Theres alot of what about women arguments which is fair, but I think it's entirely fair to view this as a harm to men's mental health, and it should be the focus.
Right now, we have a societal issue of isolating, vulnerable men. Men have the highest suicide rate, are statistically more likely to commit violent offenses and have a high rate of domestic abuse. 'Incels' who are likely these companies target market have committed domestic terrorism. I think this is an issue we should 100% be looking at, imagine what happens when there is AI misalignment? They could become a risk to themselves and others, the last thing we need is a unreliable tool for someone in that situation to sooth the pains of social isolation.
> Men have the highest suicide rate, are statistically more likely to commit violent offenses and have a high rate of domestic abuse. 'Incels' who are likely these companies target market have committed domestic terrorism.
Do everyone a favor and stop equating 'incel' with things like terrorism.
Largely agree, given your definitions and clarifications, but I see some things are just co-related issues not directly a death of that programming approach. Where I see it is the gap between programmers and end users, scope of 'users' expanding to other programmers, and the increased complexity causes more abstract soft skill code delivery/management roles are entirely co-existing issues. Where they didn't cause the death directly, more a co-morbidity situation, didn't help, but it didn't cause the death. I'd say the primary cause is cost and complexity of operations, forcing the perspective shift from 'help at least one actual human being' to 'help at least <MINIMUM VIABLE MARKET SHARE> of users/developers'. I'd also as an aside argue frameworks and items directed at devs (that are well-designed), are still abstractly utilitarian, because, if they didn't exist a human would have to do the work of programming or doing the work manually so it would directly help at least 1 human.
Looks like they want to build up and support middle men to do the apps more than them, and act more like a platform or operating system position. Which makes sense giant corporations reporting 95% failed AI projects and the core success cases are specialist companies tuning the platform to a specific problem are successful. Then there are a ton of snake oil AI apps that are over promising under delivering hurting the image of AI's usefulness
This is probably purely a pivot in market strategy to profitability to increase token usage, increase consumer/public's trust more than farming ideas for internal projects.
I think that MIT study of 95% of internal AI projects failing has scared off a lot of corporations from risking time in it. I think they also see they are hitting a limit of profitable intelligence from their services. (with the growth in inelegance the past 6–8 months being more realistic, not the unbelievable like in the past few years)
I think everyone is starting to see this as a middle man problem to solve, look at ERP systems for instance when they popped up it had some growing pains as an industry. (or even early windows/microsoft 'developers, developers, developers' target audience)
I OpenAI see it will take a lot of third party devs to take what OpenAI has and run with it. So they want to build a good developer and start up network to make sure that there are a good, solid ecosystem of options corporations and people can use AI wise.
The MIT study as released also does not really provide any support for the 95% failure rate claim. Until we have more details, we really don't know where that number came from:
Yea from what I understand 'Chats' and AI coding are something they already have market domination/are a leader on and are a good/okay product. It's the other use cases they haven't delievered on in terms of other companies using them as a platform to deliver AI apps, which I would imagine would have been a huge vertical in their pitches to investors and internal plans.
These third-party apps get huge token usage with agenentic patterns. So losing out on them and being forced to make more internal products to tune to specific use cases is not something they want to biuld out or explore
AI coding is mid(okay) yes, my main point is people use it and it's a good line of business right now for them. They expected bigger break throughs like gpt-2 to 3 to 4, and that's not happening so they have to lean on the other aspects of the business more.
The fact it is mid is why they are really needing all the other lines of business to work. AKA selling tokens to AI apps the specialize in other mid products, and limit the snakeoil AI products that are littering the market ruining AI's image of being the new catch all solution.
I was a big user of IntelliSense and more heavily, IntelliJ, for most of my career. It truly seemed like magic back then. I recall telling a colleague who preferred Emacs that it felt like having an editor that could read your mind, and would joke that my tab key was getting worn out.
Then I discovered LLMs.
If you think IntelliSense is comparable to what LLMs can do, you really, really need to try giving an AI higher-level problems to solve. Throwaway example I gave in a similar thread a few weeks ago: https://news.ycombinator.com/item?id=44892576
I think a big part of simonw's shtick is trying to get people to give LLMs a proper try, and TBH that's what I end up doing a lot too, including right now! The problem is a "proper try" takes dedicated effort, because it's not obvious where the AI will excel or fail for your specific context, and people legitimately don't have enough time for that.
But once you figure it out, it feels like when you first discovered IntelliSense, except you already know IntelliSense, so it's like... IntelliSense raised to the power of IntelliSense.
The things is that languages that need intellisense that much are language that made it too easy to construct complex systems. For lisp and C, you can get autocompletion for free, and indexing to offer docs preview and signature can be done quite easily as well. There's also an incentive to keep things short and small.
Then you have Java and C# where you need a whole IDE if you're writing more than 10 lines. Because using anything brings the whole jungle with it.
Hmm, I think all languages, regardless of verbosity, could be better with IntelliSense. I mean, if the IDE can reliably predict what you intend to type based on the context, regardless of the complexity of the application involved, why not have it?
Seems like languages like Java and C# that encourage more complexity just aim to provide richer context to mine. Simple example, given an incomplete line like "TypeA foo = bar.", the IDE can very easily figure out you want "bar.getBlah(baz)" because getBlah has a return type of "TypeA" and "baz" is the only variable available in the scope. But to have all that context at that point requires a whole bunch of setup beforehand, like a fine-grained types supported by a rich type system and function signatures and so on, which incentivizes verbosity that usually scales with the complexity of the app.
So yes, that's a lot of verbosity, but also a lot of context. To your point, I feel like the philosophy of languages like Java and C# is deliberately based on providing enough context for sophisticated tooling like IntelliSense and IntelliJ.
Unfortunately, the languages came before such sophisticated tooling existed, and when good tools did exist they were expensive, and even with those tools now being widely and freely availble, many people still don't use them. (Plus, in retrospect, the language designs themselves genuinely turned out to be more complex than ideal in some aspects.)
So the current reputation of these languages encouraging undue complexity is probably due to their philosophies being grounded in sound reasoning but based on predictions that didn't quite pan out as expected.
The thing is we did have nice tooling before those languages came to be. If you look at Smalltalk, it has this type of context in an even more powerful way. You can browse the whole library in a few click and view its code. And it has a Playground element where you can try and design stuff. And everything was inspectable.
Same with Lisp. If you take emacs has an example, you have instant documentation on every functions.
Another example can be python where there’s an help system embedded into the language.
Java is basically unwritable without a full indexer and completion. But it has a lot of guardrails and its verbosity discourages deviation.
And today we have Swift and kotlin which is barely better. They do a lot of magic behind the scene to reduce verbosity, but you’re still reliant on the indexer which is now coupled with a compiler for the magic stuff.
Better languages insists on documentation, contextual help, shorter programs, no magic unless created by the programmer, and visibility (inspection with a debugger and traceability with the system source available, if possible).
I never used SmallTalk but from what I heard about it, I feel like Java/C# etc were a deliberate push towards that kind of environment via IDEs. I am not sure why SmallTalk didn't catch on, but it may have something to do with resistance from the C++ programmers that Guy Steele mentioned they had to drag towards Lisp via Java. It seems to me that the current crop of languages is the result of this forced evolution of a reluctant developer market from relatively barebones languages like C/C++ towards a SmallTalk-like future.
Same, my personal theory where it excels and overachieves is where there is already really fleshed out and oversaturated developer ecosystems (and experienced developer pool) that organizations have alot of legacy software built on it. I think it will gain momentum as we see more need for distributed LLM agents and tooling pick up. (Or when people need extreme cost savings on front facing apis/endpoints that run simple operations)
I really hope this stays up, despite the politics involvement to a degree. I think this is a situation that is a perfect example of how AI hallucinations/lack of accuracy could significantly impact our lives going forward. A very nuanced and serious topic with lots of back and forth being distilled down to headlines by any source, it is a terrifying reality. Especially if we aren't able to communicate how these tools work to the public. (if they even will care to learn it) At least when humans did this they knew at some level at least they skimmed the information on the person/topic.
I've had multiple people copy and paste AI conversations and results in GitHub issues, emails, etc., and there are I think a growing number of people who blindly trust the results of any of these models... including the 'results summary' posted at the top of Google search results.
Almost every summary I have read through contains at least one glaring mistake, but if it's something I know nothing about, I could see how easy it would be to just trust it, since 95% of it seems true/accurate.
Trust, but verify is all the more relevant today. Except I would discount the trust, even.
> I've had multiple people copy and paste AI conversations and results in GitHub issues, emails, etc.,
A growing number of Discords, open source projects, and other spaces where I participate now have explicit rules against copying and pasting ChatGPT content.
When there aren’t rules, many people are quick to discourage LLM copy and paste. “Please don’t do this”.
The LLM copy and paste wall of text that may or may not be accurate is extremely frustrating to everyone else. Some people think they’re being helpful by doing it, but it’s quickly becoming a social faux pas.
> When there aren’t rules, many people are quick to discourage LLM copy and paste. “Please don’t do this”.
This doesn't seem to be universal across all people. The techier crowd, the kind of people who may not immediately trust LLM content, will try to prevent its usage. You know, the type of people to run Discord servers or open-source projects.
But completely average people don't seem to care in the slightest. The kind of people who are completely disconnected from technology just type in whatever, pick the parts they like, and then parade the LLM output around: "Look at what the all-knowing truth machine gave me!"
The problem is that ChatGPT results are getting significantly better over time. GPT-5 with its search tool outputs genuinely useful results without any glaring errors for the majority of things I throw at it.
I'm still very careful not to share information I found using GPT-5 without verifying it myself, but as the quality of results go up the social stigma against sharing them is likely to fade.
Have had some conversations where the other person goes into chatgpt to answer a question while I’m in the process of explaining a solution, and then says “GPT says this, look…”
Use an agent to help you code or whatever all you want, I don’t care about that. At least listen when I’m trying to share some specific knowledge instead of fobbing me off with GPT.
If we’re both stumped, go nuts. But at least put some effort into the prompt to get a better response.
I don't have an issue quoting LLMs in and of itself, but the context and how you present it both matter.
"ChatGPT says X" seems roughly equivalent to "some random blog I found claims X". There's a difference between sharing something as a starting point for investigation and passing off unverified information (from any source) as your own well researched/substantiated work which you're willing to stake your professional reputation on standing by.
Of course, quoting an LLM is also pretty different from merely collaborating with an LLM on writing content that's substantially your own words or ideas, which no one should care about one way or another, at least in most contexts.
In the old days, when somebody asked a stupid question on a chat/forum that was just a search away, you would link them to "let me google it for you" (site seems down, but there is now a "let me google that for you"), where it'd take the search query in the URL and display an animation of typing the search in the box and clicking the "search" button.
Every time somebody pastes an LLM response at work, it feels exactly like that. As if I were too fucking stupid to look something up and the thought hadn't even occurred to me, when the whole fucking point of me talking to you is that I wanted a personal response and your opinion to begin with.
I think you messed up something with your time-travelling setup. We're in the timeline where GPT5 did not become the all powerful sentient AI that Ai boosters promised us. Which timeline are you from?
Because arguing with people who are wrong on the internet. It's no fun doing the same with an LLM because you're either actually wrong or it will assume you're right without putting up a fight
We all think ourselves as understanding the tradeoffs of this tech and that we know how to use it responsibly. And we here may be right. But the typical person wants to do the least amount of effort and thinking possible. Our society will evolve to reflect this, it won't be great, and it will affect all of us no matter how personally responsible some of us remain.
I consider myself pretty technically literate, and not the worst at programming (though certainly far from the very best). Even so I can spend plenty of time arguing with LLMs which will give me plausible looking but extremely broken answers to some of the programming problems.
In the programming domain I can at least run something and see it doesn't compile or work as I expect, but you can't verify that a written statement about someone/something is the correct interpretation without knowing the correct answer ahead of time. To muddy the waters further, things work just well enough on common knowledge that it's easy to believe it could be right about uncommon knowledge which you don't know how to verify. (Or else you wouldn't be asking it in the first place)
Even with code, "seeing" a block of code working isn't a guarantee there's not a subtle bug that will expose itself in a week, in a month, in a year under the right conditions.
I've pointed this out a lot and I often get replies along the lines of "people make mistakes too". While this is true, LLMs lack institutional memory leading to decisions. Even good reasoning models can't reliably tell you why they wrote some code they did when asked to review it. They can't even reliably run tests since they'll hardcode passing values for tests.
The same code out of an intern or junior programmer you can at least walk through their reasoning on a code review. Even better if they tend to learn and not make that same mistake again. LLMs will happily screw up randomly on every repeated prompt.
The hardest code you encounter is code written by someone else. You don't have the same mental model or memories as the original author. So you need to build all that context and then reason through the code. If an LLM is writing a lot of your code you're missing out on all the context you'd normally build writing it.
It's just mass stupidity really. Technology is just a lever for what already existed.
The same people blindly trusting AI nonsense are the same people who trusted nonsense from social media or talking heads on unreputable news channels.
Like, who actually reads the output of The Sun, etc? Those people do, have always done and will continue to do so. And they vote, yaaay democracy - if your voter base lives in a fantasy world of fake news and false science is democracy still sancrosact?
> I've had multiple people copy and paste AI conversations and results in GitHub issues, emails, etc., and there are I think a growing number of people who blindly trust the results of any of these models... including the 'results summary' posted at the top of Google search results.
I like the term "echoborg" for these people. I hope it catches on.
If you were considering purchasing a Biology text book, and spot read two chapters, what if you found the following?:
In the first chapter it claimed that most adult humans have 20 teeth.
In the second chapter you read that female humans have 22 chromosomes and male humans have 23.
You find these claims in the 24 pages you sample. Do you buy the book?
Companies are paying huge sums to AI companies with worse track records.
Would you put the book in your reference library if somebody gave it to you for free? Services like Google or DuckDuckGo put their AI-generated content at the top of search results with these inaccuracies.
[edit: replace paragraph that somehow got deleted, fix typo]
I mean... Yes? That looks correct to me°, but it's been a minute since I worked with Temporal, so I'd run it myself and examine the output before I cut and paste.
Or have I missed your point?
---
°Missing a TZ assertion, but I don't remember what happens by default. Zulu time? I'd hope so, but that reinforces my point.
I would also read the documentation. In the given example, for example, you don't know if the desired fixed format "YYYY-MM-DD" might depend on some locale setting and only works because you happen to have the correct one in that test console.
Part of my point is this: If you have to read through the docs to see if the answer can be trusted, why didn't you just read the docs to begin with instead of asking the AI?
It's just dawned on me that one possible reason is that you don't know which docs to read. I've recently been forced into learning some JavaScript. Given the original question I wouldn't have known where to start. Now the AI has given me a bunch of things I can look at to see if it's the right thing
If you didn't know enough to provide the original prompt, the AI sends you down the Date path instead of Temporal. But you could have used a 20 year-old search engine that would lead you to Javascript docs. You don't need error-riddled AI-generated code to find your way to the docs.
Are you saying there is more than one way to solve most problems in life? I don't believe it! Obviously, only what you personally like is the one true way. /s
Because you skip the step of finding out which docs to read. You get the specific functions and what parts of it you should check presented to you. Also, you only need to do this when it isn't already clear.
Whenever I use AI in social settings to fact check or do research or get advice, I always trust but verify, but also disclaim it so that people know to trust but verify.
I think this is a good habit to get people into, even in casual conversations. Even if someone didn't directly get their info from AI and got it online, the content could have still been generated by AI. Like you said, the trust part of trust but verify is quickly dwindling.
One of the arguments used to justify the mass-ingestion of copyrighted content to build these models is that the resulting model is a transformative work, and thus fair use.
If this is indeed true, it seems like Google et al must be liable for output like this according to their own argument, i.e. if the work is transformative, they can’t claim someone else is liable.
These companies can’t have their cake and eat it too. It’ll be interesting to see how this plays out.
I mean that's what they have been getting already - the average joe had to deal with draconian copyright all this time but not that it's inconvenient for big tech they get to hand-wave it away. The social contract has already been broken.
And companies have always been able to get away with relatively minor fines for things that get individuals locked up until they rot.
American attitudes to free speech mean that only the most dramatic, damaging libels can be held accountable, years after the effect. I think the only one I can think of that got justice was Alex Jones libelling Sandy Hook victims.
(no easy answers: UK libel law errs in the other direction)
This story will probably become big enough to drown out the fake video and the AI (which is presumably being fed top n search results) will automatically describe this fake video controversy instead...
I think this is a situation that is a perfect example of how AI hallucinations/lack of accuracy could significantly impact our lives going forward.
This has been a Google problem for decades.
I used to run a real estate forum. Someone once wrote a message along the lines of "Joe is a really great real estate agent, but Frank is a total scumbag. Stole all my money."
When people would Google Joe, my forum was the first result. And the snippet Google made from the content was "Joe... is a total scumbag. Stole all my money."
I found out about it when Joe lawyered up. That was a fun six months.
Sorry but I don’t see how what you mention is a problem. A search engine is just surfacing the content you have in your site. That is very different from it making stuff up.
>a perfect example of how AI hallucinations/lack of accuracy could significantly impact our lives going forward.
how about stop forming judgments of people based on their stance on Israel/Hamas, and stop hanging around people who do, and you'll be fine. if somebody misstates your opinion, it won't matter.
probably you'll have to drop bluesky and parts of HN (like this political discussion that you urge be left up) but that's necessary because all legitimate opinions about Israel/Hamas are very misinformed/cherry picked, and AI is just flipping a coin which is just as good as an illegitimate opinion.
(if anybody would like to convince me that they are well informed on these topics, i'm all ears, but doing it here is imho a bad idea so it's on you if you try)
This has very little to do with Israel/Hamas. It could be false information about a lewd act, a violent crime, a racist comment, an affair, gross incompetence, a medical condition, religious blasphemy, etc, etc, etc.
People make judgments about people based on second hand information. That is just how people work.
>because all legitimate opinions about Israel/Hamas are very misinformed/cherry picked
Sure, there is plenty of misinformation being thrown in multiple different directions, but if you think literally "all legitimate opinions" are "misinformed/cherry picked", then odds are you are just looking at the issue through your own misinformed frame of reference.
Whenever someone asks my opinion on something related to the conflict in the holy land, I ask them if they want the anti-Jewish answer, the anti-Muslim answer, the pro-Jewish answer, of the pro-Muslim answer.
And it took me decades of studying this to determine what to call the two sides.
I also like "the Holy land" as a less political name for that region.
It's out of fashion and perhaps identified with Christianity, and some people think I'm being tongue-in-cheek or gently trolling by using it. But IMO it's neutral and unambiguous: that's a part of the world that is sacred to all the major religions of the Western hemisphere, while not being tied to any particular set of boundaries.
> how about stop forming judgments of people based on their stance on Israel/Hamas
I really don’t need to do much more than compare ‘number of children killed’ between Israel and Palestine to see who is on the right side of history here. I’ll absolutely form judgements of people based on how they feel about that.
I used to think it was kinda pander-y, but then after participating in some of these communities it was just obnoxious when it wasn't stated, the cultural wedge between people. Where randomly there was drama from someone posting an unrelated yet offensive meme/joke, then it was a huge discussion on if it was ban worthy, if it was okay to joke about, or xyz. When really I just wanted to be nerdy with others.
I get that. You can put up the signs, but it doesn't need to be a regular, loud topic in the community. In fact, the signs should serve to prevent the need to discuss it in the community and make moderation cleaner and easier.
> It's just another programming language, not a political platform
Politics is baked into everything we do, like the lack of any political messaging is still a political message. With this approach, it weeds out those that don't align with the core community which is ideal for an organization that only thrives with volunteer involvement.
As someone that has transitioned, I went into it with this opinion but now on year 7 of hormones, It's more complicated. I honestly now align more with the Olympic rulings of time in the hormone ranges of what you're competing under. Where at 5 years I lost all my strength, my bone density has dropped to cis womans (dexa scans). For instance, now I work out 3 times a week, 50 pounds still feels like 'get my body totally involved' heavy, before transitioning not working out it was a non-issue kinda heavy. (transitioned at 23) So its more having the time and atrophy of those muscles gained during that time period, then all that is left is probably an advantage of having a higher rate of fast twitch muscle fibre which probably is with in variance ranges of normal genetic advantages. Which also isn't a nice message to deliver to collage athletes of you have to muscle detox for 3-5 years.
Right now, we have a societal issue of isolating, vulnerable men. Men have the highest suicide rate, are statistically more likely to commit violent offenses and have a high rate of domestic abuse. 'Incels' who are likely these companies target market have committed domestic terrorism. I think this is an issue we should 100% be looking at, imagine what happens when there is AI misalignment? They could become a risk to themselves and others, the last thing we need is a unreliable tool for someone in that situation to sooth the pains of social isolation.