Also when you use Google via an alternative frontend like startpage you get a slightly better version without the crap than the one currently serving at google.com. If you use Kagi which most of the time only shows results from the google search index it becomes the best quality Google ever had.

I believe it is really only the "frontend" that is responsible for the current quality and very much an intentional choice. If Google wants to be better, it could choose so tomorrow. Actually it doesn't, because it already does today, just not on a modern browser on google.com .

Literally every shitty corporate behaviour is amplified by this technology fad.

Can you elaborate on this point?

At least lawyers can lose their bar license.

We had a discussion in a group chat with some friends about some random sports stuff and one of my friends used ChatGPT to ask for some fact about a random thing. It was completely wrong, but sounded so real. All you had to do was to go on wikipedia or on a website of the sports entity we were discussing to see the real fact. Now considering that it just hallucinated some random facts that are on Wikipedia and on the website of an entity, what are the chances that the legal advice you will get will be real and not some random hallucination?

AI is just a really good bullshitter. Sometimes you want a bullshitter, and sometimes you need to be a bullshitter. But when your wealth are at risk due to lawsuits or you're risking going to prison, you want something rock solid to back your case and just endless mounds of bullshit around you is not what you want. Bullshit is something you only pull out when you're definitely guilty and need to fight against all the facts, and even better than bullshit in those cases is finding cases similar to yours or obscure laws that can serve as a loophole. And AI, instead of pulling out real cases, will bullshit against you with fake cases.

For things like code, where a large bulk of some areas are based on general feels and vibes, yeah, it's fine. It's good for general front end development. But I wouldn't trust it for anything requiring accuracy, like scientific applications or OS level code.

In fact, giving RBAC functionality to middle managers would be a key component to any strategy for AI deployment activity. You want traceability/auditability. Give the middle managers charts, logs, visibility into who is using what data with which LLMs. Give them levers to grant and deny access. Give them metrics on outcomes from the use of the data. This could eve even make legal happy.

This missing layer will exist, its just a matter of time.

Ask it about a torque spec for your car? Yup, wrong. Ask it to provide sources? Less wrong but still wrong. It told me my viscous fan has a different thread than it has. Would I have listened, I would've shredded my thread.

My car is old, well documented and widely distributed.

Doesn't matter if claude or chatgpt. Don't get me started on code. I care about things being correct and right.

At this point I literally spend 90% of my time fixing other teams AI ‘issues’ at a fortune 50.

1. Piss-poor at the brainstorming and planning phase. For the compression thing I got one halfway decent idea, and it's one I already planned on using.

2. Even worse at generating a usable project structure or high-level API/skeleton. The code is unusable because it's not just subtly wrong; it doesn't match any cohesive mental model, meaning the first step is building that model and then figuring out how to ram-rod that solution into your model.

3. Really not great at generating APIs/skeletons matching your mental model. The context is too large, and performance drops.

4. Terrible at filling in the details for any particular method. It'll have subtle mistakes like handling carryover data at the end of a loop, but handling it always instead of just when it hasn't already been handled. Everything type checks, and if it doesn't then I can't rely on the AI to give a correct result instead of the easiest way to silence the compiler.

5. Very bad at incorporating invariants (lifetimes, allocation patterns, etc) into its code when I ask it to make even minor tweaks, even when explicitly promoted to consider such-and-such edge case.

6. Blatantly wrong when suggesting code improvements, usually breaking things, and in a way you can't easily paper over the issue to create something working "from" the AI code.

Etc. It just wasn't well suited to any of those tasks. On my end, the real work is deeply understanding the problem, deriving the only possible conclusions, banging that into code, and then doing a pass or three cleaning up the semicolon orgasm from the page. AI is sometimes helpful in that last phase, but I'm certain it's not useful for the rest yet.

My current view is that the difference in viewpoints stems from a combination of the tasks being completed (certain boilerplate automation crap I've definitely leaned into AI to handle, maybe that's all some devs work on?) and current skill progression (I've interviewed enough people to know that the work I'm describing as trivial doesn't come naturally to everyone yet, so it's tempting to say that it's you holding your compiler wrong rather than me holding the AI wrong).

Am I wrong? Should AI be able to help with those things? Is it more than a ~5% boost?

A good problem to throw at AI, I thought. I handed the tools to a SOTA model and asked it to generate me some files. Garbage. Some edits to the prompts and I still get garbage. Okay, that's pretty hard to generate a binary with complex internal structure directly. Let's ask it to tell me how to make the toolchain generate these for me. It gives me back all sorts of CLI examples. None work. I keep telling it what output I am getting and how it differs from what I want. Over and over it fails.

I finally reach out to somebody on the toolchain team and they tell me how to do it. Great, now I can generate some valid files. Let's try to generate some invalid ones to test error paths. I've got a file. I've got the spec. I ask the LLM to modify the file to break the spec in a single way each time and tell me which part of the spec it broke each time. Doesn't work. Okay. I ask it to write me a python program that does this. Works a little bit, but not consistently and I need to inspect each output carefully. Finally I throw my files into a coverage guided fuzzer corpus and over a short period of time it's generated inputs that have excellent branch coverage for me.

What would effective have looked like to you in this situation?

But sometimes good data is also bad data. HIPAA compliance audit guides are full of questions that are appropriate for a massive medical entity and fully impossible to answer for the much more common small medical practice.

No AI will be trained to know the latter is true. I can say that because every HIPAA audit guide assumes that working patient data is stored on practice-owned hardware - which it isn't. Third parties handle that for small practices.

For small med, HIPAA audit guides are 100 irrelevant questions that require fine details that don't exist.

I predict that AI won't be able overcome the absurdities baked into HIPAA compliance. It can't help where help is needed.

But past all that, there is one particularly painful issue with AI - deployment.

When AI isn't asked for, it is in the way. It is an obstacle to that needs to be removed. That might not be awful if MS, Google, etc didn't continually craft methods to make that as impossible as possible. It smacks of disdain for end users.

If this one last paragraph wasn't endlessly true, AI evangelists wouldn't have so many premade enemies to face - and there would be less friction all around.

It is shoved everywhere but nobody really needs it.

The thrashing wails of people who've spent billions on something that they should have done more due diligence on and don't want to write off their investment.

Yes, drastically. This means I'll have to wear Zuck's glasses I think, because the AI currently doesn't know what was discussed at the coffee machine or what management is planning to do with new features. It's like a speed typing goblin living in an isolated basement, always out of the loop.

Extrapolation. I see the progress I already made over the last years.

For small tasks where I can anticipate that AI will handle it well, I am already multiple times more efficient with AI than without.

The hard thing to tackle these days is larger, more architectural tasks. And there I also see progress.

Humans also benefit from a better codebase that is easier to understand. Just like AI. So the changes I make in this regard are universally good.

That is assuming that it is really a force multiplier which is not totally evident at this point.

If it's a multiplier you need to either increase the requested work to keep the same humans or reduce the humans needed if you keep the same workload. It's not straightforward which way each business will follow.

Most of my social circle are non-technicial. A lot of people have had a difficult time with work recently, for various reasons.

The global economic climate feels very precarious, politics is ugly, people feel powerless and afraid. AI tends to come up in the "state of the world" conversation.

It's destroying friends' decade old businesses in translation, copywriting and editing. It's completely upturned a lot of other jobs, I know a lot of teachers and academics, for example.

Corporate enthusiasm for AI is seen for what it actually is, a chance to cut head count.

I'm into AI, I get value out of it, but decision makers need to read the room a bit better. The vibe in 2025 is angry and scared.

It’s also worth noting that while our modern use of Luddite is simply “anti-technology”, there was a lot more going on there. The Napoleonic wars were savaging the economy and Luddism wasn’t just a reaction to the emergence of a new technology but even more the successful class warfare being waged by the upper class who were holding the line against weavers attempts to negotiate better terms and willing to deploy the army against the working class. High inflation and unemployment created a lot of discontent, and the machines bore the brunt of it because they were a way to strike back at the industrialists being both more exposed and a more acceptable target for anyone who wasn’t at the point of being willing to harm or kill a person.

Perhaps most relevant to HN is that the weavers had previously not joined together to bargain collectively. I can’t help but think that almost everyone who said “I’m too smart to need a union” during the longest run of high-paying jobs for nerds in history is going to regret that decision, especially after seeing how little loyalty the C-suite has after years of pretending otherwise.

Yes, network effects and hyper scale produce perverse incentives. It sucks that Wikipedia can be gamed. Saying that, you'd need to be actively colluding with other contributors to maintain control.

Imagining that AI is somehow more neutral or resistant to influence is incredibly naive. Isn't it obvious that they can be "aligned" to favor the interests of whoever trains them?

Yeah u can download the entirety of Wikipedia if you want to. What's the single point of failure?

So makes sense that SO like users will use AI, not to mention they get the benefit of avoiding the neurotic moderator community at SO.

But yes, AI put the nail in its coffin. Sadly, ironically, AI trained off it. I mean AI quite literally stole from Stackoverflow and others and somehow got away with it.

That's why I don't really admire people like Sam Altman or the whack job doomer at Anthropic whatever his name is. They're crooks.

I use it in much the same way you describe, but I find that it doesn't save me that much time. It may save some brain processing power, but that is not something I typically need saving.

I extract more from LLM asking it to write code Infind tedious to write (unit tests, glue code for APIs, scaffolding for new modules, that sort of thing). Recently I started asking it to review the code I write and suggest improvements, try to spot bugs and so on (which I also find useful).

Reviewing the code it writes to fix the inevitable mistakes and making adjustments takes time too, and it will always be a required step due to the nature of LLMs.

Running tasks simultaneously don't help much unless you are giving it instructions that are too general that will take it a long time executiny - and the bottleneck will be your ability to review all the output anyway. I also gind that the broader is the scope of what I need it to do, the less precise it tends to be. I achieve most success by being more granular in what I ask of it.

My take is that while LLMs are useful, they are massively overhyped, and the productivity gains are largely overstated.

Of course, you can also "vibe code" (what an awful terminology) and not inspect the output. I find it unacceptable in professional settings, where you are expected to release code with some minimum quality.

For Boilerplate code we need an AI that is less creative and more secure and predictable. I have fun creative a system design with the right tables and I have implementing logic and interaction design. I don't have the biggest fun writing down dtos and entities.

I would need an AI, that can scan an image and just build me the right lego bricks. We are just getting back to an machine that can do UML from less precise sources.

It's hit or miss at the moment, but it's definitely way more than "UML code generators".