[sbt]

I have been using it for coding for some time, but I don't think I'm getting much value out of it. It's useful for some boilerplate generation, but for more complex stuff I find that it's more tedious to explain to the AI what I'm trying to do. The issue, I think, is lack of big picture context in a large codebase. It's not useless, but I wouldn't trade it for say access to StackOverflow.

My non-technical friends are essentially using ChatGPT as a search engine. They like the interface, but in the end it's used to find information. I personally just still use a search engine, and I almost always go to straight to Wikipedia, where I think the real value is. Wikipedia has added much more value to the world than AI, but you don't see it reflected in stock market valuations.

My conclusion is that the technology is currently very overhyped, but I'm also excited for where the general AI space may go in the medium term. For chat bots (including voice) in particular, I think it could already offer some very clear improvements.

[oezi]

One of the issues certainly is that Stackoverflow is absolutely over. Within the last twelve months the number of users just fell off a cliff.

[danbruc]

That might be a good thing after all, at least in a certain sense. Stack Overflow has been dying for the last ten years or so. In the first years there where a lot of good questions that were interesting to answer but that changed with popularity and it became an endless sea of low effort do my homework duplicates that were not interesting to answer and annoying to moderate. If this now gets handled by large language models, it could maybe become similar to the beginning again, only those questions that are not easily answerable by looking into the documentation or asking a chat bot will end up on Stack Overflow, it could be fun again to answer questions on Stack Overflow. On the other hand if nobody looks up things on Stack Overflow, it will be hard to sustain the business, maybe even when downscaled accordingly.

[birn559]

Is it really dying? There's only so much growth possible without having it flooded with low effort stuff. They try to instead grow by introducing more topics, but that's limited as well.

I personally didn't use it so much (meaning: Writing content) because it always felt a bit over engineered. From what I remember the only possible entry point is writing questions that get up voted. Not allowed to even write comments, vote and of course not allowed to answer questions. Maybe that's not correct but that has always been my impression.

In general stack exchange seems to be a great platform. I think "it's dying" has an unfortunate connotation. It's not like content just vanishes, just the amount of new stuff is shrinking.

[danbruc]

Maybe it changed, but you could ask and answer questions right after signing up, just commenting, voting, and moderation rights required a certain amount of points. I only answered questions and was among the top 100 contributors at my peak activity but that was at least a decade ago.

I stopped contributing when the question quality fell of a cliff. Existing contributors got annoyed by the low effort questions, new users got annoyed because their questions got immediately closed, it was no longer fun. There were a lot of discussions on meta how to handle the situation but I just left.

So admittedly things might have changed again, I do not really know much about the development in the last ten or so years.

[Izkata]

That's still how it works: https://stackoverflow.com/help/privileges

If you're logged out it's just a list, but if you're logged in it'll grey out and put a checkmark next to the ones you have.

[tom_m]

To be honest, I used Stackoverflow less and less over the years. Not sure that was because I learned more. I just think most times I went there I was looking for snippets to save time with boilerplate. As better frameworks, tools, packages, languages came along I just had less need to go to Stackoverflow.

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.

[lithos]

SO sold off their own data and made it insanely web-crawlable.

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.

[logicchains]

>I have been using it for coding for some time, but I don't think I'm getting much value out of it.

I find this perspective so hard to relate to. LLMs have completely changed my workflow; the majority of my coding has been replaced by writing a detailed textual description of the change I want, letting an LLM make the change and add tests, then just reviewing the code it wrote and fixing anything stupid it did. This saves so much time, especially since multiple such LLM tasks can be run simultaneously. But maybe it's because I'm not working on giant, monolithic code bases.

[surgical_fire]

> I find this perspective so hard to relate to. LLMs have completely changed my workflow; the majority of my coding has been replaced by writing a detailed textual description of the change I want, letting an LLM make the change and add tests, then just reviewing the code it wrote and fixing anything stupid it did.

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.

[logicchains]

>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.

Yep but this is much less time than writing the code, compiling it, fixing compiler errors, writing tests, fixing the code, fixing the compilation, all that busy-work. LLMs make mistakes but with Gemini 2.5 Pro at least most of these are due to under-specification, and you get better at specification over time. It's like the LLM is a C compiler developer and you're writing the C spec; if you don't specify something clearly, it's undefined behaviour and there's no guarantee the LLM will implement it sensibly.

I'd go so far as to say if you're not seeing any significant increase in your productivity, you're using LLMs wrong.

[surgical_fire]

> I'd go so far as to say if you're not seeing any significant increase in your productivity, you're using LLMs wrong.

It's always the easy cop out for whoever wants to hype AI. I can preface it with "I'd go so far as to say", but that is just a silly cover for the actual meaning.

Properly reviewing code, if you are reviewing it meaningfully instead of just glancing through it, takes time. Writing good prompts that cover all the ground you need in terms of specificity, also takes time.

Are there gains in terms of speed? Yeah. Are they meaningful? Kind of.

[dinfinity]

GPs main point is that you (need to) learn to specify (and document) very, very well. That has always been a huge factor for productivity, but due to the fucked up nature of how software engineering is often approached on an organisational level we're collectively very much used to 'winging it': Documentation is an afterthought, specs are half baked and the true store of knowledge on the codebase is in the heads of several senior devs who need to hold the hands of other devs all the time.

If you do software engineering the way you learned you were supposed to do it long, long ago, the process actually works pretty well with LLMs.

[surgical_fire]

> GPs main point is that you (need to) learn to specify (and document) very, very well. That has always been a huge factor for productivity

That's the thing, I do. At work we keep numerous diagrams, dashboards and design documents. It helps me and the other developers understand and and have a good mental model of the system, but it does not help LLMs all that much. The LLMs won't understand our dashboard or diagrams. It could read the design documents, but it wokdn't help it to not make the mistakes it does when coding, and definitely would not reduce my need to review the code it produces.

I said before, I'll say it again - I find it unacceptable in a professional setting to not review the code LLMs produce properly, because I have seem the sort of errors it produces (and I have access to the latest Claude and Gemini models, that I understand to be the top models as of now).

Are they useful? Yeah. Do they speed me up? Sort of, especially when I have to write a lot of boring code (as mentioned before, glue code for APIs, unit tests, scaffolding for new modules, etc). Are the productivity gains massive? Not really, due to the nature of how it generates output, and mainly due to the fact that writing code is only part of my responsibilities, and frequently not the one that takes up most of my time.

[dwaltrip]

Do you have any example prompts of the level of specificity and task difficulty you usually do? I oscillate between finding them useful and finding it annoying to get output that is actually good enough.

How many iterations does it normally take to get a feature correctly implemented? How much manual code cleanup do you do?

[rurp]

If you ever end up working on large complicated code bases you'll likely have an easier time relating to the sentiment. LLMs are vastly better at small greenfield coding than for working on large projects. I think 100% of the people I've heard rave about AI coding are using them for small isolated projects. Among people who work on large projects sentiment seems to range from mildly useful to providing negative value.

[sandworm101]

Fun test: ask chatgtp to find where Wikipedia is wrong about a subject. It does not go well, proving that it is far less trustworthy than wikipedia alone.

(Most AI will simply find where twitter disagrees with Wikipedia and spout out ridiculousness conspiracy junk.)

[tom_m]

It is over hyped for sure. This is among the biggest hype cycles we've seen yet. When it bursts, it'll be absolutely devastating. Make no mistake. Many companies will go out of business, many people affected.

However. It doesn't mean AI will go away. AI is really useful. It can do a lot actually. It is a slow adoption because it's somehow not the most intuitive to use. I think that may have a lot to do with tooling and human communication style - or the way we use it.

Once people learn how to use it, I think it'll just become ubiquitous. I don't see it taking anyone's job. The doomers who like to say that are people pushing their own agenda, trolling, or explaining away mass layoffs that were happening BEFORE AI. The layoffs are a result of losing a tax credit for R&D, over hiring, and the economy. Forgetting the tax thing for a moment, is anyone really surprised that companies over hired?? I mean come on. People BARELY do any work at all at large companies like Google, Apple, Amazon, etc. I mean that not quite fair. Don't get me wrong, SOME people there do. They work their tails off and do great things. That's not all of the company's employees though. So what do you expect is going to happen? Eventually the company prunes. They go and mass hire again years later, see who works out, and they prune again. This strategy is why hiring is broken. It's a horrible grind.

Sorry, back to AI adoption. AI is now seen by some caught in this grind as the "enemy." So that's another reason for slow adoption. A big one.

It does work though. I can see how it'll help and I think it's great. If you know how everything gets put together then you can provide the instructions for it to work well. If you don't, then you're not going to get great results. Sorry, if you don't know how software is built, what good code looks like, AND you don't "rub it the right way." Or as people say "prompt engineering."

I think for writing blog posts, getting info, it's easier. Though there's EXTREME dangers with it for other use cases. It can give incredibly dangerous medical advice. My wife is a psychiatrist and she's been keeping an eye on it, testing it, etc. To date AI has done more to harm people than it has help them in terms of mental health. It's also too inaccurate to use for mental health as well. So that field isn't adopting it so quickly. BUT they are trying and experimenting. It's just going to take some time and rightfully so. They don't want to rush start using something that hasn't been tested and validated. That's an understaffed field though, so I'm sure they will love any productivity gain and help they can get.

All said, I don't know what "slow" means for adoption. It feels like it's progressing quickly.

[fauigerzigerk]

I used to donate to Wikipedia, but it has been completely overrun by activists pushing their preferred narrative. I don't trust it any more.

I guess it had to happen at some point. If a site is used as ground truth by everyone while being open to contributions, it has to become a magnet and a battleground for groups trying to influence other people.

LLMs don't fix that of course. But at least they are not as much a single point of failure as a specific site can be.

[notarobot123]

> at least they are not as much a single point of failure

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?

[fauigerzigerk]

>Imagining that AI is somehow more neutral or resistant to influence is incredibly naive

The point is well taken. I just feel that at this point in time the reliance on Wikipedia as a source of objective truth is disproportionate and increasingly undeserved.

As I said, I don't think AI is a panacea at all. But the way in which LLMs can be influenced is different. It's more like bias in Google search. But I'm not naive enough to believe that this couldn't turn into a huge problem eventually.

[ramon156]

Can I ask for some examples? I'm not this active on Wikipedia, so I'm curious where a narrative is being spread

[kristjank]

Franklin Community Credit Union scandal is a good example, well outlined in this youtuber's (admittedly dramatized) video: https://www.youtube.com/watch?v=F0yIGG-taFI

[n4r9]

Is their argument documented anywhere in text, rather than an 8 minutes video?

[hbogert]

ask chatgpt to summarize

[n4r9]

Not convinced that would be any better, but I did just have a go and got the response:

> I'm currently unable to fetch data directly from YouTube due to a tool issue. This means I can't access the transcript, timestamps, or metadata from the video at this time.

[notarobot123]

Here are some examples from which you can extrapolate the more serious cases: https://en.wikipedia.org/wiki/Wikipedia:Lamest_edit_wars

[fauigerzigerk]

I thought about giving examples because I understand why people would ask for them, but I decided very deliberately not to give any. It would inevitably turn into a flame war about the politics/ethics of the specific examples and distract from the reasons why I no longer trust Wikipedia.

I understand that this is unsatisfactory, but the only way to "prove" that the motivations of the people contributing to Wikipedia have shifted would be to run a systematic study for which I have neither the time nor the skills nor indeed the motivation.

Perhaps I should say that am a politically centrist person whose main interests are outside of politics.

[junek]

Let me guess: you hold some crank views that aren't shared by the people who maintain Wikipedia, and you find that upsetting? That's not a conspiracy, it's just people not agreeing with you.

[fauigerzigerk]

Your guess is incorrect. I'm keeping well away from polarised politics as well as anti-scientific and anti-intellectual fringe views.

[Panzer04]

Single point of failure?

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

[fauigerzigerk]

Not in a technical sense. What I mean is that Wikipedia is very widely used as an authoritative source of objective truth. Manipulating this single source regarding some subject would have an outsize influence on what is considered to be true.