majority of-- "handmade" and "artisanal" brands are just that "brands" and are often only slightly --if that-- different from their mass produced "counter-parts".
And yes, I hate how much quote I had to use too, you can blame the marketing people for that.
I literally used Midjourney to create luxury brand quality images for one of my d2c product. The images are at par with print ads for top tier luxury brands. My friends who work in creative industries had to zoom in 300% to spot that its AI generated.
Average consumers won’t be able to tell at all
Do yourself a favor and head over to Midjourney’s explore tab after signing up for an account.
I have no idea what you're getting a commercial photography has been on the edge of obscurity for a long time, and there's little new here. Of the several photographers I've known in my years, there's only maybe a couple that managed to make an actual career out of it.
I'm not entirely sure a final nail will be from AI, but there are def. too many people chasing too few jobs in that industry.
We may have passed "peak office". Despite heavy pressure from the "get back to work or else" movement and the commercial real estate industry, office occupancy in the US is still way below pre-pandemic levels.
What if the last available job ever is a soldier? I mean, if the only scarcity becomes about owning and having place to put the machines that work for you and no human work is better than the machine work maybe those who don't have AI to work for them will try to destroy those who have and seize the machines and the resources(energy, space, other machines).
Maybe its a bit of a stretch but I don't believe that people will be like "oh crap, my skills don't make me a living anymore. bad luck, nothing I can do about it". Currently people in disadvantage(real or perceived) are already revolting against affirmative action and cheaper labor from abroad.
There is a thing in economics called "curse or resources", which is when a country discovers oil, diamonds, or some other rare mineral that can be easily recovered. One would think that's a good thing, because it brings jobs and wealth to the country.
However, history shows that it's actually bad because it brings out the worst humans, the ones with the most greed and the most curel methods, and results in things like civil war, military dictatorships, and much higher wealth inequality than before.
This is in contrast to the countries that are resource poor, they live from human labour instead and seem to have better development trajectory. Of course, there is an impact of influence from the west which benefits from cheap resources, but I'd argue that some of those effects come from the inside.
I think that with AI, our societies will look more like some african country with warlords than the developed, labour based economies we have today.
Does that mean that soldier is going to be the last job? AI is coming for those jobs too. Drones of various kinds are already used in today's wars, but they do certainly still have need for foot soldiers.
I think that the emergence of AI will have a major impact on resource distribution questions in developed economies, which until now have distributed resources based on labour force. Without intervention, it will trend more towards distribution seen in countries with the curse of resources, which is more violent.
> I think that with AI, our societies will look more like some african country with warlords than the developed, labour based economies we have today.
It's not the same. Oil and other extractable resources are not evenly spread. AI can be. You only need to train once the base models, but oil extraction is continual work.
As for consumption - both resources and AI support diverse applications creating diverse jobs. AI has this interesting property that it adapts to the user by itself, the benefit flows mostly to the user as they prompt to solve their problems. The AI provider only gets cents per million tokens.
When talking about job market comparing humans with AI is not very useful, humans will all have AI support, the real comparison will be between AI alone and AI+human.
I agree but I think the solders will be the solders of the army that wants to seize the resources of the people with AI and themselves become people with AI. A temporary period until AI becomes AGI that can do everything, including building fighting machines. Then maybe can be anther period where people are trying to destroy AGI factories which build robots that fight humans.
True but the the current drones are made by people. If the thing beocmes a political movement, think a charismatic populist leader rallying people against the people with the smart machines, people can destroy the AI folks way before the AI folks can make machines that make other machines autonomously.
If anything, history has gone in the opposite direction. The Luddites who broke into factories and smashed machines? Publicly executed.
The same argument you’re making about automation could’ve been made 70 years ago before outsourcing was the norm. Those factory workers whose jobs were sold for pennies on the dollar? They will rise up against the factory owners. Actually, it turns out that racism is a pretty powerful distraction.
Luddites might have lost but the governments and the industrialists also got lots of restraints against actual problems that the industrial revolution caused. Pensions, working condition requirements, environment protections, restrictions on child labour etc. Luddites were early reactionaries without a real solution to actual problems, then the world has seen successful movements that proposed workable solutions for the real problems.
I don't believe even a bit that the progress can be stopped, can't even be slowed IMHO. However the society will be dramatically transformed and that transform wouldn't end as a libertarian dream world.
The problem is the thinking that either stopping or slowing is obtainable or even desirable. “Progress” isn’t a single path thing. The goal is to progress in directions that minimize externalities and maximize benefit to all. Instead of trying to retard or stop a trend entirely, instead how can we counterbalance negative outcomes and direct its growth in a positive way?
John Henry fought the steam digger and won, but it killed him. Fast forward to the present I don’t see tunnel projects unmanned - instead they’re covered in people working on various things. The difference is the people aren’t hired for their strong backs but for their strong mind and skill. Machines aren’t replacing people - they’re replacing the people doing the job now with new people doing an adjacent job. The churn in labor is definitely painful, but it’s disingenuous to say automation has eliminated work for humans. It’s just changing the nature of work.
Finally, what’s so awesome about working? I welcome a time when humans are freed from the chains of toil. And if robots fight each other, as long as they do it away from people, that’s one job I’m thrilled to see end.
The general consensus seems to be that low income and low skill workers will be affected real soon.
I don't know where we currently are in the hockey stick curve in ML/AI development, but if it keeps developing at this pace it will soon start hitting high-skill and high-income workers, and then we're going to have a huge economic problem.
Most low-paid workers are safe for the time being, since they do jobs that cannot be done by LLMs, for example, cooking, cleaning, caring for children, driving, etc. The current AI wave threatens only mid- and higher-income workers, for example, junior analysts in finance.
I have a friend who is a partner at a big law firm. They are launching a platform that facilitates the creation, analysis, and revision of legal contracts.
Platforms like this will have a huge impact on law firms, small and large. The biggest deals will always get human scrutiny, but midsize deals will probably be done with tools like these. Law school graduates who expected to be able to make six-figure salaries to pay off their six-figure debts are going to be negatively impacted.
The smart ones are already trying to figure out how to use these tools, what their strengths and weaknesses are, and which areas of law are going to be affected less (litigation versus transaction).
> The smart ones are already trying to figure out how to use these tools, what their strengths and weaknesses are, and which areas of law are going to be affected less (litigation versus transaction).
I don't think it matters much. Eventually most AI tools will be as easy to use as a smartphone. Why waste time learning ever changing AI workflows and APIs at this point.
I often hear the argument "the destroyed jobs will be replaced easily, as has been the case in all technological revolutions", but I think it's deeply flawed:
Up to know, all technological revolutions have occurred in an exponentially growing economy. That's why Schumpeter's "creative destruction" was not a problem: it was compensated by the opening of new areas for growth, mostly unlocked by consuming more resources.
But now we are hitting the limits of the planet, and amongst others the limit of extraction rhythm for most our materials. So the exponential growth can't continue. (apart from the growth from technological progress, which will be much lower)
That's where the argument doesn't hold: we do not have a growing supply of work anymore, so the increased productivity of some workers will just prevent the others from working.
Let's define AI here means assisted intelligence. AI will likely replaces jobs and create jobs at the same time. As most optimist will tell you. The problem is the skill set mismatch will never be caught up fast enough for most human being.
This will definitely create or widen inequality. However regulate AI aka EU also wouldn't help. You may slow it down but the result is still the same.
I don't have a solution. I know the basic common theme to this question on HN is UBI. But I have yet to see or read anyone's answer to how UBI solves housing issues. ( Which is a problem slightly different in different location / country / politics ) I guess that is a different topic.
> I don't have a solution. I know the basic common theme to this question on HN is UBI. But I have yet to see or read anyone's answer to how UBI solves housing issues. ( Which is a problem slightly different in different location / country / politics ) I guess that is a different topic.
The first thing would be to de-profitize places for humans to live.
Interest rates for humans with 0,1, or 2 homes could be set to low or preferably 0 interest.
Taxes for homes could skyrocket after 2 homes, as to make parasitical rental companies uneconomical.
AirBNB and similar rental structures can be required to business and hotelier standards (aka: make it uneconomical).
Banks can also be compelled to take evidence of "paid $2000/mo rental for X years" as proof of paying same for mortgage.
Companies who want to be in the rental industry must follow stringent cost-lowering structures, like rent control and other de-profiting strategies.
And also, we need to build housing of all sorts, and not just single home dwellings and apartments. We're missing the various sorts of "middle urbanized housing" cause it's illegal.
I am sure this will happen eventually but there is extreme uncertainty as to the actual timelines. Any attempts to aggressively manage it much too early (our current position in the timeline) or much too late will cause extraordinary harm to the entire population.
I would argue that the timing of policy related to this issue can do at least as much damage as the technology itself.
Wasn't the preliminary consensus from current studies that LLMs improve the performance more for the workers with lower initial performance, in various fields (copywriters, developers)? I.e. C worker becomes a B+ worker, but A worker becomes A.1 worker. So if anything, it should reduce inequality...
