The tools are good, but the ideas are not. These tools basically see many people building the same things independently, rapidly, and over and over again.
I'm not seeing a lot of unique applications. I am seeing some optimization, plenty of rehashing the same crap. Lots of thought outsourcing.
Honestly, I'm mixed as someone who also uses these tools. Is it really efficient for 100s of people to be spending tokens and electricity to all make, effectively, the same bad app from their shower thought/idea? I love that we've democratized coding so that it's more accessible to laypeople - but it is mostly all slop and lack of creativity being produced. And, in the rare instances where someone genuinely makes something good, 100x people will then spin their own AI to rip off the same idea.
Where does this all go at the end of the day for humanity when nothing is new or creative and we are going backwards in critical thinking along the journey? Especially since the journey IS the thing for humanity - not the end product of the journey.
> The tools are good, but the ideas are not. These tools basically see many people building the same things independently, rapidly, and over and over again.
Isn't this just every bubble technology since the 80s? I remember being at parties in the early 2010s and hearing awful web site ideas that were certainly feasible with an RoR stack but had a <1% survival rate. The only difference now is less engineers acting as gatekeepers.
I'm certain it was the same way with static HTML webstores in the 90s, social media since early blogs, and other industries that received funding spikes that eliminated financial barriers (construction, film, video games, restaurant chains, etc)
Eventually the Gartner Hype Cycle proves true and the hysteria dies down, but out of those promised elevators of progress come a few incremental steps upward.
I don’t think there was enough money in the web in the static HTML days to have the same bubble. But I do recall domain squatting being a lot more prevalent then than it is now.
This is why I feel like any AI usage is making us all crabs in a bucket when it comes to progressing as a society. Any genuine creativity or effort is immediately stolen and repackaged as someone else's, which inherently means that its incredibly difficult for the original author to get any credit and devalues any genuine progress. And most importantly, people are doing this to deny the author any profit
We seem to be in a situation where 0.1% of people are creating anything new in the programming space, which isn't a change at all, but the difference now is that there's an absolute army of people waiting in the wings for the second they finish to take credit - and profit off their work without compensation. Because apparently that's progress, according to some people
This has been happening the most in realtime with news sites, where any news they publish is fed into chatgpt and recycled to avoid giving them any clicks or advertising money. What happens when all the news sites die because chatgpt ate their ability to survive? We'll go backwards as a society because we won't have news, and then suddenly chatgpt will lose its ability to report the news. Lose lose, unless you're an extreme short termist
It extends to everything with AI, its giving us negative progress as a society, because its society eating itself in exchange for the shortest time line go up
>Any genuine creativity or effort is immediately stolen and repackaged
So pretty much how things have worked almost all the time. For example after the US violently branched off from the UK we stole every idea we could. China operates the same way. Copy and make cheaper.
>Because apparently that's progress, according to some people
According to the people in power keeping information locked up for entire lifetimes is a good idea. They love that kind of progress to stay rich forever.
Yeah sure but now cat is out of the bag and we are in a prisoners dilemma where this inevitable. You can personally not use AI and get your shit stolen in milliseconds either way
The problem is that the tool and the marketing around it make people think they can circumvent the time it takes to refine ideas into a good experience or product.
You cannot, in the end, eliminate time. There are no shortcuts. It's like Euclid said, there's no "royal road" to mathematics, nor to a good product.
LLMs can speed up several aspects of execution, but actually honing an idea, figuring out what constraints make sense, how the experience needs to be tweaked, in brief, design, will always take time. The one shot revolutionary product notion is total myth.
I blame the subsidies and free AI. Can't blame people for building dumb things when it's so cheap and disconnected from the actual costs. That demand then manufactures shortages that wouldn't otherwise exist... The public gets swamped with slop which they view as all AI can make, so they hate the technology. It's the worst of all worlds.
I don't think there is anything wrong with people being empowered to build dumb things, even lots of dumb things. Especially if we believe that there will be a market correction that will force the good things to rise to the top, and everything else to go away.
You're still left with a filtering problem. App review, Job applications, games on steam - any huge increase in volume means it's harder to find the good stuff now.
I think friction to distribution is an interesting thing in this. Pretty unbaked thought that isn't really saying anything, so you'll forgive me: When we were getting deviantART off the ground, folks wanted the user growth to speed up, so we reduced the amount of work to upload things. Then, the quality took a dive and it became too difficult to moderate with the tools that we had, so we increased the friction a bit, over the time I worked on it, the amount and type of friction increased and decreased as we tired to dial in the right amount of growth and quality. It was also interesting what friction filtered out what type of people... add 6 screens you have to click through with drop downs etc, lazy people drop off but creative people get fatigued, add 3 screens but require a minimum of a 200 word description, creative people love it and copy/paste artists drop off, etc. The internet itself has had various versions of this in different aspects of it. I think that is also true in reverse, things can sustain mediocrity through friction. I'm sure there are loads of interesting examples of this all over! Just a thought. :)
I expected tools when animator can just put a bunch of keyframes, sketch the transitions and have AI fill it in, getting all of the artistic vision with massive work saved.
Instead industry went full on dystopia "let's replace entire jobs" instead of settling on just massive efficiency improvements on top of keeping the quality and we got million slop generators and little of the actual good tooling that would change the industry, outside of coding tools.
> I expected tools when animator can just put a bunch of keyframes, sketch the transitions and have AI fill it in, getting all of the artistic vision with massive work saved.
I actually tried that multiple times for a simple b/w icon animation. All the tools of the Big AI labs gave me crap.
At the higher end, AI hallucinations don't hold up. I really do think professional artists and animation studios are using AI at least for rapid prototyping and many are building custom pipelines and constantly iterating until it's good enough for production, or using it in small areas of production (e.g. build a base and manually draw/edit imperfections)
I think if it has key frames and the sketched movement there is very good chance to get it near perfect. And any small problems still there can be fixed after.
On my drive to work this morning I passed by a bus and it had a poster ad on the side for a new musical. Completely uninspired obvious slop. I don't understand why when in the past they would pay better artists and not settle for cheaper ones, they now settle for that gross ai "look". Maybe the average joe / new gen is fine with it? I can actually comprehend outsourcing code to ai more than I can outsourcing your branding / visual identity
>I expected tools when animator can just put a bunch of keyframes, sketch the transitions and have AI fill it in, getting all of the artistic vision with massive work saved.
You expected it to replace jobs, and you're surprised when it replaced jobs?
Life has an abundance boring tedious jobs. Unfortunately the animation in between keyframes is almost more important than the keyframes themselves for making the animation flow correctly. Which is why AI animation often looks so wrong and off-putting.
Quality Control is tedious and boring, Do we want AI to take over there as well?
Actually don't answer that I already know most of the sloperators are in favor.
We got those tools too! Not just diffusion and descedants. Cascadeur is a good example of "tasteful" application of AI, meant to help authors realize their vision, instead of replacing it or just making it as trivial as possible.
> Instead industry went full on dystopia "let's replace entire jobs" instead of settling on just massive efficiency improvements on top of keeping the quality and we got million slop generators and little of the actual good tooling that would change the industry, outside of coding tools.
You could try, collectively, uplifting the voices who saw it coming and called it out clearly in advance. They're the ones generally derided and dismissed as impractical weirdos, though, so that might be hard.
As long as we instead uplift the greedy and craven, we will see greedy and craven results from otherwise neutral technologies.
What you wanted were essentially specialized traditional models. I still have hope that when this bubble bursts and they hype dies down and people have to pay the real price of the do-everyting LLMs, the industry will pivot towards specialized traditional models, which are give much better results for far far far cheaper.
BorisFX does this great. They don’t even call things AI just ML (which is correct) and the product I use, Optics, has this insane masking model that allows you to just paint with green and red colors very crudely (and I mean that) and get almost perfect masks immediately.
These kinds of tools are extremely valuable and worth paying for, compared to Photoshop who has gone all in on generative AI and it’s all terrible, slow and token based.
This really isn't limited to software. A more egregious example is music: we already had every genre, every mood, at our fingerprints practically for free. And now, we're just giving a big middle finger to human artists for no benefit whatsoever. Well, except that Spotify might be able to pay $0.0000 per stream instead of $0.0001.
On some level, I understand where the individual impulse is coming from - "I couldn't do this and now I can, it's amazing" - and I can understand why amoral VCs would throw money at these platforms. But I will never understand why so many intelligent, sensible people are cheering for this future.
Look at all the new book releases on Amazon. They are 95%+ slop, and not good slop. They're made with zero effort. Some of them sell because people judge books by the title & cover. What does this achieve? Once you eliminate all revenue for real authors, what next?
Everyone makes crayon scribbles as a child. What does this achieve?
Everyone learns the same basic curriculum in school. What does this achieve?
Everyone makes mediocre attempts to play instruments/sports as an adolescent. What does this achieve?
Everyone treads the same highs and lows of friendship and romance that have been tread by billions of others. What does this achieve?
It achieves the human experience. Not everything has to be best-in-class to be meaningful, and that's a dreadful expectation to heap upon anything that doesn't strictly demand it.
The things you listed are considered to be smaller version of 'proper' expression in art or mastery of some skills that humans value. Even if most people go nowhere with some of them, they're still seen as commendable things to do and parts of the universal experience.
What is AI generation a smaller version of? While you can make worse and better AI outputs, do you think the average Kindle slop author is going to one day become a proper bestselling writer? Will an average Suno music generator turn into a musician?
"Passive income" grifters spamming amazon with books that they literally spent 0 time or creative effort on are not "achieving the human experience". There is no creative/personal satisfaction or development from this, it's just people making things worse for others trying to make a quick buck.
Difference being people did those things compared to a prompt, there’s no experience to be had from prompting “make music” in relation to trying to play an instrument or lay out a track.
Playing someone else's music doesn't teach you how to compose. Laying out a track in fruityloops doesn't teach you how to play the instruments.
Complaining about negative space is easy but not very informative because the attention doesn't go missing, it gets reallocated to something that someone found more engaging. I remember the vitriol aimed at DAWs 25 years ago, but it was obvious then and correct in hindsight that what was lost in forced mediocre specialization would be more than made up for in expanded capacity, and that anyone who wanted would still be able to drill down and master an instrument, but that they would now be in a position to determine where this would be most impactful. It's the same fight all over again.
Unpopular (and unrequested) opinion:
I still think that the world's best DJ mixing music live on stage in a club is not the same class of job as an amateur pop band playing in a pub in front of family & friends. The latter being much more meaningful.
Unpopular, unrequested opinion: band kids messing around in FL studio after practice are doing something far more creatively meaningful than the activity the adults organized for them and recognize as virtuous.
Learning how to compose and play are two different skills though, they are likely to reinforce each other, whereas a prompt is just something completely disconnected.
It wasn’t that long ago people were thinking prompt engineering would be a profession of the future, and now you barely need to be able to spell to get the same output.
If you think we have every kind of music people want, then you lack imagination.
I for example generate catchy music that helps me remember topics. Those are hyper personalized.
Even outside of hyper personalized music, for the styles of music I like, something like 90%+ are topically centered around love/lust and I prefer more topical variety. So I make them.
AI books on Amazon will eliminate revenue for authors in the same way that YouTube eliminated all revenue for television. Most of YouTube is absolute garbage and even the algorithm recommending content is garbage. Yet "real" video producers that make good content still make an absolute killing. You make good stuff people will be attracted to it. The much bigger threat to authors is that people just don't read anymore.
Democratizing coding via LLMs is democratizing cooking via frozen foods. It's gonna be unhealthy to the end user despite the convenience. And the taste level will not be there. Sure, you "made" dinner just like you "made" an app.
There are startling parallels today to how news and information was 'democratized' when social media emerged in the mid to late oughties. The issue is the same: once it's cheap and easy no one seems to care about how healthy it is.
I can't write more than a few prompts without the LLM (any of the latest SOTA models) starts doing complete absolute nonsense.
I do agree that the ideas are not good to be clear, but the gap between the promises of what LLMs can do and what they can actually do is huge.
Anyway, it's funny to see the insane amount of astroturfing on HackerNews and how people try at all cost to make it seem like the technology isn't the problem but humans actually are. Literally every top-level comment is about that, or downvoted.
Original ideas are not "allowed" in a sense that the infrastructure will fight or directly forbid you. Your original ideas will have hard time to go through Apple AppStore's review or Gooogle's Android or Chrome processing. You will have hard time getting paid for it, you will have hard time to make people hear about it because x, FB, TikTok, reddit etc all have an agenda and they don't spread things that don't fit it(AKA medium is the message).
We don't live in a free world where you can just do things anymore, you are supposed to fit the machinery that is churning hundreds of billions of dollars every year.
I scanned the video and transcript and it argues (poorly, imo) that competition produces parity and mediocrity.
I also don’t really see how it applies here — AI in this case is removing what most people would understand to be competition. Do people driving cars still compete on running speed?
> AI in this case is removing what most people would understand to be competition. Do people driving cars still compete on running speed?
We're not talking about business competition, we're talking about human competition. Species level, subconscious, and driven to seek (perceived) safety in the hierarchy amongst other humans.
It is interesting I think a hard societal split is emerging between those who understand AI is a tool that can be used and can use it, and those who embed themselves in the view that AI itself competes with them as a person.
Sure. Take a look at the current landscape. What do you see? A lot of copying. How many companies do we need building harnesses that act, look, and behave the same? How many clone apps with the same exact design and aesthetic? How many half-finished games that are just created in order to copy the success of someone on X? How many genuinely novel, quality ideas go unnoticed, unrecognized, or flat out ignored because they don't align with the consensus?
Humans are conditioned to be mimetic: seek an authority, seek someone more knowledgeable outside yourself, but never, ever, ever learn for yourself or do for yourself (especially when doing so goes against the status quo). In practice, this leads to an endless sea of copies that delude themselves into thinking they're unique ("me too" apps, etc). In other words "if they're working on this or doing it this way, it must be important/safe for us to do it this way."
Most people are terrified of being original and authentic and so default to the "short blade of grass" direction of "we'll compete on the same terms as everyone else so we can fit in."
It took me several years to fully understand this but once I did, I saw/see it everywhere. And practically nobody you share this insight with is comfortable with it which tells you it's closer to the truth than not.
Edit: thank you for showing legitimate curiosity and asking me to explain instead of treating me like an idiot.
Just for entertainment in the last month or so we've had a Nintendo Switch 2 emulator and a Playstation 5 emulator that have grown by leaps and bounds. I'd usually put an estimate at 2 years before we'd have these things working properly now it's weeks. But the reason you're not seeing a lot of unique applications (though they are out there for sure) is because people aren't sharing them. I make a bunch of stuff that isn't good enough to distribute widely but serves my purposes just fine. Are they 'unique'? Maybe not entirely, but I haven't been able to find apps that do the same things in the same way or I would just use those. It's really nice to have software that works exactly how you want it and that you can tweak any way you want with a prompt.
The industry is just busy overall with day to day work and it just takes ages to implement things.
Which also means that it is slow in adoption not because the quality is not here but because you need to do things first.
You need an env to run an agent, you need connectors, you need to get that stuff through security etc.
The only question is timeline not investment or bubble because AI as is will not go away anymore.
I asked before for headcounts or consultants and today I can just use AI for it. Its not the AI which is the problem, its the stuff i mentioned above. We will probably still need a few month in my team alone to close the loop.
And whlie we close the loop, models are continesly getting better.
I don't think anyone is suggesting that AI isn't useful - that's been proven. What is being questioned is the business model and whether it's worth the investment necessary to train huge frontier models. How does one get ROI when models are continuously improving and being surpassed and people can distill your weights with a fraction of the cost and resources? How to insulate against the threat of local open weight models as hardware improves?
This is why it didn’t made sense to make OpenAI go private, instead of leaving it as a non-profit. If there’s no moat around building a dense frontier model, and others (like Google) already have a way to monetize a LLM better than you do, and yet everyone benefits from it, then it’s best left as a shared good.
Moats take many forms. In the early days there were many search engines but really there is only space for just one to efficiently make money. Being the last one standing builds a natural moat.
I will replace one person in my Team with AI. Thats one full time salary.
I'm also sure that we will forced to leverage AI a lot more in the next few years which will increase token usage even more and it will be a lot harder for us to get a headcount or consultants.
Open Chinese models have the risk of being manipulated and they are distilling Frontier.
HArdware and compute is an issue, you need partners you are allowed to use (export bans, national security) and who can provide a certain qualith/uptime and trust. After all you are sending all the private data from everyone to them.
the idea that a bubble means that when it pops AI will go away entirely is silly. the web was a bubble. we still have a web, and it's still useful, but a lot of people badly overinvested or invested in capabilities that never manifested and when the market corrected the price of those investments the bubble burst. same with real estate in 08-ish. A profitable idea drew more investment than it can support, prices went artificially high, then they came back down and we still buy and sell land.
Gotta setup your agentic software factory to generate requirements to feed into your factory to build your software that then needs CI/CD pipelines to deploy to production which requires more requirements for the factory to produce a working IaC repo against your repo and…
> The only question is timeline not investment [...]
This is not the only question.
As my own organization has rushed/struggled to implement AI workflows, I have seen this same sentiment in many places in slightly different forms. The inevitability of it. The obvious benefits. "The revolution is here." Ok. Except why is it always around the corner? Why does the investment keep growing but the returns keep pushing to the right? Why is the answer to all the problems just "more AI"? From what I can see, it is very much in question that these intelligence products can ever integrate (in their current form) into our organization in a way that pays dividends on the investment. I'm really not confident that the current products are going in a direction that will ultimately help us. Instead what I have seen is groups of users that believe they are much more productive, but their numbers don't actually reflect any improvement.
I mean it took our team alone until now to start the proper implementation for the avg corporate reason: Too much other things to do.
But the result of our PoC were very very good and we are doing this now and it will be great. But our case is analysing log files, incidents, pre analysis of issues, code review, security code scanning and team rules.
We also have to restructure our repositories as a monorepo now makes more sense than before.
I will replace one person in our team because its a lot of effort to teach that person stuff and explain stuff to that person.
AI is better that this person and whatever I teach the AI, it will not quit and I can finetune it.
This isn't "AI is useless" or "AI is dead". They are saying it's not worth the cost.
Further, I would argue that coding is the only place it's made solid headway and even then it either scaffolds to less than perfect, or the code base grows to be unmaintainable. Not saying you can't ship a product with it, just saying that while you are moving fast and breaking stuff someone else is shipping a solid product marginally more slowly.
Last point "continesly" is not a word. Continuously maybe?
Recently, I tried to automate a tedious aspect of my day job (renaming scanned check/invoices using information from each) and after a while, Microsoft imposed upload limits which reduced the number which could be processed each day to 80--100.
I've tried to ask how much it would be necessary to pay to remove that, but haven't found an answer with a solid price.
I managed to get a local LLM running and processing such files --- it took 405 seconds to process two on my work-issued laptop (by hand, with the aid of a batch file cobbled together from a spreadsheet I can do ~200/hr.) --- I need to do almost 5,000 of these for Q4.....
I haven't found one which was reasonably priced and which would cope with handwritten checks and invoices which ranged from crisp printout through rough fax through inkjet print of photo of only partially unfolded invoice with a reasonable error rate.
If you know of an opensource/free tool which does, please let me know.
Not sure where "signature" came from --- if you're concerned about PII, then I have the account line covered over w/ a pen when the scan (really a photo) is taken, and everything on the invoice is public record, and well, identifying information on checks feels pretty non-private to me --- it's not like anyone still puts their SSN on them as was common when I was young.
The renamed files are then opened one-by-one and the filename verified as the check is entered into a system which tracks payments, so the ~8% failure rate isn't a big deal and is corrected as things are entered.
Everyone I know who isn't anywhere near code uses AI daily. Law, telecommunications, real estate,. Some more effectively than others, law seems to be getting huge bump specifically (GLM law, Astra for law). I'm of the opinion if you aren't reliant on it now you will be soon.
Does it cause problems? No doubt. People get way in over their heads with AI when they don't know what they're doing. But when you know your field and it's simply a conversation about busywork that needs to be done it's better and much cheaper than most assistants you would hire (unfortunately). It can save 10s of hours a week easily.
I find it fascinating that you could hold this position. Pick a single popular SaaS today and you can duplicate it in a week. Of course that's not the only hard part of running a business, you need to market it and sell it. But that's millions of dollars and months to years of work reduced to nothing.
Even marginal labor replacement in knowledge work leads to pretty significant revenue (not profit) figures split between the labs. I do think diffusion is relevant. It takes awhile to find product market fit and for new ways to work to become common place. We've seen this with most revolutionary technology (usually on order of decade+) and it's only been a year or so that these things have been good enough to replace some labor. I think we'll see incumbents who are slow to move on this get replaced by new entrants with much better margins.
Not to say the hype isn't fucking annoying and often bullshit.
The difference is that the Mona Lisa is art, created by a human. People generally value art because of the human factor - the artist was trying to say something, there is something interesting about the piece, etc. This is part of why generated images get so much pushback.
The same cannot be said for SaaS: your business needs certain productivity tools in order to operate, and the only options tend to be these hyper-expensive SaaS options whose UX has been clunky since 2005. Historically, you could not build these in-house because it's more expensive to hire a full engineering team. Today, the math being done is whether it is worth it to build something 90% as good for 1% the price of the SaaS.
Generally, enterprise contracts with the AI assisted coding companies have protections against training on the inputs.
You or I probably can't fight it if they lie on their terms of service. A Fortune 500 would go for blood (metaphorically speaking) if their IP was violated by one of these companies.
> In a few years it will replace tham
There isn't anybody to replace in this scenario. You're either buying the SaaS, hiring a full engineering team you didn't have before, or slop coding it. But if you're already on a SaaS, someone higher up has already made the decision that you cannot increase headcount to support something internal.
Personally, I'd prefer if they hired more people. But the cold math of your average business is going to choose to ditch the SaaS in favor of the cheap slop job.
Clones are not frozen in time. If you've already slopped out some vibe-coded abomination, you can just modify it again.
The cost differential is so great when it comes to B2B SaaS that you could generate a brand new every month and still come out ahead.
I know there is probably some axiomatic "AI Bad" at play here (I avoid vibe coded solutions as well), but it's myopic to claim that we aren't at a point where the traditional B2B SaaS model is under serious risk.
Costs to maintain many of these items aren't that much different than self-hosting open source. Every week on HN you hear someone saying something has been enshittified. Well then just maintain your own shit with half the spend.
We used to buy software on floppys. It provided value without a sub. Pretending that every single software solution requires a 50mn operation is silly. I suspect HN tends to think of problems framed at massive scale.. but most people just ain't Google.
Yep, software rusts. And when something like, I dunno, a popular self-hosted BI tool has 1000s of customers with 1000s of more needs then of course you'll have an update every week. They also have to stay ahead of their competitors with features. But when you're supporting 1 customer on 1 platform and have a choice of dependencies the update cadence ain't going to be nearly as bad is it?
A frontier model has no problem keeping a VM and a list of dependencies up to date. When one of those dependencies goes unmaintained it can fork or rewrite it and do the same there.
> I suspect HN tends to think of problems framed at massive scale
I think this is an industry-wide problem. People (especially managers and juniors) tend to read blogs or watch talks by FAANG discussing how they saved millions or achieved 5-nines uptime, and go on to cargo cult it.
There seems to be a perception that everything needs a sub-3 hour recovery (plus on-calls), when realistically, the business will go on just fine if some QC tech can't log their passing test for a couple days (the plant manager makes a call to ship it now and update the record later), or a project manager can't update their portfolio until Monday (they can just message someone if a status update is immediately needed).
Yeah, some things need resiliency. But a key aspect of engineering is balancing the complexity and cost of a certain design, with the risks of that design failing.
> Pick a single popular SaaS today and you can duplicate it in a week.
This is a common claim but I have not seen it played out in practice.
The most well-known example people cite for this is Photocraft[0] which went viral a month ago; the claim was, at the time, that it was 80% feature complete upon release, and people took this to mean that Adobe was dead. The creator claimed that within a month there would be feature parity with Photoshop.
Currently, their feature parity tracker[1] has them at 43%; a whopping negative 37% improvement in compatibility. Yes, this was a change in measurement, but by their own admission:
> the 2026-10-03 estimate (82% "of the way") [...] counted feature surface and was far too optimistic
The impending "SaaSocalypse" was predicted earlier this year, but never materialized, and you can see it in the stocks of Atlassian, Salesforce and others. They're getting hammered in the past few weeks as with all other tech stocks, but if you put all your money into Atlassian in June you would still have tripled your money right now.
When I saw Photocraft for the first time, my thought wasn't "Adobe is done" but "wtf this will change software as we know it in the next 5 years".
Then the whole reingineering of firmwares and fixing bugs in software you don't have the source code for. Was it possible to break software before? Yeah sure but even me i was too lazy spending weeks on disassembling, analysing and doing it if I was able to throw a little bit of money at it. Now you don't need days or weeks anymore, you just let it run and get back to it to see its solved.
You will not see a normal company selling software as non SaaS solution. You will see companies protecting themselves and closing off APIs etc.
And you know, the more people are jobless and more young people just using these tools without any second thought, will fill all the niches which are now accessable because of AI. That small company somewere in a more rural area building and operating some small tool? Someone else will vibe an alternative.
I already here stories around my non it bubble of non it people vibecoding themselves tools they would have paid before. On TV an architect said "i vibed our hour tracking tool, i also did this and that". This alone makes software so much more accessable.
Of course. As I said it takes time. Writing the software is different than replacing an encumbant. A new software company starting with AI simply has less labor costs than any time in history. Claiming a competitor starting from the same place without AI is marginally slower is ridiculous.
Well you said it takes a week. Which is some amount of time, to be fair, but short enough we would have seen a major SaaS rewrite by now.
> Writing the software is different than replacing an encumbant. A new software company starting with AI simply has less labor costs than any time in history. Claiming a competitor starting from the same place without AI is marginally slower is ridiculous.
Of course, but while you're using AI to build e.g. Jira from scratch, Atlassian is also using AI to add more features. They may be niche features that most users don't care about like "scheduled immutable backups for compliance reporting" or "logging Confluence searches to internal employee monitoring software" but they're the features that enterprises are paying for.
> Well you said it takes a week. Which is some amount of time, to be fair, but short enough we would have seen a major SaaS rewrite by now.
You're confusing things. Writing the software with replacing an encumbant. I also said diffusion typically takes on the order of a decade. I absolutely don't think you're going to replace encumbants like intercom or zendesk in a week, despite the fact you could write the software relevant to your business to replace it, in a week. If not only for the fact that someone has to want to do that work.
To be completely clear, I agree with the premise. Overhyped product under delivering is nearly a tautology. The issue is with expectations and timelines.
> Further, I would argue that coding is the only place it's made solid headway
Why? The company you work for seriously doesn’t do anything productive with LLMs?
How about search? Every big company I’ve worked for has some shitty enterprise search. LLMs have massively improved that situation.
They’ve also almost completely replaced L1 support at my company, the numbers there are way down just from people being able to find information they need. I don’t have broad numbers higher level support, but for a couple niches that I know about they are down as well.
For example, in a global company of 100k+ employees you’d be surprised how many support requests HR got just for PTO related queries, not even real problems. Now we have an agent that can fully manage your time off, has the policy documents for every country, pulls your data from the HR system, their number of support tickets are significantly down.
Another project I worked on involved automating some risk analysis stuff as part of the IT process everyone has to do when they develop or procure anything. After developing it for a couple months, then a few months running human in the loop validating all of the assessments, we have analysts manually audit 5% of them now. All of this used to be a shitty excel process, now is a slop web app. Cycle time is down from weeks to minutes because nobody has to wait on reviewers and the app can prefill most of the data from your general system documentation. Overall everything is just as or more accurate,
I know these aren’t very glamorous, barely considered real work by some, but in my feeble imagination there are many such processes being automated and improved everywhere.
> How about search? Every big company I’ve worked for has some shitty enterprise search. LLMs have massively improved that situation.
Using Glean at work and it's not much better than bad searches. It frequently gives me outdated information, or uses questions I asked in slack as answers to the same questions I asked in the LLM.
China seems to always have viewed AI as "normal technology". Maybe they know that they're falling behind and won't hit RSI like Anthropic thinks they will, or don't think that it's possible at all. Also notable that Baidu/Xiaomi don't have any frontier models, so it's a bit like Samsung commenting on the AI race
It's not about frontier models at all. They're commenting on AI as end users. They are saying the returns from the use of AI do not match the hype that has been created by the frontier labs.
So you are saying Chinese firms are now going through the sour grapes phase.
I am thinking maybe your blunt take is right. It is true their primary objective with distilling western models and open sourcing stuff has seemed to be to undermine the US economy. Cold war tactics, instead of running a normal business.
Otherwise per their own admission why would they invest huge amounts, that too on student models.
> Cold war tactics, instead of running a normal business.
There's a bit of a double standard here: When a Chinese company does it, it's a cold war tactic. When an American business does it, it's capitalism. China engages in some shady shit, but this is not a case of that IMO.
Let's not forget that Meta founded the open LLM industry when they released Llama for free. Microsoft, Huggingface, Google, and even Amazon (all American companies) have released their own.
The business strategy here is to disrupt the closed-weight leaders (OpenAI and Anthropic) by delivering an LLM that is good enough for day to day use. This commoditizes AI, driving the cost down and shifting the business model to providing off the in infra.
It's no mistake that the companies releasing these open-weight models are by and large cloud providers.
> their primary objective with distilling western models
This is another double standard: it's simultaneously wrong to train on the output of a closed model without permission, but somehow okay for the closed model to have been trained by reading everything that has ever been written without permission.
The difference being that there is a Politburo officer sitting on the Alibaba board. The amount of direct control the Chinese enforce on their businesses to ensure their goals are aligned with the state goes leagues behind what's done in Western Democracies.
Sure, but that still does not change the fact that everyone who isn't OpenAI or Anthropic are trying to commodify AI.
It's completely irrelevant here. It would be just as logical to try and tie them drinking water to some conspiracy to undermine Coke, because some Chinese public ad campaign encouraged people to drink healthy options.
I agree with you that some of the most outlandish attacks (especially the ones directed at open-source/weight) come from US companies that don't appreciate the competition but don't be fooled for a second that just like the US with its chip exports controls, China is not aggressively trying to undermine US interests.
> China seems to always have viewed AI as "normal technology". Maybe they know that they're falling behind and won't hit RSI like Anthropic thinks they will, or don't think that it's possible at all.
If a leading AI company warns that there is a bubble, perhaps people should listen?
The likes of OpenAI and Anthropic have an interest in keeping the hype going, as losing the hype poses an existencial threat. They are highly unsustainable as-is, and losing momentum won't make that problem any better
> Also notable that Baidu/Xiaomi don't have any frontier models, so it's a bit like Samsung commenting on the AI race
Nonsense. The ugly truth is that nowadays the cutting edge frontier models might unblock niche applications, but the cheapest models in the Pareto front are already handling the bulk of the prompts/sessions.
And nowadays the only reason why cheap models such as GPT6 Luna are on the Pareto front is due to the fact that OpenAI rushed to slash it's costs in half right after it's release, in a desperate attempt to be close to the Pareto front and remain competitive.
I'm confident that once hardware costs return to normal, the bulk of heavy AI users will jump head first to local models and US companies will face an existencial crisis. There is a reason why US government officials accuse Chinese open weight models of being economic warfare.
This is a bad take. The China market is different.
Baidu and Xiaomi don't create models. That's handled by other entities in the China market.
Baidu and Xiaomi experiment with models. Test the efficacy of different models. Get an idea of the relative economics of different models. Maybe create a model of their own to see if they can improve the business case in one dimension or another.
What they're saying is that, in their role as customers for these models, they're a bit confused about the business case. I think that's a fair statement. They've presumably put a lot into finding out how to make money out of these models, and presently, the returns are not there with respect to the bottom line. Additionally, their experiments have given them enough background to feel comfortable asking, openly, where exactly everyone else feels the returns will come from?
For them, yeah, just keep the little experiments going and don't go all in until someone offers you a model that can answer questions for less in cost than the person positing the question is willing to pay you. Maybe stay in verticals where the questions are valuable enough to be worth the cost for right now. Medical, Finance. And limit exposure to the larger market of less valuable questions: "ChatGPT, where is the closest pizza by the slice?"
>Maybe create a model of their own to see if they can improve the business case in one dimension or another.
All done in the context of experimentation to improve economics not intelligence.
Baidu and Xiaomi are trying to figure out how to make money. Not how to make models. Making models is not their business and they don't want to be in that business. China is full of dime a dozen PhDs who can make models. They don't want to do that, they want to make money. That's the point.
They want to understand what are the conditions required for model providers to offer them a model, any model, that performs in an economic sense at scale. That may require creating a model or two as experiments, but that is not at all an indication that they want to change their business to model creation. Nor even get into the business of model creation.
Model creation is just not a good business for the likes of Xiaomi, Samsung, Apple et al. If they have to, fine. But it's clear from their statements that they really don't want to because it's, currently, a terrible business.
The problem isn’t AI, it’s capitalism. I use AI constantly and have built my own personal universe of software. Instead of buying a solution, I build it myself. Instead of using open source, I build exactly what I need and nothing else. Instead of using a library and introducing a dependency, I just use AI to build it in my own code base.
This has been revolutionary and game changing for me, but what it’s not is profitable in any way. In fact, quite the opposite. I also don’t even bother open sourcing or distributing what I build. It’s my computing dream, but it is just for me. Capitalism nor our idea of open source fit this model.
Not OP: GPT-5.6 made a XApp tool that lets me add or modify rows in a Google spreadsheet I use to track time working. It has a menu that lets me start and stop timers, edit the most recent row (in case I realize I forgot to start or end), and open the spreadsheet. Simple but probably not something I'd build myself.
I also have an XApp that when clicked lists recent messages received in a private messaging app that has an API and CLI but no easy way to see if there are new pending messages.
I used Claude to write a tool that lets me plan rebalancing portfolios across multiple brokerages. I was doing this by hand for a few years and I sure hated it. It's not smart about taxes (I still have to use human thinking for that part) but it reduces the effort quite a bit.
None of it is revolutionary stuff, but no software I write for myself will be revolutionary. Must it be, for it to be worthwhile? It helps me out and that's good enough.
I saw your other reply and I get the impression you wouldn't like anything I'd design. That's OK.
I've made a AviSynth/VaporSynt image manipulation program which runs on WebGPU in the browser. I wanted a script-like experience for reproducibility (apply exactly same transformations to several images) but also interactivity for fooling around with.
I made my own astrophotography camera controller for my Canon camera. It runs on a Pi and has a browser UI, so I just need the phone and the Pi. It does autofocus, helps reframe the same shot on a different day, polar alignment assistance, and sends data to Stellarium so I can see exactly what's in frame when back inside.
I made a true-color version of the Quake II software renderer[1], I've always preferred the gritty look of the software renderer over the GL renderer, but not the poor shadows and such which comes from the limited 8-bit palette. So now I'm playing Quake II the way I've dreamed of all these years.
I've made a Rust-based implementation of JPEG XR, not to to be confused with XL, so I could more easily convert OpenEXR images to and from JPEG XR. This format is what Windows requires for HDR background images, and existing tools didn't handle color spaces right.
I've built a family recipe cookbook taylored to our needs to eliminate any pain for weekend groceries and evening cooking. No more slow Airtable, no more shitty bloggers storytelling around the recipe.
I've also a wine management app, with an integration to the cookbook one so I can track which wine I've had when and on which meal. And where I can find in my two wine cellars the kind of bottle that would match with tonight's donner.
For work I've built a stand-up tool that display exactly what we need to prioritize work, on top of the ticketing SaaS tool we've been using for years.
We don't need to settle anymore for someone else product decisions and tradeoffs, we can build made to measure tools.
i'm building a wpf text editor that fit my work flow with agentic coding.
i am a long time 10+ years Jetbrains subscriber. i tried Jetbrains's new ai product and it just doesn't work for me and the IDE is becoming less important for me. i only use it to review ai written code. i do not plan to renew my Jetbrains subscription.
Could you elaborate on how you did the image editor port?
I'm wondering if you need a 200 dollar monthly subscriptions to get enough tokens for "sane image editor for linux" and what harness and model you used?
Also did it require days of human testing and refinement or were you able to automate that part with A.I.?
I used a claude pro subscription back in the opus 4.x days. If the usage limits filled up you just have to wait until next week. I wasn't in a hurry so this was fine. For testing, I wrote (this was hand coded by me since I found that part mentally stimulating) a couple of scripts that the clanker could use to drive the image editor headlessly: A headless sway instance, a daemon implementing the wayland virtual keyboard and mouse protocols that accepts commands via an unix socket, and a couple of bash scripts to tie everything together. Then I gave the agent instructions to use this setup in claude.md, the basic command set is `./headless.sh {start COMMAND|screenshot|mouse move X Y|mouse click X Y|type STUFF}`. My first attempt was using accessibility APIs by giving the agent access to pyatspi2, but there were a lot of things that didn't work properly. Using input mocking was far more reliable.
So the development was done with opus, the interactive testing was done with a sonnet subagent for cost reasons (cheaper model & doesn't inherit the long opus context). Most of this testing was completely automatic, sonnet caught most bugs itself. I still had to test everything manually however because headless sway doesn't behave exactly like a real DE with kde's platform theme, but this was just a cursory check. Most of the work was me specifying how I want every tool and panel to work.
Many things from command line tools, to web services, to full blown gui environments. I do it to explore ideas and/or solve problems. Once I started, I found it incredibly powerful to control my own release cycles and feature development. I also found I don’t actually need a good chunk of what most products do.
It’s funny because they say the first 80% is easy, but the last 20% of polish to get a shippable product takes most of the time. Well, if you just build for yourself and never ship, you can just crank your own solutions out at hyper speed.
Do you really believe everything people say on the internet?
The claims made are overly bombastic and quite hard to accept at face value. If you have used any AI tool to build anything non-trivial you would know how long it takes and the money involved.
Generally, it is good to be skeptical when people make claims like these particularly when it comes to AI. Looking at their profile, you see that "slowin" was created just over 6 months ago but has already amassed over 2.5k karma points! This is a red flag that this account is just "karma farming" pushing AI and similar other popular subjects in a concerted group.
Solving problem isn’t a needless whim. And such solving can range from fully specified software to hacky scripts. But playing around, like rebuilding 90% of vim, is needless. Because why not just remove the 10%?
It’s kinda like few really attempt to reimplement ffmpeg. It’s just easier to disable all the codecs you don’t need than to reimplement the codecs from scratch.
I don't understand why you see this as an "anti-AI goal post". It's precisely because AI is powerful at building software that we can no longer assume software is motivated by something beyond a needless whim.
Capitalists are known for pushing their customers into unhealthy over consumption to maximize profits. I would argue that 99% of all vibe coded software is nothing but throwaway toys, that gets used a couple of times and thrown in a pile of other unused toys which is quickly starting to take up all the space from your room.
Everything needs to exists in the "capitalistic environment". People, organizations, healthcare, church, software, activism. It's the environment we have.
AI is capitalism. It is build by stealing labor, the models (i.e. the means of production) are owned by owners of capital who hand it to us if we pay them, i.e. rent seeking. If I am wrong, then AI is feudalism, which is not any better.
Have you seen the prices of gpu/memory lmao? Virtually no one can run anything remotely useful locally and won't be able for the foreseeable future.
There are maybe 100m people worldwide ready to pay $20 a month for a top of the line llm, no one will drop $2k+ on a machine to do the same at home with a gimped local model which brings 1% of the value since it won't be tightly integrated with anything
Poor people didn't usher in the computer age or the personal computer age which later branched off of mainframe computing there were plenty of people computer users, students (Steve Jobs, Bill Gates and others for example that ruined IBM's big iron dreams) and there are many small/medium companies today that can and they will push forward irrespective of Sam Altman's ram declarations.
I would put a finer point on that, and say the problem is specifically highly-tuned financialized capitalism. Where "highly tuned" means that a LOT of resources are cranked to N degrees of precision about making money from calculated risk, where the actual value of those risk calculations is largely bollocks.
I'm not throwing shade on the entire risk field, but I AM throwing shade at betting trillions of dollars on what you think a tiny fraction of future costs might be going off into infinity. It's what made image sharing or bond insurance appear more profitable than asteroid mining or solar power, and it's the leprechaun promising the pot of gold under the AI rainbow[1] for financialists.
This happens to be a puzzle piece that fits pretty well together with several well-smashed shards of the Old European Right, but that's another subject and much more dubious. The "financialist thesis" of Western Decline is on much more solid ground - you can see it with graphs and shit.
[1] I.e., eliminating all future labor costs forever
In that sense, plagiarising was always part of programming, a smart programmer would be able to analyse and lift parts of code to be remixed and re-applied.
This moral ship has sailed, as much as you and I would like for it to not be true, if this is the world we need to live in I will do as OP and use these tools to achieve stuff that wasn't possible with my free time before. In that sense I think it's an accomplishment to be able to generate code for ideas I had but didn't have the time to properly code them.
I'm not defending it at all, I'm with you on the morality of it but this cheap shot about plagiarising is not a good argument as a counter. Software has always been plagiarised, that's not the moral issue with LLMs being trained by gobbling any available data.
Would be nice if you don't use a throwaway to write low-effort comments though, that's also quite cheap and weak.
I mean, yeah. They're deeply connected to the Chinese state and the Chinese state is trying to undermine the American economy, which has stupidly put a lot of its eggs in the AI hyperscaler basket.
It makes total sense that they'd warn of this. They're also somewhat right. Not for the right reasons, mind, but they're right.
It doesn't seem China is very agressive towards the Western economy? I'd argue it's much clearer that America is acting agressively towards China's (and other's).
They remember the Opium Wars, Boxer Rebellion and Japanese occupation and they have the pressure of 1.3 billion people living in a area the size of the United States from the Atlantic Ocean to the Mississippi river with a massive barren plateau above 10,000 feet. Imagine 1.3 billion Americans living/restricted in the same space that would make you plan faster.
This all comes down to execs looking for a reason to be relevant. Just like every single "revolution" in software development, the executive set thinks that this is going to be the innovation that allows them to finally hire fewer programmers and deliver better software than ever before, and it's never been true, and won't be true this time.
AI has become an incredibly useful tool for competent engineers to build more, better software. Is it orders of magnitude more impactful than, say, the widespread introduction of OOP? yeah, probably. Is it still the same kinda thing? Yeah, absolutely. You can do some _incredible_ stuff as a solo developer with a good coding agent and good domain knowledge these days; as we speak I'm working on a complete astrophotography toolkit that can run off an android smartphone, and this is possible mostly because 1) I already knew enough android to reason about the ecosystem and 2) I know enough about astrophotography to reason about the work that needs doing. Domain knowledge is what separates AI slop from good software that happens to be written using AI, and you can't just magick that out of existence, no matter how much the pointy haired's want it gone.
I don't think we're particularly close to an AI system that can fully substitute for human taste, and if we stop pretending we can already have "fully autonomous AI employees" and instead see AI as a new paradigm for building software in the hands of professionals, we have a good chance of actually changing the industry. But stupid hype cycles are gonna stupid hype cycle, and it'll probably take a few more MSFT-level crash outs for us to get to a level of sanity around it.
I'm more likely to attribute this to the layoff culture that swept through all tech related businesses after Elon showed Twitter could more or less continue running after laying off whatever huge percent they did.
they're already doing damage control by saying "the code was written by a human" which to me sounds so damn funny, you can tell the ai boosters are scared of reality catching up
What do any of those issues have to do with AI other than they happened loosely in an environment that contains AI and a bunch of conspiracy theory "it's obvious bro" connections?
they’re happening in an environment where companies and employees are claiming that ai writes all the code and is often handling the entire process through so-called “software factories”.
The tools are good, but the ideas are not. These tools basically see many people building the same things independently, rapidly, and over and over again.
I'm not seeing a lot of unique applications. I am seeing some optimization, plenty of rehashing the same crap. Lots of thought outsourcing.
Honestly, I'm mixed as someone who also uses these tools. Is it really efficient for 100s of people to be spending tokens and electricity to all make, effectively, the same bad app from their shower thought/idea? I love that we've democratized coding so that it's more accessible to laypeople - but it is mostly all slop and lack of creativity being produced. And, in the rare instances where someone genuinely makes something good, 100x people will then spin their own AI to rip off the same idea.
Where does this all go at the end of the day for humanity when nothing is new or creative and we are going backwards in critical thinking along the journey? Especially since the journey IS the thing for humanity - not the end product of the journey.
> The tools are good, but the ideas are not. These tools basically see many people building the same things independently, rapidly, and over and over again.
Isn't this just every bubble technology since the 80s? I remember being at parties in the early 2010s and hearing awful web site ideas that were certainly feasible with an RoR stack but had a <1% survival rate. The only difference now is less engineers acting as gatekeepers.
I'm certain it was the same way with static HTML webstores in the 90s, social media since early blogs, and other industries that received funding spikes that eliminated financial barriers (construction, film, video games, restaurant chains, etc)
Eventually the Gartner Hype Cycle proves true and the hysteria dies down, but out of those promised elevators of progress come a few incremental steps upward.
I don’t think there was enough money in the web in the static HTML days to have the same bubble. But I do recall domain squatting being a lot more prevalent then than it is now.
More recently was blockchain hype.
This is why I feel like any AI usage is making us all crabs in a bucket when it comes to progressing as a society. Any genuine creativity or effort is immediately stolen and repackaged as someone else's, which inherently means that its incredibly difficult for the original author to get any credit and devalues any genuine progress. And most importantly, people are doing this to deny the author any profit
We seem to be in a situation where 0.1% of people are creating anything new in the programming space, which isn't a change at all, but the difference now is that there's an absolute army of people waiting in the wings for the second they finish to take credit - and profit off their work without compensation. Because apparently that's progress, according to some people
This has been happening the most in realtime with news sites, where any news they publish is fed into chatgpt and recycled to avoid giving them any clicks or advertising money. What happens when all the news sites die because chatgpt ate their ability to survive? We'll go backwards as a society because we won't have news, and then suddenly chatgpt will lose its ability to report the news. Lose lose, unless you're an extreme short termist
It extends to everything with AI, its giving us negative progress as a society, because its society eating itself in exchange for the shortest time line go up
>Any genuine creativity or effort is immediately stolen and repackaged
So pretty much how things have worked almost all the time. For example after the US violently branched off from the UK we stole every idea we could. China operates the same way. Copy and make cheaper.
>Because apparently that's progress, according to some people
According to the people in power keeping information locked up for entire lifetimes is a good idea. They love that kind of progress to stay rich forever.
Yeah sure but now cat is out of the bag and we are in a prisoners dilemma where this inevitable. You can personally not use AI and get your shit stolen in milliseconds either way
The problem is that the tool and the marketing around it make people think they can circumvent the time it takes to refine ideas into a good experience or product.
You cannot, in the end, eliminate time. There are no shortcuts. It's like Euclid said, there's no "royal road" to mathematics, nor to a good product.
LLMs can speed up several aspects of execution, but actually honing an idea, figuring out what constraints make sense, how the experience needs to be tweaked, in brief, design, will always take time. The one shot revolutionary product notion is total myth.
I blame the subsidies and free AI. Can't blame people for building dumb things when it's so cheap and disconnected from the actual costs. That demand then manufactures shortages that wouldn't otherwise exist... The public gets swamped with slop which they view as all AI can make, so they hate the technology. It's the worst of all worlds.
I don't think there is anything wrong with people being empowered to build dumb things, even lots of dumb things. Especially if we believe that there will be a market correction that will force the good things to rise to the top, and everything else to go away.
I don't disagree, but I think we got the ordering wrong. This step should've come later in its evolution.
You're still left with a filtering problem. App review, Job applications, games on steam - any huge increase in volume means it's harder to find the good stuff now.
I think friction to distribution is an interesting thing in this. Pretty unbaked thought that isn't really saying anything, so you'll forgive me: When we were getting deviantART off the ground, folks wanted the user growth to speed up, so we reduced the amount of work to upload things. Then, the quality took a dive and it became too difficult to moderate with the tools that we had, so we increased the friction a bit, over the time I worked on it, the amount and type of friction increased and decreased as we tired to dial in the right amount of growth and quality. It was also interesting what friction filtered out what type of people... add 6 screens you have to click through with drop downs etc, lazy people drop off but creative people get fatigued, add 3 screens but require a minimum of a 200 word description, creative people love it and copy/paste artists drop off, etc. The internet itself has had various versions of this in different aspects of it. I think that is also true in reverse, things can sustain mediocrity through friction. I'm sure there are loads of interesting examples of this all over! Just a thought. :)
I'm honestly massively disappointed.
I expected tools when animator can just put a bunch of keyframes, sketch the transitions and have AI fill it in, getting all of the artistic vision with massive work saved.
Instead industry went full on dystopia "let's replace entire jobs" instead of settling on just massive efficiency improvements on top of keeping the quality and we got million slop generators and little of the actual good tooling that would change the industry, outside of coding tools.
> I expected tools when animator can just put a bunch of keyframes, sketch the transitions and have AI fill it in, getting all of the artistic vision with massive work saved.
I actually tried that multiple times for a simple b/w icon animation. All the tools of the Big AI labs gave me crap.
Well, filling transitions between key frames is an entire job: inbetweener.
At the higher end, AI hallucinations don't hold up. I really do think professional artists and animation studios are using AI at least for rapid prototyping and many are building custom pipelines and constantly iterating until it's good enough for production, or using it in small areas of production (e.g. build a base and manually draw/edit imperfections)
I think if it has key frames and the sketched movement there is very good chance to get it near perfect. And any small problems still there can be fixed after.
On my drive to work this morning I passed by a bus and it had a poster ad on the side for a new musical. Completely uninspired obvious slop. I don't understand why when in the past they would pay better artists and not settle for cheaper ones, they now settle for that gross ai "look". Maybe the average joe / new gen is fine with it? I can actually comprehend outsourcing code to ai more than I can outsourcing your branding / visual identity
>I expected tools when animator can just put a bunch of keyframes, sketch the transitions and have AI fill it in, getting all of the artistic vision with massive work saved.
You expected it to replace jobs, and you're surprised when it replaced jobs?
I'm all for replacing the boring parts of the jobs. Making a bunch of nearly-same animations in between keyframes is definition of boring.
Life has an abundance boring tedious jobs. Unfortunately the animation in between keyframes is almost more important than the keyframes themselves for making the animation flow correctly. Which is why AI animation often looks so wrong and off-putting.
Quality Control is tedious and boring, Do we want AI to take over there as well?
Actually don't answer that I already know most of the sloperators are in favor.
We got those tools too! Not just diffusion and descedants. Cascadeur is a good example of "tasteful" application of AI, meant to help authors realize their vision, instead of replacing it or just making it as trivial as possible.
> Instead industry went full on dystopia "let's replace entire jobs" instead of settling on just massive efficiency improvements on top of keeping the quality and we got million slop generators and little of the actual good tooling that would change the industry, outside of coding tools.
You could try, collectively, uplifting the voices who saw it coming and called it out clearly in advance. They're the ones generally derided and dismissed as impractical weirdos, though, so that might be hard.
As long as we instead uplift the greedy and craven, we will see greedy and craven results from otherwise neutral technologies.
> Instead industry went full on dystopia "let's replace entire jobs"
That's the only thing the elites care about. They dream of not needing pesky labor anymore.
No wonder Hitler had so much support. They even called him man of the year. Then he turned on them.
He wasn't man of the year because he was so great, but because he had a significant influence on world affairs that year.
He was part of the Elite.
What you wanted were essentially specialized traditional models. I still have hope that when this bubble bursts and they hype dies down and people have to pay the real price of the do-everyting LLMs, the industry will pivot towards specialized traditional models, which are give much better results for far far far cheaper.
BorisFX does this great. They don’t even call things AI just ML (which is correct) and the product I use, Optics, has this insane masking model that allows you to just paint with green and red colors very crudely (and I mean that) and get almost perfect masks immediately. These kinds of tools are extremely valuable and worth paying for, compared to Photoshop who has gone all in on generative AI and it’s all terrible, slow and token based.
This really isn't limited to software. A more egregious example is music: we already had every genre, every mood, at our fingerprints practically for free. And now, we're just giving a big middle finger to human artists for no benefit whatsoever. Well, except that Spotify might be able to pay $0.0000 per stream instead of $0.0001.
On some level, I understand where the individual impulse is coming from - "I couldn't do this and now I can, it's amazing" - and I can understand why amoral VCs would throw money at these platforms. But I will never understand why so many intelligent, sensible people are cheering for this future.
Look at all the new book releases on Amazon. They are 95%+ slop, and not good slop. They're made with zero effort. Some of them sell because people judge books by the title & cover. What does this achieve? Once you eliminate all revenue for real authors, what next?
Everyone makes crayon scribbles as a child. What does this achieve?
Everyone learns the same basic curriculum in school. What does this achieve?
Everyone makes mediocre attempts to play instruments/sports as an adolescent. What does this achieve?
Everyone treads the same highs and lows of friendship and romance that have been tread by billions of others. What does this achieve?
It achieves the human experience. Not everything has to be best-in-class to be meaningful, and that's a dreadful expectation to heap upon anything that doesn't strictly demand it.
The things you listed are considered to be smaller version of 'proper' expression in art or mastery of some skills that humans value. Even if most people go nowhere with some of them, they're still seen as commendable things to do and parts of the universal experience.
What is AI generation a smaller version of? While you can make worse and better AI outputs, do you think the average Kindle slop author is going to one day become a proper bestselling writer? Will an average Suno music generator turn into a musician?
"Passive income" grifters spamming amazon with books that they literally spent 0 time or creative effort on are not "achieving the human experience". There is no creative/personal satisfaction or development from this, it's just people making things worse for others trying to make a quick buck.
Difference being people did those things compared to a prompt, there’s no experience to be had from prompting “make music” in relation to trying to play an instrument or lay out a track.
Playing someone else's music doesn't teach you how to compose. Laying out a track in fruityloops doesn't teach you how to play the instruments.
Complaining about negative space is easy but not very informative because the attention doesn't go missing, it gets reallocated to something that someone found more engaging. I remember the vitriol aimed at DAWs 25 years ago, but it was obvious then and correct in hindsight that what was lost in forced mediocre specialization would be more than made up for in expanded capacity, and that anyone who wanted would still be able to drill down and master an instrument, but that they would now be in a position to determine where this would be most impactful. It's the same fight all over again.
Unpopular (and unrequested) opinion: I still think that the world's best DJ mixing music live on stage in a club is not the same class of job as an amateur pop band playing in a pub in front of family & friends. The latter being much more meaningful.
Unpopular, unrequested opinion: band kids messing around in FL studio after practice are doing something far more creatively meaningful than the activity the adults organized for them and recognize as virtuous.
Learning how to compose and play are two different skills though, they are likely to reinforce each other, whereas a prompt is just something completely disconnected. It wasn’t that long ago people were thinking prompt engineering would be a profession of the future, and now you barely need to be able to spell to get the same output.
What does having machines do stuff instead of Everyone making that stuff on a not-best-in-class level achieve? What is the Human Experience in that?
Apart from more wealth inequality and environmental pollution, of course, which is what we call The American Dream nowadays.
If you think we have every kind of music people want, then you lack imagination.
I for example generate catchy music that helps me remember topics. Those are hyper personalized.
Even outside of hyper personalized music, for the styles of music I like, something like 90%+ are topically centered around love/lust and I prefer more topical variety. So I make them.
AI books on Amazon will eliminate revenue for authors in the same way that YouTube eliminated all revenue for television. Most of YouTube is absolute garbage and even the algorithm recommending content is garbage. Yet "real" video producers that make good content still make an absolute killing. You make good stuff people will be attracted to it. The much bigger threat to authors is that people just don't read anymore.
Democratizing coding via LLMs is democratizing cooking via frozen foods. It's gonna be unhealthy to the end user despite the convenience. And the taste level will not be there. Sure, you "made" dinner just like you "made" an app.
There are startling parallels today to how news and information was 'democratized' when social media emerged in the mid to late oughties. The issue is the same: once it's cheap and easy no one seems to care about how healthy it is.
Or maybe the tools are not good?
I can't write more than a few prompts without the LLM (any of the latest SOTA models) starts doing complete absolute nonsense.
I do agree that the ideas are not good to be clear, but the gap between the promises of what LLMs can do and what they can actually do is huge.
Anyway, it's funny to see the insane amount of astroturfing on HackerNews and how people try at all cost to make it seem like the technology isn't the problem but humans actually are. Literally every top-level comment is about that, or downvoted.
Original ideas are not "allowed" in a sense that the infrastructure will fight or directly forbid you. Your original ideas will have hard time to go through Apple AppStore's review or Gooogle's Android or Chrome processing. You will have hard time getting paid for it, you will have hard time to make people hear about it because x, FB, TikTok, reddit etc all have an agenda and they don't spread things that don't fit it(AKA medium is the message).
We don't live in a free world where you can just do things anymore, you are supposed to fit the machinery that is churning hundreds of billions of dollars every year.
> I'm not seeing a lot of unique applications. I am seeing some optimization, plenty of rehashing the same crap.
http://youtube.com/watch?v=nMfHYUFeiJo
I'd prefer to hear what you have to say vs some video which I don't really have time to watch. Care to summarize?
I scanned the video and transcript and it argues (poorly, imo) that competition produces parity and mediocrity.
I also don’t really see how it applies here — AI in this case is removing what most people would understand to be competition. Do people driving cars still compete on running speed?
> AI in this case is removing what most people would understand to be competition. Do people driving cars still compete on running speed?
We're not talking about business competition, we're talking about human competition. Species level, subconscious, and driven to seek (perceived) safety in the hierarchy amongst other humans.
It is interesting I think a hard societal split is emerging between those who understand AI is a tool that can be used and can use it, and those who embed themselves in the view that AI itself competes with them as a person.
Sure. Take a look at the current landscape. What do you see? A lot of copying. How many companies do we need building harnesses that act, look, and behave the same? How many clone apps with the same exact design and aesthetic? How many half-finished games that are just created in order to copy the success of someone on X? How many genuinely novel, quality ideas go unnoticed, unrecognized, or flat out ignored because they don't align with the consensus?
Humans are conditioned to be mimetic: seek an authority, seek someone more knowledgeable outside yourself, but never, ever, ever learn for yourself or do for yourself (especially when doing so goes against the status quo). In practice, this leads to an endless sea of copies that delude themselves into thinking they're unique ("me too" apps, etc). In other words "if they're working on this or doing it this way, it must be important/safe for us to do it this way."
Most people are terrified of being original and authentic and so default to the "short blade of grass" direction of "we'll compete on the same terms as everyone else so we can fit in."
It took me several years to fully understand this but once I did, I saw/see it everywhere. And practically nobody you share this insight with is comfortable with it which tells you it's closer to the truth than not.
Edit: thank you for showing legitimate curiosity and asking me to explain instead of treating me like an idiot.
Just for entertainment in the last month or so we've had a Nintendo Switch 2 emulator and a Playstation 5 emulator that have grown by leaps and bounds. I'd usually put an estimate at 2 years before we'd have these things working properly now it's weeks. But the reason you're not seeing a lot of unique applications (though they are out there for sure) is because people aren't sharing them. I make a bunch of stuff that isn't good enough to distribute widely but serves my purposes just fine. Are they 'unique'? Maybe not entirely, but I haven't been able to find apps that do the same things in the same way or I would just use those. It's really nice to have software that works exactly how you want it and that you can tweak any way you want with a prompt.
The industry is just busy overall with day to day work and it just takes ages to implement things.
Which also means that it is slow in adoption not because the quality is not here but because you need to do things first.
You need an env to run an agent, you need connectors, you need to get that stuff through security etc.
The only question is timeline not investment or bubble because AI as is will not go away anymore.
I asked before for headcounts or consultants and today I can just use AI for it. Its not the AI which is the problem, its the stuff i mentioned above. We will probably still need a few month in my team alone to close the loop.
And whlie we close the loop, models are continesly getting better.
I don't think anyone is suggesting that AI isn't useful - that's been proven. What is being questioned is the business model and whether it's worth the investment necessary to train huge frontier models. How does one get ROI when models are continuously improving and being surpassed and people can distill your weights with a fraction of the cost and resources? How to insulate against the threat of local open weight models as hardware improves?
This is why it didn’t made sense to make OpenAI go private, instead of leaving it as a non-profit. If there’s no moat around building a dense frontier model, and others (like Google) already have a way to monetize a LLM better than you do, and yet everyone benefits from it, then it’s best left as a shared good.
Moats take many forms. In the early days there were many search engines but really there is only space for just one to efficiently make money. Being the last one standing builds a natural moat.
I will replace one person in my Team with AI. Thats one full time salary.
I'm also sure that we will forced to leverage AI a lot more in the next few years which will increase token usage even more and it will be a lot harder for us to get a headcount or consultants.
Open Chinese models have the risk of being manipulated and they are distilling Frontier.
HArdware and compute is an issue, you need partners you are allowed to use (export bans, national security) and who can provide a certain qualith/uptime and trust. After all you are sending all the private data from everyone to them.
It's a matter of national security at this point. So they'll probably find money.
the idea that a bubble means that when it pops AI will go away entirely is silly. the web was a bubble. we still have a web, and it's still useful, but a lot of people badly overinvested or invested in capabilities that never manifested and when the market corrected the price of those investments the bubble burst. same with real estate in 08-ish. A profitable idea drew more investment than it can support, prices went artificially high, then they came back down and we still buy and sell land.
Gotta setup your agentic software factory to generate requirements to feed into your factory to build your software that then needs CI/CD pipelines to deploy to production which requires more requirements for the factory to produce a working IaC repo against your repo and…
…deploy.
(Sir, The airspace is shutting down)
Crap.
I thought the airspace issue was a broadcast storm on a juniper switch some humans installed.
I have seen so much crap in our industry over the last 15 years, there is a reason why all the nice things we have are crap and its not AI
> The only question is timeline not investment [...]
This is not the only question.
As my own organization has rushed/struggled to implement AI workflows, I have seen this same sentiment in many places in slightly different forms. The inevitability of it. The obvious benefits. "The revolution is here." Ok. Except why is it always around the corner? Why does the investment keep growing but the returns keep pushing to the right? Why is the answer to all the problems just "more AI"? From what I can see, it is very much in question that these intelligence products can ever integrate (in their current form) into our organization in a way that pays dividends on the investment. I'm really not confident that the current products are going in a direction that will ultimately help us. Instead what I have seen is groups of users that believe they are much more productive, but their numbers don't actually reflect any improvement.
I mean it took our team alone until now to start the proper implementation for the avg corporate reason: Too much other things to do.
But the result of our PoC were very very good and we are doing this now and it will be great. But our case is analysing log files, incidents, pre analysis of issues, code review, security code scanning and team rules.
We also have to restructure our repositories as a monorepo now makes more sense than before.
I will replace one person in our team because its a lot of effort to teach that person stuff and explain stuff to that person.
AI is better that this person and whatever I teach the AI, it will not quit and I can finetune it.
That's some big cope you've got there.
This isn't "AI is useless" or "AI is dead". They are saying it's not worth the cost.
Further, I would argue that coding is the only place it's made solid headway and even then it either scaffolds to less than perfect, or the code base grows to be unmaintainable. Not saying you can't ship a product with it, just saying that while you are moving fast and breaking stuff someone else is shipping a solid product marginally more slowly.
Last point "continesly" is not a word. Continuously maybe?
Recently, I tried to automate a tedious aspect of my day job (renaming scanned check/invoices using information from each) and after a while, Microsoft imposed upload limits which reduced the number which could be processed each day to 80--100.
I've tried to ask how much it would be necessary to pay to remove that, but haven't found an answer with a solid price.
I managed to get a local LLM running and processing such files --- it took 405 seconds to process two on my work-issued laptop (by hand, with the aid of a batch file cobbled together from a spreadsheet I can do ~200/hr.) --- I need to do almost 5,000 of these for Q4.....
There were literally dozens of products and open source projects that did that before AI.
I haven't found one which was reasonably priced and which would cope with handwritten checks and invoices which ranged from crisp printout through rough fax through inkjet print of photo of only partially unfolded invoice with a reasonable error rate.
If you know of an opensource/free tool which does, please let me know.
LLMs and the general progress of ML (probably also through LLMs) made a LOT of things relevant better.
OCR with an LLM and context icnreased our OCR quality by over 20% over night. Finetuning Gemma on semantic data also at least 15%
and do you consider yourself responsible or liable if something goes wrong with those signatures?
Not sure where "signature" came from --- if you're concerned about PII, then I have the account line covered over w/ a pen when the scan (really a photo) is taken, and everything on the invoice is public record, and well, identifying information on checks feels pretty non-private to me --- it's not like anyone still puts their SSN on them as was common when I was young.
The renamed files are then opened one-by-one and the filename verified as the check is entered into a system which tracks payments, so the ~8% failure rate isn't a big deal and is corrected as things are entered.
Of course, exactly as much responsibility as there was having Microsoft do it.
Or did you think he could have told the customer that MSFT messed up so no harm no foul?
you're not the one I asked
Would you be more pleased to hear "AI bad" ?
do you also sign 5000 documents per month with an LLM? if so then, I apologize
For the specifics of what I am doing, please see:
https://news.ycombinator.com/item?id=50520440
> Or did you think he could have told the customer that MSFT messed up so no harm no foul?
that is exactly the value proposition of software as a service
Software engineering, math, physics (is coming now), law, media. The list is already big with plenty of money in it.
Even if you pay for a legal GPT as much as one employee, if the legal GPT is 100x faster, you will replace the human salary with tokens.
"Continuously" yes i'm not a native speaker and people complained when I was using AI to correct my text so currently I see it as human proof.
Everyone I know who isn't anywhere near code uses AI daily. Law, telecommunications, real estate,. Some more effectively than others, law seems to be getting huge bump specifically (GLM law, Astra for law). I'm of the opinion if you aren't reliant on it now you will be soon.
Does it cause problems? No doubt. People get way in over their heads with AI when they don't know what they're doing. But when you know your field and it's simply a conversation about busywork that needs to be done it's better and much cheaper than most assistants you would hire (unfortunately). It can save 10s of hours a week easily.
> marginally more slowly
I find it fascinating that you could hold this position. Pick a single popular SaaS today and you can duplicate it in a week. Of course that's not the only hard part of running a business, you need to market it and sell it. But that's millions of dollars and months to years of work reduced to nothing.
Even marginal labor replacement in knowledge work leads to pretty significant revenue (not profit) figures split between the labs. I do think diffusion is relevant. It takes awhile to find product market fit and for new ways to work to become common place. We've seen this with most revolutionary technology (usually on order of decade+) and it's only been a year or so that these things have been good enough to replace some labor. I think we'll see incumbents who are slow to move on this get replaced by new entrants with much better margins.
Not to say the hype isn't fucking annoying and often bullshit.
you can repaint the monalisa with some robots but to what end
The difference is that the Mona Lisa is art, created by a human. People generally value art because of the human factor - the artist was trying to say something, there is something interesting about the piece, etc. This is part of why generated images get so much pushback.
The same cannot be said for SaaS: your business needs certain productivity tools in order to operate, and the only options tend to be these hyper-expensive SaaS options whose UX has been clunky since 2005. Historically, you could not build these in-house because it's more expensive to hire a full engineering team. Today, the math being done is whether it is worth it to build something 90% as good for 1% the price of the SaaS.
These people are not using AI they're training AI. In a few years it will replace tham and their jobs just like it did programmers
Generally, enterprise contracts with the AI assisted coding companies have protections against training on the inputs.
You or I probably can't fight it if they lie on their terms of service. A Fortune 500 would go for blood (metaphorically speaking) if their IP was violated by one of these companies.
> In a few years it will replace tham
There isn't anybody to replace in this scenario. You're either buying the SaaS, hiring a full engineering team you didn't have before, or slop coding it. But if you're already on a SaaS, someone higher up has already made the decision that you cannot increase headcount to support something internal.
Personally, I'd prefer if they hired more people. But the cold math of your average business is going to choose to ditch the SaaS in favor of the cheap slop job.
problems evolve, hence why a saas is a subscription and not a disc you can buy with a license code
clones are frozen in time and tend to rust and a financial incentive emerges to just purchase something that doesn't rust
Clones are not frozen in time. If you've already slopped out some vibe-coded abomination, you can just modify it again.
The cost differential is so great when it comes to B2B SaaS that you could generate a brand new every month and still come out ahead.
I know there is probably some axiomatic "AI Bad" at play here (I avoid vibe coded solutions as well), but it's myopic to claim that we aren't at a point where the traditional B2B SaaS model is under serious risk.
Costs to maintain many of these items aren't that much different than self-hosting open source. Every week on HN you hear someone saying something has been enshittified. Well then just maintain your own shit with half the spend.
We used to buy software on floppys. It provided value without a sub. Pretending that every single software solution requires a 50mn operation is silly. I suspect HN tends to think of problems framed at massive scale.. but most people just ain't Google.
> We used to buy software on floppys. It provided value without a sub.
that was a different world with different problems
> self-hosting open source
I get update notifications almost every week or so with most of my self hosted software.
Yep, software rusts. And when something like, I dunno, a popular self-hosted BI tool has 1000s of customers with 1000s of more needs then of course you'll have an update every week. They also have to stay ahead of their competitors with features. But when you're supporting 1 customer on 1 platform and have a choice of dependencies the update cadence ain't going to be nearly as bad is it?
A frontier model has no problem keeping a VM and a list of dependencies up to date. When one of those dependencies goes unmaintained it can fork or rewrite it and do the same there.
> I suspect HN tends to think of problems framed at massive scale
I think this is an industry-wide problem. People (especially managers and juniors) tend to read blogs or watch talks by FAANG discussing how they saved millions or achieved 5-nines uptime, and go on to cargo cult it.
There seems to be a perception that everything needs a sub-3 hour recovery (plus on-calls), when realistically, the business will go on just fine if some QC tech can't log their passing test for a couple days (the plant manager makes a call to ship it now and update the record later), or a project manager can't update their portfolio until Monday (they can just message someone if a status update is immediately needed).
Yeah, some things need resiliency. But a key aspect of engineering is balancing the complexity and cost of a certain design, with the risks of that design failing.
Who wants to draw the Monalisa with a robot?
You know who wants AI? Your boss, my boss. They see us as effort. Something you need to control, manage, keep happy. Something you need to cater to.
Whenever I talk to others not in IT, its crazy what working conditions they have.
> Pick a single popular SaaS today and you can duplicate it in a week.
This is a common claim but I have not seen it played out in practice.
The most well-known example people cite for this is Photocraft[0] which went viral a month ago; the claim was, at the time, that it was 80% feature complete upon release, and people took this to mean that Adobe was dead. The creator claimed that within a month there would be feature parity with Photoshop.
Currently, their feature parity tracker[1] has them at 43%; a whopping negative 37% improvement in compatibility. Yes, this was a change in measurement, but by their own admission:
> the 2026-10-03 estimate (82% "of the way") [...] counted feature surface and was far too optimistic
The impending "SaaSocalypse" was predicted earlier this year, but never materialized, and you can see it in the stocks of Atlassian, Salesforce and others. They're getting hammered in the past few weeks as with all other tech stocks, but if you put all your money into Atlassian in June you would still have tripled your money right now.
[0] https://github.com/storytold/photocraft/
[1] https://github.com/storytold/photocraft/blob/main/docs/targe...
Because people don't understand timelines.
When I saw Photocraft for the first time, my thought wasn't "Adobe is done" but "wtf this will change software as we know it in the next 5 years".
Then the whole reingineering of firmwares and fixing bugs in software you don't have the source code for. Was it possible to break software before? Yeah sure but even me i was too lazy spending weeks on disassembling, analysing and doing it if I was able to throw a little bit of money at it. Now you don't need days or weeks anymore, you just let it run and get back to it to see its solved.
You will not see a normal company selling software as non SaaS solution. You will see companies protecting themselves and closing off APIs etc.
And you know, the more people are jobless and more young people just using these tools without any second thought, will fill all the niches which are now accessable because of AI. That small company somewere in a more rural area building and operating some small tool? Someone else will vibe an alternative.
I already here stories around my non it bubble of non it people vibecoding themselves tools they would have paid before. On TV an architect said "i vibed our hour tracking tool, i also did this and that". This alone makes software so much more accessable.
> I have not seen it played out in practice.
Of course. As I said it takes time. Writing the software is different than replacing an encumbant. A new software company starting with AI simply has less labor costs than any time in history. Claiming a competitor starting from the same place without AI is marginally slower is ridiculous.
> Of course. As I said it takes time.
Well you said it takes a week. Which is some amount of time, to be fair, but short enough we would have seen a major SaaS rewrite by now.
> Writing the software is different than replacing an encumbant. A new software company starting with AI simply has less labor costs than any time in history. Claiming a competitor starting from the same place without AI is marginally slower is ridiculous.
Of course, but while you're using AI to build e.g. Jira from scratch, Atlassian is also using AI to add more features. They may be niche features that most users don't care about like "scheduled immutable backups for compliance reporting" or "logging Confluence searches to internal employee monitoring software" but they're the features that enterprises are paying for.
> Well you said it takes a week. Which is some amount of time, to be fair, but short enough we would have seen a major SaaS rewrite by now.
You're confusing things. Writing the software with replacing an encumbant. I also said diffusion typically takes on the order of a decade. I absolutely don't think you're going to replace encumbants like intercom or zendesk in a week, despite the fact you could write the software relevant to your business to replace it, in a week. If not only for the fact that someone has to want to do that work.
To be completely clear, I agree with the premise. Overhyped product under delivering is nearly a tautology. The issue is with expectations and timelines.
> Further, I would argue that coding is the only place it's made solid headway
Why? The company you work for seriously doesn’t do anything productive with LLMs?
How about search? Every big company I’ve worked for has some shitty enterprise search. LLMs have massively improved that situation.
They’ve also almost completely replaced L1 support at my company, the numbers there are way down just from people being able to find information they need. I don’t have broad numbers higher level support, but for a couple niches that I know about they are down as well.
For example, in a global company of 100k+ employees you’d be surprised how many support requests HR got just for PTO related queries, not even real problems. Now we have an agent that can fully manage your time off, has the policy documents for every country, pulls your data from the HR system, their number of support tickets are significantly down.
Another project I worked on involved automating some risk analysis stuff as part of the IT process everyone has to do when they develop or procure anything. After developing it for a couple months, then a few months running human in the loop validating all of the assessments, we have analysts manually audit 5% of them now. All of this used to be a shitty excel process, now is a slop web app. Cycle time is down from weeks to minutes because nobody has to wait on reviewers and the app can prefill most of the data from your general system documentation. Overall everything is just as or more accurate,
I know these aren’t very glamorous, barely considered real work by some, but in my feeble imagination there are many such processes being automated and improved everywhere.
> How about search? Every big company I’ve worked for has some shitty enterprise search. LLMs have massively improved that situation.
Using Glean at work and it's not much better than bad searches. It frequently gives me outdated information, or uses questions I asked in slack as answers to the same questions I asked in the LLM.
China seems to always have viewed AI as "normal technology". Maybe they know that they're falling behind and won't hit RSI like Anthropic thinks they will, or don't think that it's possible at all. Also notable that Baidu/Xiaomi don't have any frontier models, so it's a bit like Samsung commenting on the AI race
It's not about frontier models at all. They're commenting on AI as end users. They are saying the returns from the use of AI do not match the hype that has been created by the frontier labs.
Xiaomi has MiMo, which is not "frontier" but still a good model.
Xiaomi no frontier model? https://artificialanalysis.ai/models/mimo-v2-6-pro
Samsung is in the business of selling RAM and storage for all these data centres. They adore the AI capex and don’t want it to end.
So you are saying Chinese firms are now going through the sour grapes phase.
I am thinking maybe your blunt take is right. It is true their primary objective with distilling western models and open sourcing stuff has seemed to be to undermine the US economy. Cold war tactics, instead of running a normal business.
Otherwise per their own admission why would they invest huge amounts, that too on student models.
> Cold war tactics, instead of running a normal business.
There's a bit of a double standard here: When a Chinese company does it, it's a cold war tactic. When an American business does it, it's capitalism. China engages in some shady shit, but this is not a case of that IMO.
Let's not forget that Meta founded the open LLM industry when they released Llama for free. Microsoft, Huggingface, Google, and even Amazon (all American companies) have released their own.
The business strategy here is to disrupt the closed-weight leaders (OpenAI and Anthropic) by delivering an LLM that is good enough for day to day use. This commoditizes AI, driving the cost down and shifting the business model to providing off the in infra.
It's no mistake that the companies releasing these open-weight models are by and large cloud providers.
> their primary objective with distilling western models
This is another double standard: it's simultaneously wrong to train on the output of a closed model without permission, but somehow okay for the closed model to have been trained by reading everything that has ever been written without permission.
Companies and Governments do things to their own interests but those interests are wildly different.
Yes, China as a nation state has different goals than Alibaba, just as OpenAI has different goals than the US.
This is different than taking a basic market play of commodification and painting it as some Commie conspiracy
The difference being that there is a Politburo officer sitting on the Alibaba board. The amount of direct control the Chinese enforce on their businesses to ensure their goals are aligned with the state goes leagues behind what's done in Western Democracies.
Sure, but that still does not change the fact that everyone who isn't OpenAI or Anthropic are trying to commodify AI.
It's completely irrelevant here. It would be just as logical to try and tie them drinking water to some conspiracy to undermine Coke, because some Chinese public ad campaign encouraged people to drink healthy options.
I agree with you that some of the most outlandish attacks (especially the ones directed at open-source/weight) come from US companies that don't appreciate the competition but don't be fooled for a second that just like the US with its chip exports controls, China is not aggressively trying to undermine US interests.
Xiaomi MiMo is quite good.
> China seems to always have viewed AI as "normal technology". Maybe they know that they're falling behind and won't hit RSI like Anthropic thinks they will, or don't think that it's possible at all.
If a leading AI company warns that there is a bubble, perhaps people should listen?
The likes of OpenAI and Anthropic have an interest in keeping the hype going, as losing the hype poses an existencial threat. They are highly unsustainable as-is, and losing momentum won't make that problem any better
> Also notable that Baidu/Xiaomi don't have any frontier models, so it's a bit like Samsung commenting on the AI race
Nonsense. The ugly truth is that nowadays the cutting edge frontier models might unblock niche applications, but the cheapest models in the Pareto front are already handling the bulk of the prompts/sessions.
And nowadays the only reason why cheap models such as GPT6 Luna are on the Pareto front is due to the fact that OpenAI rushed to slash it's costs in half right after it's release, in a desperate attempt to be close to the Pareto front and remain competitive.
I'm confident that once hardware costs return to normal, the bulk of heavy AI users will jump head first to local models and US companies will face an existencial crisis. There is a reason why US government officials accuse Chinese open weight models of being economic warfare.
This is a bad take. The China market is different.
Baidu and Xiaomi don't create models. That's handled by other entities in the China market.
Baidu and Xiaomi experiment with models. Test the efficacy of different models. Get an idea of the relative economics of different models. Maybe create a model of their own to see if they can improve the business case in one dimension or another.
What they're saying is that, in their role as customers for these models, they're a bit confused about the business case. I think that's a fair statement. They've presumably put a lot into finding out how to make money out of these models, and presently, the returns are not there with respect to the bottom line. Additionally, their experiments have given them enough background to feel comfortable asking, openly, where exactly everyone else feels the returns will come from?
For them, yeah, just keep the little experiments going and don't go all in until someone offers you a model that can answer questions for less in cost than the person positing the question is willing to pay you. Maybe stay in verticals where the questions are valuable enough to be worth the cost for right now. Medical, Finance. And limit exposure to the larger market of less valuable questions: "ChatGPT, where is the closest pizza by the slice?"
> Baidu and Xiaomi don't create models. That's handled by other entities in the China market.
12 days ago: https://huggingface.co/baidu/ERNIE-5.5-Flash-Base 20 days ago: https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL
You didn't read further..
>Maybe create a model of their own to see if they can improve the business case in one dimension or another.
All done in the context of experimentation to improve economics not intelligence.
Baidu and Xiaomi are trying to figure out how to make money. Not how to make models. Making models is not their business and they don't want to be in that business. China is full of dime a dozen PhDs who can make models. They don't want to do that, they want to make money. That's the point.
They want to understand what are the conditions required for model providers to offer them a model, any model, that performs in an economic sense at scale. That may require creating a model or two as experiments, but that is not at all an indication that they want to change their business to model creation. Nor even get into the business of model creation.
Model creation is just not a good business for the likes of Xiaomi, Samsung, Apple et al. If they have to, fine. But it's clear from their statements that they really don't want to because it's, currently, a terrible business.
The problem isn’t AI, it’s capitalism. I use AI constantly and have built my own personal universe of software. Instead of buying a solution, I build it myself. Instead of using open source, I build exactly what I need and nothing else. Instead of using a library and introducing a dependency, I just use AI to build it in my own code base.
This has been revolutionary and game changing for me, but what it’s not is profitable in any way. In fact, quite the opposite. I also don’t even bother open sourcing or distributing what I build. It’s my computing dream, but it is just for me. Capitalism nor our idea of open source fit this model.
Relatable, computers are clay now. It’s a revolution but from the ground up and a quiet one
Its honestly a dream come true.
> have built my own personal universe of software.
what are these and why did you do this? I am the only one that doesnt need custom software in my life.
Not OP: GPT-5.6 made a XApp tool that lets me add or modify rows in a Google spreadsheet I use to track time working. It has a menu that lets me start and stop timers, edit the most recent row (in case I realize I forgot to start or end), and open the spreadsheet. Simple but probably not something I'd build myself.
I also have an XApp that when clicked lists recent messages received in a private messaging app that has an API and CLI but no easy way to see if there are new pending messages.
I used Claude to write a tool that lets me plan rebalancing portfolios across multiple brokerages. I was doing this by hand for a few years and I sure hated it. It's not smart about taxes (I still have to use human thinking for that part) but it reduces the effort quite a bit.
None of it is revolutionary stuff, but no software I write for myself will be revolutionary. Must it be, for it to be worthwhile? It helps me out and that's good enough.
I saw your other reply and I get the impression you wouldn't like anything I'd design. That's OK.
Not op but...
I've made a AviSynth/VaporSynt image manipulation program which runs on WebGPU in the browser. I wanted a script-like experience for reproducibility (apply exactly same transformations to several images) but also interactivity for fooling around with.
I made my own astrophotography camera controller for my Canon camera. It runs on a Pi and has a browser UI, so I just need the phone and the Pi. It does autofocus, helps reframe the same shot on a different day, polar alignment assistance, and sends data to Stellarium so I can see exactly what's in frame when back inside.
I made a true-color version of the Quake II software renderer[1], I've always preferred the gritty look of the software renderer over the GL renderer, but not the poor shadows and such which comes from the limited 8-bit palette. So now I'm playing Quake II the way I've dreamed of all these years.
I've made a Rust-based implementation of JPEG XR, not to to be confused with XL, so I could more easily convert OpenEXR images to and from JPEG XR. This format is what Windows requires for HDR background images, and existing tools didn't handle color spaces right.
[1]: https://news.ycombinator.com/item?id=50529463
I've built a family recipe cookbook taylored to our needs to eliminate any pain for weekend groceries and evening cooking. No more slow Airtable, no more shitty bloggers storytelling around the recipe.
I've also a wine management app, with an integration to the cookbook one so I can track which wine I've had when and on which meal. And where I can find in my two wine cellars the kind of bottle that would match with tonight's donner.
For work I've built a stand-up tool that display exactly what we need to prioritize work, on top of the ticketing SaaS tool we've been using for years.
We don't need to settle anymore for someone else product decisions and tradeoffs, we can build made to measure tools.
Why am I not surprised…
Every time someone gives an example of the ‘amazing’ things they’ve accomplished with AI, it’s bullshit like this.
Tracking food/wine pairings? Seriously? You could accomplish that with a notebook and pen.
I am now fully convinced that AI is a bunch of bullshit hype.
It's great you can make this and I don't want to put you down personally, but to me this sounds like you automated a bunch of non-issues.
Recipe management, well, OK... but tracking which wine you had when and paired with what meal? I guess that's an issue for some folks, but come on.
Time and time again I wonder what everyone is building and all they're doing is dogwalking management, recipe management and wine tracking.
Surely we as a group can do better? Does anyone have truly ambitious ideas? Is there anything left in software?
Sometimes I wonder if we are at peak software and everything's been done before.
I agree with you, it's very micro optimizations for very first world problems.
i'm building a wpf text editor that fit my work flow with agentic coding.
i am a long time 10+ years Jetbrains subscriber. i tried Jetbrains's new ai product and it just doesn't work for me and the IDE is becoming less important for me. i only use it to review ai written code. i do not plan to renew my Jetbrains subscription.
NTA but I had the clanker build this for me:
- A sane image editor for Linux, targetting the parts of Paint.NET that I need and skipping the rest (not having this annoyed me for decades)
- Closed caption overlay using a local nemotron-3.5-asr-streaming-0.6b
The latter I could've easily hacked together on a weekend myself, but I never had the energy to do it. Now it's (basically) free.
Could you elaborate on how you did the image editor port?
I'm wondering if you need a 200 dollar monthly subscriptions to get enough tokens for "sane image editor for linux" and what harness and model you used?
Also did it require days of human testing and refinement or were you able to automate that part with A.I.?
I used a claude pro subscription back in the opus 4.x days. If the usage limits filled up you just have to wait until next week. I wasn't in a hurry so this was fine. For testing, I wrote (this was hand coded by me since I found that part mentally stimulating) a couple of scripts that the clanker could use to drive the image editor headlessly: A headless sway instance, a daemon implementing the wayland virtual keyboard and mouse protocols that accepts commands via an unix socket, and a couple of bash scripts to tie everything together. Then I gave the agent instructions to use this setup in claude.md, the basic command set is `./headless.sh {start COMMAND|screenshot|mouse move X Y|mouse click X Y|type STUFF}`. My first attempt was using accessibility APIs by giving the agent access to pyatspi2, but there were a lot of things that didn't work properly. Using input mocking was far more reliable.
So the development was done with opus, the interactive testing was done with a sonnet subagent for cost reasons (cheaper model & doesn't inherit the long opus context). Most of this testing was completely automatic, sonnet caught most bugs itself. I still had to test everything manually however because headless sway doesn't behave exactly like a real DE with kde's platform theme, but this was just a cursory check. Most of the work was me specifying how I want every tool and panel to work.
Many things from command line tools, to web services, to full blown gui environments. I do it to explore ideas and/or solve problems. Once I started, I found it incredibly powerful to control my own release cycles and feature development. I also found I don’t actually need a good chunk of what most products do.
It’s funny because they say the first 80% is easy, but the last 20% of polish to get a shippable product takes most of the time. Well, if you just build for yourself and never ship, you can just crank your own solutions out at hyper speed.
ok its more like a hobby like model trains. I've been building stuff like that too, ngl its fun.
> I do it to explore ideas and/or solve problems
that's awesome. I wonder if you can name, like, a single problem you solved this way?
Why are you being sarcastic? He solves problems he encounters, for himself. Why do you care?
Do you really believe everything people say on the internet?
The claims made are overly bombastic and quite hard to accept at face value. If you have used any AI tool to build anything non-trivial you would know how long it takes and the money involved.
Generally, it is good to be skeptical when people make claims like these particularly when it comes to AI. Looking at their profile, you see that "slowin" was created just over 6 months ago but has already amassed over 2.5k karma points! This is a red flag that this account is just "karma farming" pushing AI and similar other popular subjects in a concerted group.
Wait, what? Is the anti-AI goal post moving, on Hacker News nonetheless, at a point where we claim that building software is a needless whim?
Solving problem isn’t a needless whim. And such solving can range from fully specified software to hacky scripts. But playing around, like rebuilding 90% of vim, is needless. Because why not just remove the 10%?
It’s kinda like few really attempt to reimplement ffmpeg. It’s just easier to disable all the codecs you don’t need than to reimplement the codecs from scratch.
I don't understand why you see this as an "anti-AI goal post". It's precisely because AI is powerful at building software that we can no longer assume software is motivated by something beyond a needless whim.
Capitalists are known for pushing their customers into unhealthy over consumption to maximize profits. I would argue that 99% of all vibe coded software is nothing but throwaway toys, that gets used a couple of times and thrown in a pile of other unused toys which is quickly starting to take up all the space from your room.
Things thats can be built with no effort have no value by definition.
You wouldn't have AI without capitalism.
You wouldn't have AI without anti-capitalism (open source).
Maybe Free Software or copyleft software would be more accurate/poignant.
open source is not anti-capitalist and in fact has thrived in our current capitalistic environment
Everything needs to exists in the "capitalistic environment". People, organizations, healthcare, church, software, activism. It's the environment we have.
Doing skilled information work for free for the benefit of the people around you is very easy to rationalize as an anti-capitalist act.
Built on open source computers of course.
AI is capitalism. It is build by stealing labor, the models (i.e. the means of production) are owned by owners of capital who hand it to us if we pay them, i.e. rent seeking. If I am wrong, then AI is feudalism, which is not any better.
You wouldn't have indoor toilets for everyone either, what's your point ?
So the future is everyone throwing $x00/months to megacorps to rebuild the wheel forever?
When you say "everyone", how much of the general population do you believe is currently using AI to write their own software for daily use?
No, I think local models will be good enough for this as long as Anthropic doesn’t lobby to lock them down.
I can’t foresee what happens to the economy, I just know that I’m far more empowered than I’ve ever been before, and I’m not going back.
Do you code for a living? I'm asking because I'm curious if this is trend in general or among programmers
I do, and have for nearly 30 years.
Then you will be giving money to the electric company instead. The poignant part of the post was "re-inviting the wheel forever"
Have you seen the prices of gpu/memory lmao? Virtually no one can run anything remotely useful locally and won't be able for the foreseeable future.
There are maybe 100m people worldwide ready to pay $20 a month for a top of the line llm, no one will drop $2k+ on a machine to do the same at home with a gimped local model which brings 1% of the value since it won't be tightly integrated with anything
Poor people didn't usher in the computer age or the personal computer age which later branched off of mainframe computing there were plenty of people computer users, students (Steve Jobs, Bill Gates and others for example that ruined IBM's big iron dreams) and there are many small/medium companies today that can and they will push forward irrespective of Sam Altman's ram declarations.
It depends on how you define foreseeable future, I think we’re in a temporary state.
“Instead of using open source…” - you might be surprised to discover what your models were trained on.
I would put a finer point on that, and say the problem is specifically highly-tuned financialized capitalism. Where "highly tuned" means that a LOT of resources are cranked to N degrees of precision about making money from calculated risk, where the actual value of those risk calculations is largely bollocks.
I'm not throwing shade on the entire risk field, but I AM throwing shade at betting trillions of dollars on what you think a tiny fraction of future costs might be going off into infinity. It's what made image sharing or bond insurance appear more profitable than asteroid mining or solar power, and it's the leprechaun promising the pot of gold under the AI rainbow[1] for financialists.
This happens to be a puzzle piece that fits pretty well together with several well-smashed shards of the Old European Right, but that's another subject and much more dubious. The "financialist thesis" of Western Decline is on much more solid ground - you can see it with graphs and shit.
[1] I.e., eliminating all future labor costs forever
I do the same for my company. Behold, it's profitable. That happens when you put those efforts in the middle of something that generates profits.
You don't build anything. You are plagiarizing by the press of a button and are delusional enough that you think you accomplished anything.
In that sense, plagiarising was always part of programming, a smart programmer would be able to analyse and lift parts of code to be remixed and re-applied.
This moral ship has sailed, as much as you and I would like for it to not be true, if this is the world we need to live in I will do as OP and use these tools to achieve stuff that wasn't possible with my free time before. In that sense I think it's an accomplishment to be able to generate code for ideas I had but didn't have the time to properly code them.
"It has always been this way." Do these talking points have chapter and verse numbers, like: Book of Amodei, 3:16?
I'm not defending it at all, I'm with you on the morality of it but this cheap shot about plagiarising is not a good argument as a counter. Software has always been plagiarised, that's not the moral issue with LLMs being trained by gobbling any available data.
Would be nice if you don't use a throwaway to write low-effort comments though, that's also quite cheap and weak.
I mean, yeah. They're deeply connected to the Chinese state and the Chinese state is trying to undermine the American economy, which has stupidly put a lot of its eggs in the AI hyperscaler basket.
It makes total sense that they'd warn of this. They're also somewhat right. Not for the right reasons, mind, but they're right.
It doesn't seem China is very agressive towards the Western economy? I'd argue it's much clearer that America is acting agressively towards China's (and other's).
China just has much better leadership and is not being loud about it.
They remember the Opium Wars, Boxer Rebellion and Japanese occupation and they have the pressure of 1.3 billion people living in a area the size of the United States from the Atlantic Ocean to the Mississippi river with a massive barren plateau above 10,000 feet. Imagine 1.3 billion Americans living/restricted in the same space that would make you plan faster.
How're things going over at Volkswagen?
Seems mostly "hahaha, look at those fuckers undermine themselves and cripple their own economy just to give us money for some cheaper labor"
I'll be a believer when the technology allows them to crank out a new GTA every month.
This all comes down to execs looking for a reason to be relevant. Just like every single "revolution" in software development, the executive set thinks that this is going to be the innovation that allows them to finally hire fewer programmers and deliver better software than ever before, and it's never been true, and won't be true this time.
AI has become an incredibly useful tool for competent engineers to build more, better software. Is it orders of magnitude more impactful than, say, the widespread introduction of OOP? yeah, probably. Is it still the same kinda thing? Yeah, absolutely. You can do some _incredible_ stuff as a solo developer with a good coding agent and good domain knowledge these days; as we speak I'm working on a complete astrophotography toolkit that can run off an android smartphone, and this is possible mostly because 1) I already knew enough android to reason about the ecosystem and 2) I know enough about astrophotography to reason about the work that needs doing. Domain knowledge is what separates AI slop from good software that happens to be written using AI, and you can't just magick that out of existence, no matter how much the pointy haired's want it gone.
I don't think we're particularly close to an AI system that can fully substitute for human taste, and if we stop pretending we can already have "fully autonomous AI employees" and instead see AI as a new paradigm for building software in the hands of professionals, we have a good chance of actually changing the industry. But stupid hype cycles are gonna stupid hype cycle, and it'll probably take a few more MSFT-level crash outs for us to get to a level of sanity around it.
To me this reads much like “Buick CEO warns of growing gap between electric car hype and returns.”
It took them 2 years for their people to report the hype? Or did they never follow what's been going on last 2 years? Hype's been there for years!
It really looks like we’re starting to see the first failures of vibecoding ( vibe dev ops too) in prod.
Can’t send emails, fly or check your bank balance. Ohh well.
I would love to see 2027 be the return of QA across tech.
I'm more likely to attribute this to the layoff culture that swept through all tech related businesses after Elon showed Twitter could more or less continue running after laying off whatever huge percent they did.
they're already doing damage control by saying "the code was written by a human" which to me sounds so damn funny, you can tell the ai boosters are scared of reality catching up
What do any of those issues have to do with AI other than they happened loosely in an environment that contains AI and a bunch of conspiracy theory "it's obvious bro" connections?
they’re happening in an environment where companies and employees are claiming that ai writes all the code and is often handling the entire process through so-called “software factories”.
And?
Does AI get credit for all the advances that are happening too, or do you just cherry pick things that suit the narrative?
It seems pretty reasonable considering the constant refrain of "nobody is handwriting code anymore" and such things.
Mass layoffs justified by AI “productivity gains “ and essential services falling shortly after.
Seems connected.
Democrats sweep to victory opposing "AI data centers" and essentially services falling shortly after.
Also connected?