Fixing my furnace: https://news.ycombinator.com/item?id=48417845

New software for a retro keyboard: https://news.ycombinator.com/item?id=48418158

Customizing my camper van: https://news.ycombinator.com/item?id=48417379

Porting my astronomy app from an old Nokia phone: https://news.ycombinator.com/item?id=48419242

Fixing my kid’s science fair project: https://news.ycombinator.com/item?id=48419364

Unborking the family printer: https://news.ycombinator.com/item?id=48419480

Learning to draw anatomy (!): https://news.ycombinator.com/item?id=48418716

Lowering my electrical bill: https://news.ycombinator.com/item?id=48417949

Making classic guitar pedals programmable: https://news.ycombinator.com/item?id=48418006

Avocado armchair guy victory lap: https://news.ycombinator.com/item?id=48417658 (<-- oops, wrong: https://news.ycombinator.com/item?id=48418274)

Putting an overlay on enemies in a video game: https://news.ycombinator.com/item?id=48420635

It just goes on and on. I was a little nervous when I saw that post originally, but it's amazing what happens when a title is somehow just right.

My next project was a 2d game with random terrain, physics, sound, music, multiple levels, a day/night cycle with transitions high score tracking... (not uploaded anywhere, but it works, and I refined it a good bit.). That was more like 8 hours and maybe a 100 prompts.

Here are a few screenshots:

https://imgur.com/a/vhUXBu3

One thing that I have found to make a pretty big difference is using both the latest models and higher thinking levels. Opus 4.8 with thinking on Extra or even Max is genuinely mind blowing. The thing I hadn't really appreciated, having a sort of naive impression formed mainly from using free early versions of stuff like ChatGPT and Stable Diffusion was sort of that "Type a big ass prompt and it craps out a result" experience. But Claude is really great at refining from feedback, and it's way more flexible and responsive than I would have ever expected. I can do something like take a screenshot of a small portion of the running app or website or whatever and just say "This button needs to be bigger" or "make this red" or something like that, or even sometimes just "fix this", and Claude both correctly identifies what I'm talking about, and actually does the thing.

here I've found it really, incredibly game changing is my health. I have a pretty, to put it mildly, complex medical profile at this point. I haven't worked in over a year and pretty much every sign is pointing towards permanent disability at this point. Tons of symptoms, long med list, and I live in a smaller town with not great access to care. I'm also autistic and have not the greatest verbal communication, especially under stress or time pressure. I dumped all my info at it, in bits and bobs over several days (Side note... it's memory is pretty limited, but it will quite happily right out everything it knows from a session into a markdown file it can later re-read. I've found it very good for things like screening for drug interactions, or talking through and logging symptoms (and it can log those into human readable markdown files too). Biggest win (other than having unlimited time and interactions) is that it thinks across specilaties, versus the "real world" where the gastro only wants to deal with gastro stuff, neurology only wants to do neuro.

I certainly don't (and wouldn't) use it as a replacement for a doctor, but as an adjunct it's phenomenal. For instance, it flagged a possible drug interaction with a symptom I was having, and then offered to draft a portal message to my GP about it. I have poor executive function so lowering the friction from "type up a message and send it" to "copy and paste" is actually a pretty big deal. Turns something (I probably won't do) later into something I will do now.

It wouldn't surprise me if my very direct, literal, autistic communication style is particularly well suited to interacting with AI. I actually find talking to it rather refreshing as, while of course it's not perfect, it tends to actually respond to what I say rather than the all the assumed subtext NTs tend to expect/react to.

The alternative isn't "run everything by a human doctor", it's nothing.

Jokes aside, thanks for your selection. I had read some, but missed others until your comment.

If it matters, I think there's some people that hasn't decided yet what tribe (pro-AI/anti-AI) they belong to. There's probably dozens of us!

There is a really important question that is been lost between the anti and pro AI camps which is really answering what AI is good at and what AI is bad at, and what is the root cause of the weaknesses. Is it intrinsic to the models that LLM's use, or the way thy have been trained. In this knowledge is where the gold mine is for the next start up.

For example AI is very good at answering well defined questions, but suffers from premature closure. It will not know if it has all the information to answer the question. So whilst AI will score better than a doctor or a lawyer on a domain question it will not necessarily gather all the evidence needed to answer the question properly. Knowing this whilst using a LLM is a super power.

There is also a large gap Usability issue in that often the LLM does not really know the humans context, in other words a context collapse. It does not know if you specilised in your domain or just asking for fun.

We should be exploring and debating these gaps.

I think some of this pro/anti AI discussion is really a proxy for something else though. There is another unstated, societal disagreement about personal responsibility and the behaviors of the AI tool users.

It isn't entirely new, as people have always been blaming abstractions which lack agency (society, bureaucracy, "the machine", "the process", "the game", etc.) to conduct themselves in an ethically or logically flawed manner and then shirk responsibility. But, I think it is accelerating dramatically with the way AI agent usage is being adopted. I don't see how this can possibly continue to be normalized without leading to a catastrophic outcome.

Right now, AI tool usage is being applied in ways that would have been considered misconduct, negligence, or fraud if the actors were doing the same thing via traditional outsourcing. People who have a contractual and ethical responsibility to apply analysis, judgement, and oversight are rapidly turning into dumb pipes who just relay content. They essentially phone it in, trying to take credit for apparently moving KPIs "up and to the right". They pretend to themselves and others that they are reviewers and still in control, but when things go wrong they expect forgiveness and for others to lean in and clean up their rapidly spreading mess.

In software circles, many of us are appalled by this slop that is poisoning the well for our existing organizations. It seems like a massive, toxic externality. To us, people aiming their AI agents into our existing collaborations are pretty much flipping a switch to "roll coal", or worse.

AI in theory could be a transformative technology. In practice it's the final product of a self-destructive system that has lost contact with the real world.

Higher-ups are literally saying “This company WILL use AI, you are required to use AI, now go use it whether you want to or not”.

I wonder if this will change now with the model cost explosions that are happening.

And IIRC, the same thing happened to the "oh shit" moment thread you linked to. Did the mods have to intervene to get it back on the front page?

HN might not be anti-AI, but I feel like the way flags are weighted by the ranking allows a minority that is extremely anti-AI to create the impression that it is.

EDIT: And now it's back.

https://hn.algolia.com/?dateRange=all&page=0&prefix=true&que...

I’m happy that my friend who works in plastics manufacturing can move his monstrosity of an Excel spreadsheet to something more predictable and maintainable. I’m deeply annoyed by my coworker who’s trying to put a chatbot in our UI.

I did see your thread earlier today and I admit was pleasantly surprised. Maybe HN is turning over a new leaf? I hope so. I honestly considered switching to X it was getting so bad :P

I know others also do this - though often they are kind enough to auto-fold.

I think this would be a very useful feature to add to HN itself. I nearly emailed dang a few weeks ago to suggest it myself rather than roll my own browser extension to do it, though maybe it's my own responsibility to roll my own software too.

The Gearspace forums (vBulletin based?) have an "Ignore User" feature that helps make that forum vastly more tolerable.

If LLMs were truly able to replace me, I'd be disappointed, sure - I've spent 30 years developing mastery of a craft, it's sad to see it go. But I'd resign myself and move on.

But that's not what makes me mad.

I get angry because I simply don't believe it is true. I have a reasonable math and tech background to where I grok how this stuff works at a fundamental level, and I'm utterly unconvinced that it is performing any sort of reasoning.

Call it a stochastic parrot, a token extrusion machine, whatever, but these things are not thinking or reasoning.

That doesn't mean they aren't useful- they clearly are very useful for many tasks.

My anger comes from the global attempt to replace things that require human reasoning with LLMs.

There's this push to use this stuff way past the point of "helpful accelerator" to "why do we need programmers/doctors/lawyers/etc..."

I think it's incredibly dangerous.

So while the tool is useful, I don't think we've figured out how to use it appropriately. For all the short term appearance of productivity boost it provides (whether this is real in a total-cost sense is still an active question), I think the risks to skill development and quality outweigh those benefits in many cases, and are being overlooked.

It wasn't just that crypto was an obvious grift; it was that you didn't need to be an experienced developer to confirm that 99% of the "web 3.0" nonsense that what was being thrown around literally made no technical sense.

You might reject LLMs on principle, or find that they don't work for you. But I think we're well past any debate of whether they do anything at all, which is exactly where crypto was sitting at peak hype.

And a tool that serves only to subvert the law (at enormous expense) does not strike me as valuable or worth supporting.

NB: I support the outcome of the 1990's crypto war (in the good old days when "crypto" meant cryptography, not crime-token), namely the right to private uncensored unrestricted communication. But a private uncensored unrestricted way of sending around money is a terrible idea, as we can see empirically.

The State Machine Replication problem in distributed computing has been studied since the 1960s or so, and indeed can be solved using Byzantine Broadcast. That has been solved for various settings:

* Permissioned (+ PKI), synchronous: SMR possible for any f, Dolev Strong (1983)

* Permissioned (no PKI), synchronous: SMR impossible if f≥n/3 (PSL 1980, FLM 1985)

* Permissioned, asynchronous setting: SMR impossible even with f=1 (FLP 1985)

* Permissioned, partially synchronous: deterministic SMR with "eventual liveness" possible if f<n/3 (late 1990s), stochastic if f<n/2, impossible otherwise.

The theory to achieve reliable state machine replication even in the presence of byzantine failures was sorted out in the 1980s, and the practice followed in the late 1990s (PBFT, byzantine Paxos, etc.)

So, just to drive the point home: the Byzantine Generals problem (more precisely: Byzantine Broadcast & SMR) was a solved problem by the late 1990s.

What Bitcoin (ie LCR + PoW) added was to do it in the permissionless setting, just as I've said. (It also increased f from n/3 to n/2, modulo "selfish mining", and while earlier solutions were consistent always, available eventually, LCR + PoW is available always, consistent eventually, which is arguably the wrong tradeoff for finance.)

> The lack of CS knowledge on this board is pretty staggering sometimes.

You don't say.

Was it universally obvious? Hindaight is 20/20 There were many block-chain startups funded, and even FAANG got caught up in the hype. FWIW, I was a crypto sceptic, but I had many arguments with believers online and in my social circles. Side note: a crypto enthusiast colleague bought a house off their crypto gains, it may be a grift, but a small number of crypto-believwrs got really wealthy, and you're not going to convince them.

Yes. Crypto was never productive.

But nice that your buddy financed a house with other people's dumb money.

In fact as the act of paying itself has become more restricted, it's often also a good way to make illegal payments for completely legal goods and services.

I was asked to audit a few proposals during that era and in every case I had to go back to the person asking me to say that it was a technical architecture L. Ron Hubbard would have admired. Just straight up making up words in most cases.

Far beyond "wait, is this actually a parody? how is this not a parody?" territory. 80% of them were effectively "tamagotchis + pyramid scheme".

This is why I say that you don't have to love LLMs, but you also can't compare them in terms of also bad. Jay-walking is bad and Jeffrey Dahmer was bad; they are not "both bad".

>Mentioning AI brings out a sharply negative side of HN that I had not seen before 2023

And the parent said,

>I would say that HN was at least as sharply negative during the cryptocurrency craze

And your response was,

>The false equivalency in this explanation is off the scale.

The parent in fact said a very straightforward, non-controversial thing, and you responded in anger, as if they said something like "AI is the same as crypto".

Feel free to respond to what I actually said, though!

More seriously, I think this is a true reflection of a cultural phenomena. All discussions have become more polarized. There is a more of a generational divide in perception and discussion. I would also say there is a loss of nuance.

To complicate this even further there is a real diversity of experiences depending on many factors.

I mean we had flame wars on USENET but somehow it feels to me that most discourse even on controversial topics was civil. When we had tabs vs. spaces flamewars (or whatever the fun topic of the day was) everyone knew they were in a flame war (and often acknowledged that). Or maybe I'm just being nostalgic/biased.

I see the anti-AI sentiments in my work place. I think people are genuinely worried/concerned and don't know how this is going to change our world or even where we are exactly. This is also spilling into adjacent areas where people have strong emotional responses to (the rich, the economy, job market, politics, environment etc.).

The youth are facing an enormous employment crisis. Many have found themselves completely unemployable through no fault of their own.

And then AI leaders go around to commencement speeches to rub it in.

There's no loss of nuance, the situation has just escalated a lot.

This is a good example of anti-AI bias.

This didn’t happen. Out of thousands of commencement speeches across the country, a handful of speakers, none of whom are “AI leaders”, mentioned AI in passing and students booed.

So yeah, I’d say there’s a loss of nuance.

That you fail to understand why those students took such great offense to what was said at those speeches doesn't make it "anti-AI bias".

Those students reacted like that, and I used it as an example, because it's very emblematic of how tech companies and leading figures act.

The absolute closest example is Schmidt, who was last CEO of Google more than a decade before ChatGPT launched, and is now CEO of an aerospace company. No other "AI leaders" gave commencement speeches where they were booed.

So you're blatantly lying and spreading misinformation to feed a negative narrative about the companies and executives in the AI space, then doubling down when called out. Reminds of Vance saying that he's willing to "create stories" in order to draw media attention to his cause-du-jour.

https://en.wikipedia.org/wiki/National_Security_Commission_o...

Pretending he's some AI nobody is a bit goofy.

I think this is probably a combination of nostalgia and/or USENET prior to 1990 or so?

Forums like this, reddit, X, readers of news sites etc tend to be filled with people that don’t have much going on in their lives, have a lot of free time to comment, are less likely to exploit the benefits of AI, and more likely to have simpler skills sets that are replaceable with AI.

Talking to people in real world, I would say the overwhelming majority are excited by AI and interesting in using it more rather than less.

Do those people not deserve to be able to live and survive or are they all just replaceable?

In the end the real problem isn't AI but the way our economic system is set up. For a while it was accepted and worked to some degree that society should be structured to allow capitalists gain maximum profit and then use that money to hire people. AI and robotics will probably break this and we will either have a permanent underclass or we will have to think about a society where all citizens gain from technological progress.

You can't hold up technological progress. If once country doesn't do it, another will.

And no I dont think that generally, but it just seems like so many tech people here have been kinda mid/upper class/well off. I think a core issue around AI, around a new technology, is the class/capitalism issue, because that is the system we are stuck in.

But maybe some also havent quite connected the dots, and are starting to wake up to the reality of the situation.

It probably also gets astroturfed.

The interesting question is, if it's so consistent, how can it go unnoticed for so long (as you've reported) and/or get perceived as one-sided ("HN is so anti-A!") when in fact it is almost always two-sided ("HN users are divided on A")?

I believe there's a clear answer: it's because what you notice depends on how you feel [2]. If you like A, or (more precisely) dislike anti-A, you are far more likely to notice anti-A posts. Not only that, but they will make a stronger impression on you than the median post—even the median pro-A post.

These two factors—frequency of impression and impact of impression—combine to create a strong picture of the site as anti-A—so strong that people often use universals like "always" and "never" and "absolutely" when describing it. In reality, HN is a statistical cloud of datapoints and your pre-existing feelings are what determines which datapoints you notice (i.e. frequency) and how strongly they affect you (i.e. impact). [3]

This is why people with strongly opposing views feel the same about how biased HN is, just in opposite directions! i.e. with the A team convinced there's a strong anti-A bias and the B team equally convinced there's a strong anti-B bias. It's simply (<-- that word again) that their feelings cause them to notice different subsets of the available datapoints. Abstract out the directional bit (pro- or anti-A), and their comments become isomorphic.

Unfortunately for us, HN is more afflicted by this phenomenon than sites which allow sharding, whether by social group (e.g. Twitter's follow lists) or content (e.g. Reddit's subreddits), and thus organize the community into silos [4]. HN is non-siloed, meaning everyone is in the same big community: all the As, all the anti-As, all the Bs, and all the anti-Bs - all roaming the same threads and bumping into each other. You are therefore more likely to run into datapoints you dislike and so more likely to feel that the community is biased against your view, and - what's worse - more biased the more strongly you feel!

For me this is sad, because I believe that in reality HN is a somewhat (<-- let's not exaggerate) more thoughtful and tolerant community than most others of the same or greater size, but precisely because of this dynamic, ends up being perceived as less so. (I wrote a thing about this a few years ago, if anyone is interested: https://news.ycombinator.com/item?id=23308098.)

I believe this is why one so often hears about how toxic, nasty, negative HN is—not that it isn't those things! but the relative level of them gets distorted. Human beings simply can't take very much of what they dislike and disagree with before resorting to generalization and other internal barriers. This is basically an immune response. It often takes only a handful of datapoints (3, or 2, or maybe even just 1) before the impression burns into the retina and becomes a permanent image [5].

(Side note: in one case I saw, some people were agreeing about how terrible and mean Hacker News is, and to prove the point, one of them linked to a vile reply he had received. That reply, however, was from Twitter—not from HN at all! He hastened to add "HN is the same"—somewhat self-refutingly, since if this were true, an actual example would not have been hard to find. In reality, while such vile comments do appear on HN, the community quickly flags most, and moderators eventually flag most of the rest. This is an example of the skew in perception I'm talking about.)

This phenomenon is at its most painful when the topic is close to one's heart—for example, when one's own work is being discussed. In cases like that, it doesn't take much before one feels wounded, and such painful impressions rarely go away.

I write a long digressive post about this once or twice a year because it makes me sad, especially since I've learned that it comes from fundamentals [6] - in this case, how HN's design interacts with human hard-wiring. Because of this, it is not going to change, and cannot be affected by argument.

[1] https://hn.algolia.com/?dateRange=all&page=0&prefix=true&que...

[2] I don't mean "you" personally, of course—everyone does this. It's a double whammy of https://en.wikipedia.org/wiki/Negativity_bias and https://en.wikipedia.org/wiki/Clustering_illusion, sometimes described in this way: https://en.wikipedia.org/wiki/Hostile_media_effect.

[3] Lots of past explanation here if anyone wants it: https://hn.algolia.com/?dateRange=all&page=0&prefix=true&que....

[4] https://hn.algolia.com/?dateRange=all&page=0&prefix=true&que...

[5] https://hn.algolia.com/?dateRange=all&page=0&prefix=true&que...

[6] https://hn.algolia.com/?dateRange=all&page=0&prefix=true&que...

When TypeScript came out and for the first couple of years, I was the biggest TypeScript zealot on the planet. I loved it! There was a point when I typed T that the autosuggestions on iPhone would suggest "Typescript" :-) This all to say: when anti-Typescript sentiment popped up, I definitely noticed it, and it definitely annoyed me. (I still think back on jashkenas saying that he didn't see any point in TS because any good engineer wouldn't make the errors it catches and I just want to throttle him! But I digress...) And there definitely was a lot of it.

But there was a difference in quality between the anti-TS sentiment and the anti-AI sentiment. No one ever attacked my abilities as an engineer for saying I liked TS; no one ever said the things I built with TS were embarrassing or intern-quality like they have with AI. It never devolved into personal attacks the way that anti-AI commentariat pull out when they run out of other arguments.

I'd make a humble suggestion. I would like to suggest that when comments get that derisive, that those comments are removed faster and those users are banned faster. I genuinely think it brings down the quality of discourse site-wide. (FWIW, I think the same about pro-AI incendiary content - very low-quality comments about how people are going to lose their jobs / become obsolete without AI should equally well be flagged, removed and banned.)

I know the HN moderator team is incredibly busy and I am very thankful for all you do!

> I completely am fine with HN being a negative place

That can't be true, or else the personal attacks, derisive and incendiary content you're describing wouldn't bother you!

Also, it isn't possible to treat comment quality and negativity as separate issues because the two are related. Negativity reduces comment quality, for many reasons. One is what when people are angry or frightened they become repetitive and uncreative. Another is that such comments evoke more, and worse, from others.

> I would like to suggest that when comments get that derisive, that those comments are removed faster and those users are banned faster.

No argument there, but this is what we're already doing, or at least trying to do, regardless of the topic. If we knew a more effective way to do it, we'd be there!

FWIW, I think we agree on almost everything, but just a clarification on this point. I am perfectly fine (in fact I delight in reading) if HN is negative in the flavor of "You are completely wrong, here are sources 1 2 and 3 that show you exactly why you are wrong". I have read plenty of anti-AI writing on HN of that form and enjoyed it. OK, I don't love it when it's directed at me, but everyone needs to be told they are wrong sometimes. But I very strongly dislike "Only an inexperienced person could hold a view like that", i.e. personal attacks.

> No argument there, but this is what we're already doing, or at least trying to do, regardless of the topic. If we knew a more effective way to do it, we'd be there!

Godspeed :)

At the same time, cut raw steaks don't alert people in the same way as corpses do, and that is a double standard, in a way, I guess.

A drill is just a tool. You can screw a screw without one. AI is just a tool, you can do things without it, too.

But AI applied in a manner you could not otherwise complete on your own is not a tool. It's a slave. None of the ideas or creations presented in that instance are your own. You just cheated someone out of a paycheck they may have otherwise earned.

The AI discussion on HN is exhilarated by the fact that it can have a tangible impact on people's lives on this forum. You add to this the possibility of a minority extracting an unfathomable amount of wealth from the hype train and we lose all hope for a moderate discussion.

My own views on this are rather boring. After having tried various models, I've reached the conclusion that it adds minor benefits to my workflows, but I don't have to lose sleep from Claude reaching singularity anytime soon..

I think the folks who say 20-40% faster is typical are definitely in the right ballpark. And that's best case. There are plenty of times where my biggest blockers are not anything Claude will ever be able to fix.

Maybe the result of this whole experience will be grudging appreciation from management that software developers aren't just coders. Maybe. I'm an optimist, though. And my manager already knew that.

Now I can work from wherever, whenever, by speaking into my headphones, and have it persist. Sure, I want to be in front of a screen to do final syntax review, but I don't mind planning out architecture and guiding an LLM towards a cohesive result while going for a walk or jog. That's just not a workflow that would have been viable 3 years ago.

I do find it somewhat amusing to read the commentary here and ponder why some of these people are wildly enthusiastic about making software development a dead field but simultaneously don't think it's also going to put them out of work. Programmers may be a lucrative target, but LLMs are "good enough" at many other knowledge disciplines too.

  "Content about AI and LLMs are considered off-topic with the sole exclusion of deeply technical content about implementation."

Typically there's a lot of acknowledgement that problems are coming rapidly, if only in the form of amateurs flooding the domain, but there's also an assumption that their domain knowledge is going to matter, in terms of staying viable in the market. This is far from a safe assumption.

Even if you assume you are ONLY needing to sell products or services to humans, also assuming that humans will continue to thrive and spend money, you're assuming people will be able to identify superior products you assume you can produce and market, and that they won't be misled by false advertising, at scale, generated by other AI to optimize its effectiveness.

If you instead assume the domain experts are going to need to sell to AIs (either as owners of wealth and power, or as the agents of the humans we're assuming will still thrive and spend money) you're making a hell of a big assumption that the agents will be trustworthy market participants with perfect information able to evaluate stuff in the best traditions of market capitalism. Humans don't manage that, even when they're a lot smarter. If this assumption fails, all the money goes to whatever exploits the AI agents most effectively, also at scale. At that stage there is no use being a domain expert in anything but hacking people's AI agents to trick or puppet the agents to do your bidding. Anything else is wasted effort.

Sometimes I also see pushback along the lines of programmers criticizing the quality of the AI coding itself. This is gonna be domain-dependent and it's fair to counter it through developing less fragile languages, but we don't know that'll work as an answer. There's a lot of handwaving.

I'm pretty sure people's experience of AI, as a rule, is the bad, damaging forms, where it's burning stupid amounts of energy to make everything worse while openly promising to take away everyone's ability to work and replacing it with the need to pay the AI owners a sort of extra tax just to be able to exist and think. The HN crowd is WILDLY pro-AI compared to the general population.

It's not going to look any better in future for the general population. What I've described leads to 'AI' increasingly producing shockingly bad outcomes for people in general, through simple market dynamics. It's going to be way more profitable to produce the bad outcomes.

I expect the people in here to be domain experts, understand simple concepts like closed loop water cooling, deterministic vs non-deterministic systems, maybe some basic concept of how a GPU and vector math works and most notably the exponential pace that it’s becoming both more capable and more efficient.

Unfortunately, like OP that’s not the case and it’s the same talking points I could read in my local paper. Then everyone’s talking points change in unison like they are waiting on the latest instructions from headquarters.

But the opposite is also true: shallow basic knowledge about almost anything, instantly available in a form that directly applies to one's problem, is an incredible resource. Not only can you get pretty far on that alone, it also gets you moving in a way that inertia would otherwise have prevented. That's my experience at least.

"Exponential" is very clearly not true.

1. Comments should be thoughtful even if not coming from a place of expertise

2. Better comments have something useful to share behind the passion

3. "Me too" comments are usually not as interesting to read and rarely add to the discussion

I have personally been both praised and attacked for making detailed, reasoned comments. It's more about the mood and attitude of the reader(s) that determines which way it goes.

AI is a complicated issue because it is creating change very quickly--as many predicted it would 20-30 years ago--and that change is still accelerating.

In your words, the change is exponential--and this may indeed be demonstrated by the performance data over the past several years. It's hard to contemplate or predict when that change will plateau, but we definitely aren't there yet.

The ultimate impact on the world society is likely to be profound. Whether you like it or hate it, it is happening. Those who lose a job to AI may find that their mother is saved by a medicine or doctor aided by AI. It's complicated.

Perhaps there are ways to legislate compromises into its use or expansion in some parts of the world--but it is likely a technology that has at least as much positive potential for humanity as it does negative--so legislation is going to have to address the good and the bad and no country is an island unto itself.

Thanks for the opportunity to comment.

You might have seen some comments by me receiving votes lately that some might classify as anti-AI. I'm not for banning it, I use it at work and at home, I learn about it.

Here a few positions I hold, for example:

- There is a fundamental deviation between what we should be seeing if we were surrounded by 100x enabled engineers and reality. We're not seeing previously untackable, complete open source projects pop up everywhere like someone coding an open-source, iphone-compatible OS in a year, or companies providing 10x more. Just POCs and small apps. LLMs have been around enough that this is pointing towards inflated claims of success, even if they are actually useful, which I'm not denying.

- LLMs provide users with a strong psychological reward (making mental workload disappear). They do so only sometimes, in a chance-based outcome. Anyone with a passing interest in psychology should realize how similar that is to the mechanics of gambling, and thus how risky it is that a user misjudges when it is reasonable to use them. Mind that I'm not saying that the tool isn't worth it, just pointing at a source of major deviation between perceived and actual outcomes that few people consider.

- There are a lot of signals that humans rely on that are broken by LLMs. "Well formatted text -> text written with careful consideration" no longer works. "Large document -> significant effort" does not hold. "Good grammar -> educated speaker" is broken as well. "decent code practices -> the PR is safeish to approve" no longer true. Some of these barriers being broken can be enablers for people, but on the whole this is going to disrupt society in fundamental and unpredictable ways.

- I think the industry is drawing unreasonable and dangerous conclusions from the advent of AI. As some commenters pointed out, if code generation is now cheap we should be seeing engineers freed to deal with non coding tasks like automated QA, user research, architecture or design, and being more able to handle bug resolution for example. We are instead seeing a push for _creating code faster_, and proposing ignoring tasks like review and quality control in pursue of speed, which is fundamentally inconsistent with speed being less of a problem. To use a flawed analogy, if your car is now 10x as fast you should be putting way more attention to how you steer, rather than asking everyone to go pedal to the metal.

- LLMs products have the potential to be extremely user hostile if enshittified. We could have probabilistic insertion of promoted material. We could have subtle political steering of people. We could have a model's performance reduced without much SLA recourse. We are not tackling those issues before they appear when it is obvious that they will appeare, and society will pay the price.

If you read with attention you'll see that no point of mine is arguing against AI usage. I don't want to bury my head in the sand and pretend LLM's don't exist or are useless. I don't want to ban them. I'm just not willing to fully allign with marketing speech and turn my brain off.

Tbf I think this is probably the way most people see it at the moment from my discussions with others.

The static comes from people who have essentially staked their entire reputation on AI one way or another.

I have a suspicion that online, those are mostly bots.

Your comment reminds me of another important variable: time. We're still at the beginning of figuring out what these tools even are, let alone how best to use them.

That's a fair point, and I imagine it's going to take a long while for the dust to clear.

People are operating on semi-dark private implementations, where the approach that works today breaks tomorrow. It's like SEO, where people operate on assumptions of google's algorithm that are not provable or guaranteed to hold, so everything gets cargo culty.

Openai, anthropic et all are not fully to blame for this, I think. even if they wanted to be more transparent, it's hard to imagine what kind of determinism they could provide users with, since the product is a probabilistic emergent property more than purposeful design.

Then there's a flood of marketing and astroturfing that seeps into technical discussions, in a way that the engineering world is not really used to (other than the crypto space, that was relatively contained).

It's a really hard, multifaceted and interesting problem to solve. I just wish I saw more cold headed attempts to approach these issues scientifically.

I can tell what HN is Anti though, in 80/ 90%+ if not higher. Ads, Oracle, Facebook, MPEG Codec, MySQL,

Edit: scratch that - I was going to link you to https://news.ycombinator.com/item?id=46018271 but it turns out you were the user I was replying to there! I don't suppose repeating it would help.

I can speak from myself as an example (although n=1): I am incredibly open to machine learning and the advances it brings. On the other hand I am extremely conscious of the fact that the current LLMs do often write bad code, which becomes especially obvious once projects go much beyond "private toy" size. For me personally the consequence is that I try to make my projects even more modular, and the modules even more clearly delineated. With proper guidance, current LLMs work mostly fine when working on isolated modules. That doesn't mean they don't sometimes fail even there, but they also on rare occasions come up with surprisingly clever solutions when you let them loose on a code base, as long as the problem you want them to fix is mostly isolated in a couple of modules.

So long story short, you can be all for LLMs and still be conscious of their shortcomings, and that just vibe coding applications you intend to let loose on unsuspecting customers is probably a bad idea, and possibly outright immoral. We have already seen more than enough examples of vibe coded applications dumping sensitive user data into the lap of anyone who is inclined to pick up a stick and prod them a little bit.

We tend to ignore that reality but it hangs above all these discussions like a dirty cloud

This is just an object lesson for everyone who thought all that software guys did was write code that there is just a little bit more to the job than that. Actual developers should already have known.

I've spent an obscene number of hours learning how to get reliably good quality code out of these things. I'm actually very happy with where the tech is right now and can't imagine ever going back to typing code by hand.

But I absolutely hate how companies and society at large are acting because of this stuff. It feels like all rationality has flown out the window. So I'm just staying in my sandbox with my little toys and hoping the mass psychosis blows over at some point.

It’s also strangely bad at correctness. You can ask it to write unit tests for a project. Unless you’re careful with your prompting, it will only write the unit tests that it knows will pass.

Generally, these models are amazing tools. But they cannot be trusted to make correct, maintainable software. At least not yet. Maybe in another year or two.

What's really fun about these tools is that this is both true and false!

If you ask these tools to reason about things in-the-large you often get very useful information back out of them.

I've asked about refactors I was thinking about doing, and gotten accurate and useful information back, which as been AMAZINGLY helpful for avoiding "let me start doing this, and then realize 4 hours in that it's not gonna work as well as I hoped" traps.

But it's a very attention-on-one-thing-at-a-time thing. IMO this is fairly inherent to the models, but people have been doing great work making the tooling around them smarter in terms of how to break up tasks ahead of time to compensate, so I'm not gonna say it won't get materially better.

So if you prompt it to do a task in a certain way, especially in a "plan mode" type of usage, you can get a pretty solid recipe + execution of a properly-designed implementation of that task.

But if you're not opinionated and checking in frequently, you're gonna get the sorta median-approach or random-luck-output-of-the-day decision. And so the human-in-the-loop point is unlikely to go away as long as the human has more context. Even if it's half-baked or not-fully-realized intuition about how the code is likely to evolve in the future that you don't put into every prompt.

> It’s also strangely bad at correctness. You can ask it to write unit tests for a project. Unless you’re careful with your prompting, it will only write the unit tests that it knows will pass.

My hunch is that this is the same fundamental problem. When it's attention is fully on "produce the next string of code" the parts of the context that relate to the broader system goals are NOT being considered as much for the output. So you get things like this, even with latest Opus still, when dealing with hard-to-isolate-in-a-single-test bugs (esp when it comes to multi-service call sequences or concurrent code):

- "we need to fix this bug across eight methods in three files"

- "I found the spot! we need to do [blah blah blah]"

- "great, implement that plan"

- "I've done it!"

- "wait a sec... you moved some of the sequencing around, but didn't actually fix the fundamental issue"

- "you're right! i moved [xyz] into [func b] instead of [func a] since it needed to be called later, but actually it needs to be after [func c] since it depends on the output of func b!"

When asked about correctness it's good enough at "reasoning"-style output to spot these issues, but when generating code it's in such a pure "predict plausible code sequences" mode that this can get lost.

Its weird, working with LLMs. There are some things the LLMs are extremely good at doing autonomously. Like, reverse engineering, or reading documentation (and using that knowledge in other areas). There are things that they can do - but you need to explicitly prompt. Like, I've found Opus is quite good at optimising code. I've had a lot of success by asking it to write benchmarks (and do profiling), and use that data to improve the performance of some piece of code. Thats often enough to get quite large performance improvements. You can get even further by showing it similar code others have written which is well optimised. It's very good at copying optimisation ideas from one project to another.

But then there are very simple things it really struggles to do. Some kinds of correctness testing. Invariants. System design.

Is it bad at that stuff, or do I just need to figure out how to prompt the LLM? To return to the topic of this thread, I think we're seeing a lot of different opinions on LLM generated code for 3 reasons:

1. Some people aren't looking at claude's output at all. Some people are looking at the code and it looks fine to them. And some people (with more experience writing software) are looking at the code and judging it poorly.

2. We all prompt our LLMs very differently! It turns out that you get really different results based on how you prompt the machine. We're all figuring this out together. Some people have better instincts than others.

3. We're working on different projects. Claude is comparatively much better at end-user facing software. Its great at making a standalone website. Its much less good at finding and fixing obscure bugs in large, established pieces of software. If you work in consulting, LLMs can already do a lot of your job. If you work on Chrome or Unreal or the windows kernel, maybe not so much.

I think there's definitely a bit of both. Some things are easier to prompt "adequately." Some domains or types of requests are tougher.

The billion dollar question (well, trillion dollar, looking at the valuations of OpenAI and Anthropic) is will that change enough to actually replace the highly-paid people who currently are needed to make sure shit doesn't go sideways? They're betting that they can solve the "turn a bad prompt into a good-enough series of prompts" problem generically for everyone.

And where exactly that lands could have a lot of knock-on effects. The easy targets are things like SaaS that is only valuable because of economies of scale but the problems are "simple" if you don't have that scale.

But even there, there's a lot of echoes of the past, where things like ad-hoc Access apps or spreadsheets powered (or still power!) a bunch of business processes in lieu of SaaS products. How much appetite is there long-term for large businesses to really go back to owning all that in-house?

Also a fun irony in that the trillion-dollar-valuation world is basically "the biggest SaaS of them all" and those companies have a huge target on them and at that point practically everyone else in the world would be gunning for them. If they do find that "good enough" point, they also have to hope that nobody can replicate it for less anytime soon... (but they'll also have given those folks aiming at doing just that a great tool for helping build those systems).

For those sorts of systems, comments like this just remind me of people pitching UML + codegen-or-outsourcing 20 years ago.

> Users don’t care whether the code was written by AI or by hand, or which framework you used. They care that the product works.

> I say this as someone who has spent more than 20 years honing their craft as a software engineer.

I've worked with outsourced code, and I've worked with similarly-messy/not-forward-thinking first-iteration startup-MVP code that found success and now needs to do more and be more reliable while being cheaper to operate. It ain't pretty.

Once you start modifying working code that people rely on you quickly start to see that the code itself matters. There aren't enough tests in the world to get around this from a quality POV. And piles of copypasta turn into either inconsistent behaviors (bad from a user satisfaction POV) or a continual drag on velocity (bad from a product goals/competitivness POV). And I've run the tools on those sorts of "found MVP, now need to iterate fast" codebases, both on the messy parts and the better parts.

Sure, Claude can chew through the pain points of "let me find the ten copypasta versions of this and try to update them all" and chase down all the weird stupid bugs/janky things that that requires because each of them evolved separately much faster than a human can. But it's not gonna be as fast as it was at writing the first version, and it's gonna get more and more annoying. You hit the point where your agent is churning for hours and running more and more tests and inadvertently breaking more things trying to make the change in all those places, and making MORE code changes to fix those (and those are just the caught broken things the tests cover!).

It's wayyyyy faster and less potentially painful to have it make updates to a well-factored module. "You don't have to read the code" is all well and good until you have a 10k+ line PR that's 10x larger than it needs to be because nobody read the code in the past, and you realize that most of the relevant test files also changed substantially and you don't have that much known-unchanged permutation coverage of the actual things users do... how comfortable are you pushing the "ship it" button then?

I've had this experience with vibe-coded applications written by Claude itself. For some reason, Claude doesn't seem to use any good practices unless I tell it to, and seems to require me to walk it through a decent design up front.

But also, the algorithms push enragement equals engagement and that favors pushing content to the product (us) that enrages the product to drive up their engagement numbers so they click on more ads.

Meanwhile, the medical applications of AI alone are astounding. But that doesn't get people mad so it doesn't get promoted as near as much as the latest clueless word salad from a rich dude. Election season is going be to very interesting this year IMO.

https://erictopol.substack.com/p/the-paradox-of-medical-ai-i...

What's my position? GenAI will be subsumed into media toolchains. And reasoning is clearly no longer a hallmark of sentience or consciousness. That's going to take some time to sink in. Token costs are only going to drop further from here: https://www.navyaai.com/reports/ai-cost-report-token-prices-.... And there's opportunity to build businesses to bring down the rest of the costs in the article I cited.

Also, I'm with Martina Scorsese and Dua Lipa on GenAI. And having worked in and around Hollywood in the past, it blows my mind that such a corrupt oligarchy would be considered worth protection. Above the lines thrive while below the lines would be borderline homeless without the guilds. And the ones outside the guilds are.

Finally, build datacenters in the Dakotas, the Texas panhandle, and any other place far away from people where power can be generated onsite by whatever means suit the location best that's not coal.

And medical advances of AI are from other ML algorithms, not LLMs. They use similar technology and can run in the same DCs, but they aren't the same and people don't really have the same opinions on them both.

But sensing the free market pitchforks and torches, the pushback commenced shortly thereafter...

But you're right that this can be addressed in multiple ways such as pushing the locations to places where construction is cheap or building where it is made cheap to operate with on-site power generation.

What's your take on medical VLMs? That's pretty much an LLM with multiple classes of tokens. While nothing's FDA approved yet, Cognita CXR is listed as a breakthrough device and multiple companies providing conventional AI have both published research and signaled intent to submit for FDA approval.

https://mosaicclinical.ai/news/2026/03/mosaic-clinical-techn...

E.g., agentic development is great for solo projects and prototyping, but can become overwhelming alarmingly quickly when you’ve got multiple devs involved, especially if they’re not all super-disciplined and consistent about how they use their agents. And we’re all learning how to deal with that.

The true state of things is that the anti-AI folks are FAR more vocal about their opposition, meanwhile, the people who like AI are busy being productive and accomplishing a truly staggering amount of work.

The 84% stack overflow "currently use AI" number from a _year_ ago almost certainly still holds true today. Just look at the absurdly long and very incomplete list of "tainted by AI" projects that the list-making activists are maintaining over on Codeberg. Software is not "divided", it is simply being occupied by angry protesters.

Hacker news skews significantly older and whiter, and the blogs and such that circulate come primarily from the "established" English-speaking old-guard "white people" software days. Opposition to AI is highly concentrated among the white, first world, left-wing, established engineers.

This group punches far their weight in terms of the volume of their activism compared to the actual percentages of humans on earth.

That's all it is. Anti-AI hatred is just a noisy echo-chamber of those who are privileged enough to not need to modernize their skills.

You make a good point about the asymmetry in content: the negative ones are mostly just critical, and often denunciatory, while the positive ones are less likely to be generic, and more likely to be about specific work people are doing. That's not a new dynamic, though, and long pre-dates what today is called AI.

(and a different person will do a similar sampling technique but come away with the opposite conclusion, that HN is super pro-AI)

This is why neglecting statistics in favor of vibes is a bad idea.

[1] https://news.ycombinator.com/item?id=48035600

[2] https://news.ycombinator.com/item?id=48035860