AI and two hundred dollar tasks | Hacker News Reader

AI and two hundred dollar tasks

blog.ninlabs.com

83 pointsazhenley84 comments

This is presented like it's a good thing, but it quite probably implies a local maximum in cultural abilities for a long time.

How do experts become experts? Through time and practice. Many great photographers hone their skills or discover their talent when working at entry level and mid level positions - shooting portraits, etc.

Most great musicians, bartenders, software engineers, etc. do the same. Without entry level/apprentice jobs the only other model is patronage from the rich, and that was never a good model even before today's rich largely abandoned the belief that they had a responsibility to contribute back to the society that allowed them to become rich.

Without a pipeline of new talent, where will new state of the art come from? Where will more training data for the planet burners come from?

It's a tool, and just like other tools, it multiplies human labor and intellect.

I don't need to spend a decade becoming proficient with a pocket knife, and then another decade becoming an expert with an ax.

Instead, an experienced person can show me how to be safe with a chainsaw in an afternoon.

I won't be an expert with the chainsaw, that still takes many hours of experience, but I don't need the knife and ax prerequisites.

The calculator has allowed anyone to do complex arithmetic while they're holding a calculator, required by a job or some other formal or necessary context. At the expense of retarding most people's ability to do everyday math at places like the grocery store, a context in which they won't have the wherewithal to pull out the tool. It has not increased intellect regarding anything but rote keypressed calculation.

On the other hand, looking up trigonometric functions in a table to find precise answers is not a valuable human activity. Calculators keep you from having to do that.

That's right! And what human activity do these AI tools replace? A calculator is only useful if you are able to put in the right formula, which means you still need to understand the problem and its solution. On the other hand, you lose the ability to verify that the solution is actually correct, beyond double checking that all your inputs are okay. And if your solution cannot be found through the operations supported by the calculator then you're just stuck and need to think of another way.

Modern AI is much more advanced because users only need to ask questions. But that means people who rely too much on it will lose the ability to understand the solution. Or, even worse, they will never acquire that ability in the first place if they offload all intellectual tasks from primary school onwards to an AI. If people leave the thinking to an AI, they will lose the ability to think on their own.

Of course responsible users can and will use AI to significantly boost their learning and capabilities, but the question is what will happen to society at large over long periods of time. If the dominant attitude is that only results and outcomes matter, then I think that we are heading towards an intellectual wasteland.

And that is kind of the point. Tools both augment and limit human capabilities, so that we gradually forget what was before the tool existed. But a tool like AI could limit people's ability so much that we become collectively blind to its limitations. And that is exactly the local maximum of cultural development that started this thread of discussion.

Is using AI the equivalent of using a calculator to look up trigonometric function values, or outsourcing understanding what the functions are?

Right now, more the former, but it depends.

I'm an iOS dev. The better models exceed what I can learn about the operating system I develop for, in the same way that a trig function on a calculator exceeds the precision a human can memorise. (So do search engines, but they can't respond to natural language queries combined several topics at once).

Because LLMs perform like juniors (sychophantic juniors with Alzheimer's), if you use them blindly without knowing the domain, that's still occasionally like picking random trig buttons because you don't know which of [sin, cos, tan] you wanted.

You can show someone how to safely use a chainsaw, but the people who build replacements for or better chainsaws often have worked building, repairing, and using those tools day in and day out.

Education involves repeating things others have done, doing them wrong, and learning from that. When you give someone a tool which just gets the answer without having to understand why, you never have to build the mental model that lets you understand how things can be better.

You see this in software: tools can let you build an app in a day, but if you just use those tools you will have no idea why your app is slower than it has to be or what other things are possible.

When tools let everyone outsource their writing and even their thinking to machines, many of them are not "working faster" - they are skipping crucial steps and acting without understanding.

Tools can allow many people to become moderately proficient rapidly, but when they are introduced immediately they can hinder the development of becoming incredibly proficient.

But are we identifying what the new katas are with the new tools?

I am surprised that "professional headshot" would be a task that can be outsourced to an AI. Even with the various filters available these days, how do you work around the need to actually have a professional grade kit?

At a minimum I would expect you'd still need a professional lighting setup, a sturdy tripod, the right studio background, a good DSLR (or mirrorless), properly diffused, reflected and toned flash, and the skills to really work with RAW images to get the best possible output. And that's even before we include the skills and knowledge to guide their subject to just the right poses to get the subtle body language right. (Yes, the relative position of your spine and the stress you put on your upper body does affect how comfortable and confident you appear in the final shot. I know that now.)

More than 20 years ago, my boss at the time recommended that I get a professional headshot. We were producing printed[ß] booklets as part of our marketing strategy, and I wrote occasional columns in them. He wanted to have the author's face attached to the pieces. He recommended that I pay for the service myself so that the photos would remain with me. If the company had paid for them, getting them into any other (read: off-work, post resignation) use would have been awkward, confusing and possibly a red flag for the accountant. One of the best investments I've ever made. I paid for three slightly varied versions, and used them actively for about a decade.

As it happens, it might well be the time to get a fresh set.

ß: an anachronism, no doubt

There are a number of services that let you upload a few random photos of yourself and will then generate headshots based on them. They're not real photos though: they're just AI-generated. The quality is variable at best though, and many are either obviously AI-generated or don't look much like the person. They might be enough for some people, and they're certainly cheaper than the real thing.

the answer is - you don't need a professional grade kit.

in the past, i used to have to pay more than $10 to take some physical photos designed specifically for passport dimensions. and then, ironically, i have to then scan it in to submit for a passport application.

now? i download a free app with all the right settings that gives me instructions on how to take a proper passport photo, and then applies some filters to brighten it up, remove some imperfections on my face, etc and it looks every bit as good as any passport photo i've ever had. and it cost me $0.00.

the delta in quality between "professional" shots with professional equipment, and smartphones with increasingly capable cameras and software is becoming so small for most everyday use purposes, that the professionals are going to find it very hard to justify their rates in the future.

The parent comment did not talk about a passport shot, but something that you can use in a professional setting to represent yourself.

At the same time, the two photographers in my neighbourhood are doing better than ever. When it comes to memories (like newborn photos) surprisingly many people are willing to pay a premium.

Often the event being photographed has the participants fully involved, they don't have time to think about photos. Hence the need for a professional. DIY has limits, people are single-threaded. You need to hire professionals for extra threads.

My neighbourhood photographer charged me approximately the same for taking a passport photo and auto-submitting it via a quality-checking app as I'd have paid to use the quality-checking app myself...

She also recognised my name from the prints of my baby nieces she'd sold.

I think OP was talking about professional portraits made by a photographer in a studio in a roughly hour long session. Photo shoots like that typically cost 100-200€ and the resulting portraits are not at all comparable to a 10€ passport photo.

Correct. I was indeed.

Although this particular round fell somewhere inbetween "passport mughots at Snappy Snaps" and a full "subject all dolled up" session. IIRC the photo session was about 20 minutes and a week later I paid (IIRC) 25€ for each processed photo I eventually wanted to buy. She burned the selected high-res JPEGs on a CD while I waited.

I'd be happy to pay £120 to £150 for a fresh set these days, assuming they'd serve me for a decade again.

[Funnily enough, we ended up going to her studio for our official wedding photos a couple of years later.]

I see discussion that the OP was about headshots for a different frame of reference, but passport/visa photos are an interesting comparable. Every country has their own standard (dimensions, framing, glasses/no glasses, smiling/not, etc). Although they might not vary too much, it used to be easier for an international traveler to go to a "passport photo" shop (as I did, years ago), but now, as you point out, it's a problem that can be solved with a computer.

Chat GPT, here are some casual photos of me, please produce a professional headshot of me.

What constitutes a “professional headshot” is nothing more than a set of shared norms. Those norms will change rapidly (and have been changing, actually) to include AI Generated headshots. You can already find plenty of headshots that are an AI amalgamation of someone’s selfies. Most of the time you don’t notice until you’re looking a little more closely. That service will only get better and cheaper, and norms will shift too.

I’m calling it now: in a few years the amalgamated style of AI headshots will look desperately uncool. Even lighting, defocused background, too-perfect hair, senseless focus on the wrong details, etc.

All it will take is one trendsetter to get the ball rolling, accelerated by the backlash against scam profiles, and all the current AI models will be worthless for the task.

I have instead opted for a headshot of the goomba from the 1993 live action super Mario bros. Movie which typically gets me requests to change my headshot as people find it disturbing. Perhaps your idea is a better one. Haha.

> Most great musicians, bartenders, software engineers, etc. do the same. Without entry level/apprentice jobs the only other model is patronage from the rich

Or, you know, school? Hobby work? Activities people would engage in of they weren't under threat of poverty or death should they fail to enter the workforce as soon as possible.

> Activities people would engage in of they weren't under threat of poverty or death should they fail to enter the workforce as soon as possible.

And this is going to change, because as jobs are lost business leaders will graciously volunteer to redistribute their wealth

I wouldn't expect it to be gracious exactly, but it's that or... what? Mass indentured servitude? How else would you make a society function?

There's plenty of dystopian sci-fi out there with all sorts of ideas about that.

No need to look to fiction, various human societies were feudal for centuries.

Tech makes it sci-fi though.

That's the question...

Throughout history, it has never been gracious. Power has always flowed from literal numbers and metaphorical barrels of guns.

That's my suspicion

Typically when AI hits human capabilities they immediately overshoot and become superhuman. We're probably going to see a peak in our cultural abilities, both from AI directly and from humans having superhuman teachers to help them in their formative years.

> ... that was never a good model even before today's rich largely abandoned the belief that they had a responsibility to contribute back to the society...

And now to the meat. The rich never believed that! The evidence I've seen was that the patronage system was all about showing off how rich they were. My town ended up with a bunch of really nice churches because all the local landholders were purple-in-the-face having to put up with those damn other Catholics/Baptists/Anglicans/whatever families having nicer churches than they did.

The economy is all about taking base human emotions and using them to power improvements regardless.

Bear with me, I'm serious. When you say "Typically when AI hits human capabilities they immediately overshoot and become superhuman", what AI are you thinking of?

What I've seen suggests that in absolute terms human intelligence is fairly low variability. Officially I refer to this chart [0]. Unofficially I'm usually thinking of AlphaGo as my reference. The Fan Hui version didn't look like it was as strong as the one that won against Lee Sedol, and that version was weak vs. what the AlphaGo team bought to bear a little later online. The sub- to super- human was remarkably fast.

Given the rate of improvement of Stable Diffusion (v1 was amazing in context that it just wasn't possible before, but before around SD v3 I didn't feel like it followed the prompt at all - then Flux happened) I'd expect to see an AI system pumping out Pride and Prejudice-level works by the minute some time this decade and I have no idea what the future of art will be but I imagine pretty impressive.

[0] https://ourworldindata.org/grapher/test-scores-ai-capabiliti...

Very good point. I am convinced the only variation of capitalism that can survive the future is a drastic change to always favouring underdog participants somehow. Specific skewing of rules to help new small businesses, personal choice tending toward local and smaller, government intervention against too big and subsidy for too small and just starting.

(specifically not helping a failing dying business though) but of course all impossible, because would require social credit system that wasn't abused for evil, which is all that human coercion hierarchies know

When GenAI first started taking off a few years ago I called out that the real market set to be absolutely decimated wasn't corporations - it was the gig economy.

The quality of generative AI may not be professional level quality, but it presents an easy and cheap drop-in replacement for one-off tasks that people previously outsourced to platforms like Fiverr (voiceovers, logo design, clip art, copy editing, translation, etc).

I mean, these type of cheap gig work apps already basically destroyed such work for people who were good at it since they can't compete with a guy who will do it for a fiver

I used to frequent rentacoder back in the 2000s. Got very good reviews, landed $20-50 a day doing C++/pascal/opengl homework for the first-world dunces. Then came the horny indian programmers. (we called them that on yahoo! chat because all they wanted in life was to marry up to a US/Canada lady and migrate). They would put in half the effort for 1/10th of the price and my gig was quickly over.

The guy who is good at it and can make a $100 job for $5 will not keep charging $5 for too long

> The guy who is good at it and can make a $100 job for $5 will not keep charging $5 for too long

As long as there is a steady supply of new people willing to do it for $5, it will be a $5 job, even if it would still be a viable job at $100 if there was less competition.

Market helps take care of that!

The real dichotomy will be labor vs. capital as factors of production. AI competes with labor. Capital owns AI. Capital stands to benefit immensely -- that includes corporations. Labor stands to lose -- that includes not only gig economy workers but all people who sell their skills for money.

Eh I give most companies about 5 years before serious trouble

> Capital owns AI

DeepSeek is free and you can run it anywhere.

The price of intelligence is quickly trending to zero.

>The price of intelligence is quickly trending to zero.

Please define intelligence as you see it then, because I see little of the accurate, astute, highly selective and precisely reasoned kind from AI. If anything is caused by AI tools its that the price of semi-coherent filler content (visual and textual for now) is falling to zero. This is far from similar to intelligence.

It's not just $200 tasks. For example, it's really good with producing music... Which would have cost several thousands of dollars before. E.g: suno.com

I've been recording short tunes on my low-quality $50 Ukulele (which I self-taught myself) and then getting Suno to turn it into full songs. The quality is better than anything I could possibly have produced. Even if I paid thousands of dollars to a professional, I could not get better quality than what I'm getting from Suno. I cannot tell that it's produced by AI. AI is surprisingly suitable for music.

If I want lyrics, I type out the topic and underlying message of the song and the aesthetics I want (e.g. short lines, rhymes, etc...) inside Claude or ChatGPT and make it generate the lyrics.

Also once you have the first draft of lyrics, you can make it refine them by telling it to rhyme in particular ways. For example, the first words of each sentence can rhyme with each other as well as the last word of each sentence and you can tell it to change the rhyme pattern each paragraph. It's amazing how it can keep the meaning of the lyrics the same but totally transform the sounds of it by swapping out words with synonyms.

AI video capability is impressive but it can't get it all the way to a production-quality film. For music though, it definitely seems to be production-quality. Sometimes it takes a few attempts but it definitely seems to cross the 'professional' line regularly. Often, you don't need any tuning, trimming or intervention to the piece; it's just done.

Many people see AI as a replacement for the human, but current-gen AI is in my view a replacement for the hammer. It's a tool that needs to be wielded with skill for good results... And when that happens, the productivity can be truly amazing.

Agreed, I think music is one of the areas which it nails though (pun intended). The size of the human input can be very small and still produce very high quality output.

I mean you can produce 0 input and fully automate it and it will still sound professional... Maybe it will lack a bit of creativity but typically only a professional would be able to know that.

I think what Suno produces is often higher quality and more creative than what most modern pop stars produce... Just because the bar there happens to be quite low.

If I listen to 80s music, I can see the creativity gap (at least relative to the AI's typical ouput)... But AI output is often superior to modern artists IMO.

> Maybe it will lack a bit of creativity but typically only a professional would be able to know that.

Judging by the examples on suno.com, no. All of these are noticeably sub-par. They're "professional" in the way that slapping a lens flair png on top of a photo makes the photo look "professional."

> I think what Suno produces is often higher quality and more creative than what most modern pop stars produce

Go listen to any song from Taylor Swift's latest album. I'm not a Swift fan, and I don't particularly enjoy her music, but she's the current face of pop music. Compare the texture of the instrumentation with anything produced by suno's model—it's night and day.

I worry that the novelty of generative models that cater to their users' every whim will create a culture that is uninterested in anything challenging or new. Who would bother to look for good art when an endless fountain of bad art is right at our fingertips? For goodness' sake, you didn't even listen to a pop song before deciding that "most" modern pop artists are worse than "suno.com." I don't even like pop music, but that's just false on the face of it.

Really I think we may be a good ways towards that last paragraph just from algorithms designed to drown users in whatever content they have recently consumed.

(I hate the term "content" because it's so associated with this sort of view, but it's exactly right here.)

Taylor Swift did make a few good pieces but many of her songs are too similar to each other IMO. I can acknowledge that 'Blank Space' is very good even though it's not my favorite genre of music though.

That said, I think there are a lot of great pieces that we've never heard of. Just because something isn't popular doesn't mean it's not great. Especially true nowadays when media is so centralized and people are drowning in content from a handful of celebrities; this is making it impossible for great pieces from non-celebrities to break through...

We live in a time of information and intelligence explosion and at the same time total media monopolization; the best pieces will never be heard. Also, the best work of humanity of the coming century will be ignored.

You maybe have just terrible taste in music.

I'm actually quite difficult with music and I don't like most music. Yet, considering those few pieces I do like, I like pieces from a variety of styles ranging from techno to classical music.

My ability to appreciate a broad set of styles indicates that I am not culturally biased. The fact that I only like a very small % of popular songs indicates that I am demanding. The fact that I don't have any favorite artists indicates that I am not susceptible to cult of personality biases. The fact that I do not listen to niche music indicates that my tastes aren't contrarian; I like popular pieces which other people like but I just like a smaller subset.

So based on this it would seem I have excellent taste.

(obligatory "anecdote" regarding suno: https://www.immibis.com/ex509__immibis_uc13_shitmusic.mp3 )

I thought this was going to be about my wasted 200 OpenAI pro subscription to get operator. It’s failed all its tasks so far.

It's worth noting that spending $200 on a graphic design from someone online was already a huge reduction in cost and time from what it would take before platforms like Upwork

I don't want to sound like I am bashing the original designer of the $200 monkey mascot, but I think the author has some misunderstanding here.

The market value of that was very unlikely $200. Before AI, there were so many people offering similar services on Fiverr. And from my experience they're mostly not scammers (just novices). Of course they might not live in the US, at least not in big cities.

The price range for that was $20~$50.

Edit: the article says 2013. I don't know if Fiverr was popular back then. I'm talking about more like 2019. At that point Fiverr and similar platforms had upended this kind of $200 market.

I think you would see a quality difference between 'a monkey mascot' at $40 and $200. The $200 designer is shining a light on his personal brand.

The mascot is friendly, vaguely memorable, well-proportioned, soft, and not attention-seeking. Its expression tells a story, adds humanity, and creates unresolved tension.

The AI ones are sharp and confident and eye-catching, zero subtlety, completely missing the point. I'm willing to bet a $40 designer would drop the ball in a different, equally bad way (probably make it too corporate, or miss the precise "cute but low-effort" spot the original designer hit).

It's funny because as I read the article, one thought was "the slop results look like they came from very old models"

Use a model released this week and the results are (to my untrained eye) no longer distinguishable from a human artist: https://imgur.com/a/tgEsXq8

And it wasn't some pro image prompting magic. Even compared to 12 months ago, the text encoders have very good. I was able to use wording that came to me naturally and get an aesthetically pleasing result in seconds.

You mentioned a model released this week. Is it Lumina new model by any chance?

Imagen 3 002

The reason AI slop is slop isn't because models are not advanced enough. The slop factor is inherent in the way AI works and cannot be fixed by brute force compute power.

AI is likelihood optimization under the hood. It draws the most statistically likely image. The human brain and eye is very good at picking out "average" pictures. Turns out our in-built AI detection capabilities are very, very good. (This might actually kill the AI industry very quickly. I imagine in a couple years AI-generated pictures or texts will be the lamest thing ever, and AI companies will lose a shitload of money in this arms race.)

a) I linked to an entire album of pictures that 99% of the population would not identify as AI, including the population that knows what AI slop is.

b) This is a gross oversimplification to the point of being unhelpful and pretty much wrong. You don't seem very familar with how these models work.

There is no inherent reason why sampling from the latent space of a model limits us to average of any concept.

Not to mention the models are learning what average means and what exceptional means and can increasingly produce both at will. As they get larger the degrees of seperation between those "sub-concepts" of each concept grow larger and larger.

The reality is humans are good at convincing themselves they're good at things. The false positive and false negative rates are already going up for AI art, and it's only going to accelerate from here on out.

> that 99% of the population would not identify as AI

Absolutely not true about the "99%". Give it a couple years and not being able to spot AI slop will permanently mark you as an "okay boomer" tech illiterate slob.

AI == lame, square, uncool.

We already see this with the AI assistants they keep putting into OS updates. People hate them, and not for "privacy" reasons. They just don't want to be lame.

Again, can you point out to even a single giveaway for that album?

I'm honestly convinced you plain didn't realize the entire Imgur album was AI generated and that's why you haven't been able to address it.

AI is lame, kids hate that stuff. Positioning yourself as some sort of "AI expert connoisseur" will only make you look lamer.

You had 3 days and that was the best you could do?

This comment can't be further from reality. If you show this (https://imgur.com/a/tgEsXq8) to people, their "very, very good built-in AI detection" won't bat an eye.

> This might actually kill the AI industry very quickly

Yes it will, but not in the way you implied. In a few years AI assisted image creation will be so common that no one will bother mentioning "AI" anymore, effectively "kill" the AI hype. Just like you can't sell built-in webcam as a feature of a laptop: every laptop has it.

It's funny because I see this exact same thing in game dev. A bunch of games have now decided to outsource basic logos, banners, icons etc to AI.

They all end up looking very obvious and similar and I pass on them every time. It's THE core advertising for your product.

Supermarket Simulator is a game whose Steam capsule (the very first art players see on Steam) is blatantly AI made.

The estimated sales are > 2M copies.

[0]: https://steamdb.info/app/2670630/charts/

Wonder if it would sell more if it had a good art banner instead?

Would it sell more than the additional sales that were obtained by the work done during the time saved through generating it with AI instead of finding an artist?

Of course it's possible, but 2M copies is already the top-seller in this genre. The original Goat Simulator (arguably THE one game that made "funny goofy simulator" a genre) sold ~4M.

[0]: https://steamdb.info/app/265930/charts

A lot of those simulation games seem rather poorly made, but a couple of them have been big hits regardless.

I'm sure you'd do better on average with not terrible marketing.

I wrote a long while about mergekit. Makes one wonder if it will be possible to just "merge" experience from specifically trained LxMs going forward.

AI probably can do very nice landscaping design too. It's just that image generation models are not made for it. If we manage to give AI a more structured input, you'll get a nice design in return.

I don't think so. Good landscaping requires knowledge of the site, the local ecosystem, and an understanding of how water and soil flow. It's a whole lot more than generating an image. The image is simply an output of the process.

Those are not esoteric things, there’s generally data available for things like climate zones, soil type, plat maps, zoning and height restrictions (for trees). I was a commercial landscaper for a decade and have done my fair share of reading architectural plans as well as design some of my own. This I think it’s doable; it could even put together a shopping lest and check with local supplies for whats in stock and generate an estimate.

Now this is an interest idea, have you had any success?

I have a feeling landscaping design is not common in its dataset, not as commonly talked about online, and so it will struggle with the task. Would love to be proven wrong

I'm more happy that a guy who last posted on his blog in 2013 now posts again because of AI. I just started following your website Chris, now add an RSS feed. ;)

All I want to know is did he pay $200 for that little vector illustration of the monkey?

paying to fix your own bugs sounds like a skill issue

Reply on news.ycombinator.com