Why does AI feel so different?(blog.nilenso.com) |
Why does AI feel so different?(blog.nilenso.com) |
It's true that social benefit is incidental to the development of nearly all the big AIs, but that applies to every venture in capitalism.
Of course, the assumption is it is going to be a lot bigger as the decades give it time to fully roll out. It is impossible to guess precisely but it'll probably be bigger than the industrial revolution by the time all the 2nd order effects start settling in. We just don't know what it looks like when the entire management and political decision making apparatuses are suddenly stuffed with counterintelligence. It'll be a shock to the system compared to the current leadership.
Is that true?
Steam technology and factory machinery were created and used to make the pit, mill, and factory owners more money than with human workers. The railway mania in the 1840s was a period where vast sums of money were invested in railway lines and engine technology.
Back then and today, you find plenty of true believers, and plenty of people who are in it for the money.
Similar, when people financed eg canals and railways or factories in the past, many participants cared about the money.
(Or are you saying it's not about the money, but specifically VCs?)
Those things were tangible and their advantages/disadvantages more immediate than something intangible like ``AI''.
Now look what's happened to the Internet.
The problem, if you want to call it that, is purely social/cultural.
A friend of mine, who always found learning difficult, now works as a plumber and installer with his own one-person company.
One of his clients is the local community pool. He took over the job from an older man who showed him how everything worked. My friend noticed a lot of inefficient (manual) use of energy and resources, but the previous installer told him it would cost too much to automate, so nothing was done.
The other day we talked about AI, and he said he had never tried it. I asked him to think of some questions from his daily work. He came up with questions about the pool system and how to automate it.
ChatGPT answered his questions correctly, and after a few follow-ups, it even provided a list of the parts he would need and a plan to make it work. Last weekend he told me that, together with the previous installer, he made improvements that should save the pool thousands of euros over the next few years.
Of course, an experienced pool expert might have come up with the same solution, but it is often difficult to convince a committee to hire someone when nothing is broken, the budget is already set, and the inefficiencies are not obvious.
My friend now uses ChatGPT several times a day to solve practical problems. It helps him learn new things and explore possibilities he might not have considered otherwise.
I do not mind if ChatGPT makes the occasional mistake or gives an answer based on outdated information. As long as it is used as a tool to improve your own abilities, the subscription is worth the cost.
When I worked with colleagues, I sometimes relied on advice that turned out to be wrong. Everyone makes mistakes, even when working to the best of their abilities, and I find that my AI 'colleagues' make fewer of them.
https://replicationindex.com/2020/12/30/a-meta-scientific-pe...
This statement and similar ones along the line have a very bitter taste to me. Someone has to create the knowledge. Someone has to create the actual training data for others to be received in dumbed down and stripped out version through AI. You will always be late to the game when you try to gain knowledge that way. I really really don't understand why people overhype the learning part of AI so much. Yes, it's a neat tool, but in my opinion rather bad for __actual in depth__ learning
And we deeply need to believe something wonderous, whether delightful or terrifying, is also happening. I honestly think AI is largely mass hallucination which is supported by a system that is able to extract tremendous wealth from this hallucination continuing.
We can no longer believe that everything happening is normal, so we need to inject some, slightly less horrific, explanation for what we're experiencing.
0. https://www.wsj.com/economy/the-u-s-marches-toward-state-cap...
Literally all possible with Google and YouTube.
Literally all possible with school
Shouldn't you rather say "go to Wikipedia straight", or push for Wikipedia to improve their search with all their money already? Isn't a choice of using local open-weight LLMs not actually progress over "googling" and Google's rotten algorithms, or even freaking YouTube and their even worse ad feeds?
So what's your point (I'm tempted to insert "boomer" here)?
I thought the video images were AI at first but the guy actually does them in blender https://youtu.be/itj2t8Fv8xs?t=2040
It's something I often find when people complain about "even freaking YouTube and their even worse ad feeds" etc., when you look you find all this amazing stuff that didn't exist a few years ago. I mean I'm sure the ad crap exists but I don't really see it due to ad blockers and the like.
Proper research that compares many different pages still requires the regular Google search. Google LLM summaries are pathetic, Perplexity and Grok have to be nudged to the point that I feel I am training them with links obtained by Google.
Over time, we reduced friction and cost.
To take one example, the author speaks of using AI to chat with his favorite thinkers. Well I have favorite thinkers, so why don’t I chat with AI versions of them? Because I know it’s all fake. Whatever substance there is to chatting with a corpse or corpus is unknowable. So it amounts to playing with dolls.
I think for a living. I would LIKE AI to help. But it doesn’t. On balance, any benefit is drowned out by the effort it takes me to check its work.
I’m left wondering if the only people who value GenAI are those who haven’t experienced the satisfaction of thinking for themselves, or who have crippling impostor syndrome.
For the last 4bn or so years we've had reproducing DNA which led onto thinking humans but they get born, thinking and then drop dead a 14 millionth of a bn years later. Now we will have AI mind children that live on indefinitely. It's the biggest change since the evolution of life.
> Customer service automation, Agentic coding, Content creation
Fucking hell. Customer service automation. Auto-spaghetti code. Brainrot as a service. Truly what society needs.
(Edit: typos)
If customers want some customer service, but don't want to enough for it to be good; then very basic customer service is what they are getting.
Here is how I approach this - this might be a coping mechanism, but it’s certainly helped me personally.
LLMs are hugely impressive, no-one is denying that. But they have already been out for nearly 3 years at this point, and there are still massive gaps in their functionality that mean, in their current state, they are nowhere near being able to take over highly skilled work (e.g. software engineering at senior+ levels) from humans. They can handle grunt work well, but are unable to go beyond that: they operate at the level of a (poor) junior.
LLMs remind me to some extent of journalists and the Gell-Mann Amnesia effect. When I ask an LLM a question in an area I am not an expert in, I’m impressed. On the other hand, when I ask it something I am an expert in, I am almost always disappointed. I don’t bother using them for complex work any more, because they lie and hallucinate too much.
Furthermore, the rate of progress appears to be slowing.
All of this gives me some hope that my own life will not be completely ruined as a result of this technology.
And let's be real: we're in a bubble. There's zero evidence that LLMs have produced a measurable productivity boom. If anything, they're net negative in many contexts---encouraging shallow engagement, short-circuiting the learning process, and making knowledge workers more dependent on stochastic parrots that frequently hallucinate. We've replaced "look it up" with "ask the oracle," and the oracle can't reliably tell fact from fiction.
The hype cycle will keep burning cash until someone admits the emperor's wardrobe is ... speculative at best.
There's many models offered by many companies and labs that are close enough to state of the art. Many of them are completely open source or at least have open weights.
You might complain about outsourcing your thinking to a machine, sure. But there's no monopoly nor oligopoly.
I really hope open source models are the way, but the fact is the vast majority of day to day usage of LLMs is on models owned by multi-bullion dollar companies.
And even if they're not monopolies, people being influenced by companies to this extent should worry people.
> "Once men turned their thinking over to machines in the hope that this would set them free. But that only permitted other men with machines to enslave them."
- F Herbert, 1965
I want to emphasise the "other men with machines" part. Mr Herbert wasn't talking about an AI takeover. He was talking about a techbro takeover.
He didn't even believe in the agency of machines, just human potential - an inanimate thing that does not want, will not want to take over. It is a tool for the people who command it and do.
The problem is that people just trust what the LLM tells them, not realising they can be misled, or the model tells them "you're right" without any pushback or invitation to think further—just like an echo chamber, the consequences of which we've seen with social media in the last few years
Of course they're exciting from a tech perspective but I can't help but feel people are missing the bigger picture.
Lots of people, many amongst us here at HN, are critical of AI and LLMs in general. A lot of it has to do with two things, in my opinion - first, with the amount of money being thrown at it, and second, with the hype around it.
Many technical people, myself included, resent money being thrown at a problem rather than finding a proper solution. And we're especially skeptical of the hype cycle, because we've been through multiple hypes which didn't really live up to their claims.
But despite all that, I find regular people, using ChatGPT and other tools, just like any other online tool. Just like all of us rely on some software to find the time on the other side of the world when we have to schedule meetings, similarly, all these people are using these as tools to increase their capabilities and knowledge without the extra cruft that used to be involved in searching the web.
Additionally, AI is literally approachable by anybody who has an internet connection, and that means anyone with a smartphone. This didn't happen with devices, this didn't happen with Napster, this didn't happen with Bittorrent, and certainly not with Bitcoin. Each of these had some hurdle associated with it - cost of device, connectivity, technical knowhow.
But over the years, all those helped lay the foundation of what AI is being seen as now - as a tool that anybody can use. The more savvy among us are also using it to build their own tools as well.
So, yes, this time it certainly feels different.
Do you want that in every device or a lot of them, in every home, every service, for super cheap?
What's the paradigm being shifted here?
The effect right now of LLMs is reducing friction for some people with certain problems, but big important problems are already optimised to much higher levels.
You seem to believe the venture capital money machine that tells you technology and innovation must always be a major disruptive force and solve big, important problems.
Most problems we encounter in daily life are small. If a technology helps solve them faster, at a lower cost, or in a way that was not possible before, I believe it can scale very well. People just have to get accustomed to it.
For example, I fixed my lawnmower recently with the help of ChatGPT. I could have gone to the dealer or asked my brother for help, but instead it was a quick five minute fix that saved me from bothering anyone or spending time searching through manuals and videos for the answer. It literally couldn't have been fixed faster.
I prefer that over augmented reality goggles and other "Big Ideas".
As an extreme AI doomer, I can now respect and understand the position of AI optimists: they are both expressions of the same critique towards the status quo. What scares me then are those that feel the world is fine, we’re not collapsing, and AI is just another technology that is here to stay but won’t amount to much, and we’re all gonna be alright.
1: see HyperNormalisation by Adam Curtis
+∞. Hardly anyone wants to acknowledge that. Humans never really stopped wanting to believe in miracles and “AI” is the perfect idol.
A democratic politician only has to care about the next 4 years, at best; if any issue requires long-term planning, all you can do is hope it fixes itself.
Remind me of some of the things that Biden himself planned during his presidency.
We are living in an unprecedented golden age of peace and prosperity. What collapse are you talking about?
Yes, climate change is gonna knock a few percent points off global GDP, but it's not gonna be the end of the world. It's no worse than living in the UK rather than the US in these terms.
Keep in mind that this impact comes super imposed the our regularly scheduled growth.
We also also plenty of relatively low cost mitigation strategies. (Though many of them are not currently politically feasible.)
> [...] we're watching the United States transform into State Capitalism [...]
Hard to believe, but luckily there's more than one country on the planet.
Please vote with your feet!
Golden ages are fictional constructs, applied retroactively as part of an agenda. There is no such thing as living in a current golden age without knowing the future.
I also doubt the statements about peace and prosperity. These just sound like pithy phrases to dismiss concerns.
War in Ukraine, Israel-Palestine conflict, the recent India-Pakistan conflict, China rapidly ramping up its military and openly carrying out drills that are clearly designed to be practise runs for a Taiwan takeover, China asserting its presence near the Phillipines militarily, European countries ramping up defense spending, and so on, are clear indicators that the world is not going through a golden age of peace.
They'll win in the short term, due to the tyranny of convenience. Then, when all economic value has been extracted from them and the possibility of future value made nil, they'll lose. Of course, that doesn't mean you'll win though.
That's an uncharitable take of what he actually said, he was clearly talking about being able to get information presented as if it was explained by a particular person. The training data contains everything that person ever wrote, so it might be pretty close. Fake sure, but also often correct and presented in a familiar way.
I've often thought that many concepts I struggled with in school came down to the teacher explaining them in ways that was "incompatible" (for lack of a better word) with my way of thinking for that particular concept. Not that I was special, I think every student experiences this from time to time.
Being able to continue asking questions about dinosaurs or atoms or the roman empire long after any human teacher would get tired will help many a curious kid learn faster than they could before.
>I’m left wondering if the only people who value GenAI are those who haven’t experienced the satisfaction of thinking for themselves
I find this take strange - you still have to think but LLM's help you. Googling things or tracking down a breaking change in version 9.2 from 8.7 or why a regexp is not working correctly is not thinking. It's just busywork.
Spreadsheets removed the satisfaction of "carrying the one" with pen and paper and a whole bunch of valued skills lost value - but it turned out fine. Same with going from assembly to C or C to Java. We ended up with more programmers and more software, not less.
I think once the dust settles LLM's will be a similar change, if perhaps much larger in scope.
I’m not so sure on this one. AlphaFold2, while not quite an LLM, was based on Transformer architecture - and its implementation wasn’t a million miles away from a language model - and it massively improved the rate of protein structure prediction.
I think in general you’re correct, that we’re in a bubble, but I think it’s too extreme to say the technology is valueless.
But don't take my word for it, it's way outside of my area of research.
Historically (at least for the last century or so) we mostly stopped lowering working hours and instead focused on increased output.
Eg people still work eight hours a day, but get paid vastly more in real terms than in the 1950s. Instead of working only one hour a day (or fewer years or whatever) and taking a 1950s compensation package.
Of course, this is merely a statistical observation. There are plenty of people who eg decide to retire early on a modest nest egg.
It's also been a coherent academic discipline since at least Turing's 1950 paper introducing the Turing test and with a history going back well beyond that.
You are right that a general productivity boom is hard to detect so far.
But we already have certain industries with more than measurable productivity impact.
A very salient example: first level call centre agents. The kind of unfortunate souls who are paid to man a phone and not allowed to deviate from their script, nor authorised to actually solve your problem.
The companies renting out these services make heavy use of current AI and that has measurably affected hiring, staffing, prices etc.
Even fairly basic AI (like what we have today) is good for this use case.
Does it suck: sure. Does it suck more than the status quo with warm bodies without power: not really. And it's cheaper.
Is it useful ? Yes? Am I happy it exists yes ? Is it revolutionary. I doubt it.
I encourage you to keep up with the literature [0].
[0] https://metr.org/blog/2025-07-10-early-2025-ai-experienced-o...
That’s like saying. “So a car is just a better horse, what’s the paradigm being shifted here?”
It's definitely not wanting to talk to people, and even employment is a bit of a practical joke for now.
Comparing with Internet or electricity here is ridiculous thus far as well.
If we had actual intelligence instead of pretty nifty text robots, then it would be perhaps more true.
Therefore, current LLMs won't be a complete upheaval like some are fearing.
Things like AlphaFold might create upheavals in their respective fields but they are very specialised. I'm more enthusiastic about that than chatbots.
And people trying shading things doesn't make a monopoly. Especially if there's plenty of competition.
That's a weird self contradiction. I just contemporarily applied the Golden Age moniker, so your assertion is just plain wrong.
People have never been as rich as today, even poor people. Wars are a bit bursty, but if you smooth it out a bit, but the probability for an average human to become a casualty in a war has been going down over the centuries, and is still going down over the last decades.
I can't tell what your point is.
Btw, there were plenty of people who thought electricity or the internet were just toys or fads.
We don't even know if we want LLMs in more devices. Likewise we likely do not want Internet/email in every toaster.
LLMs staying the same would be the death of this technology. Even minor improvements don't matter enough.
Voice activated devices are not necessarily paradigm shifting either. We had these for over 25 years now.
See https://en.wikipedia.org/wiki/Electricity#Cultural_perceptio...
> It is said that in the 1850s, British politician William Ewart Gladstone asked the scientist Michael Faraday why electricity was valuable. Faraday answered, "One day sir, you may tax it."
That admittedly apocryphal anecdote is set 30 years before the turbine you mentioned was invented.
I tried Windsurf over the last 3 days. It can write complex game mechanics logic in multiple files and refactor code when told to. I can talk to it using language of the domain I just made up and I can talk to it using language that developers use when they talk about the code. It can solve bugs that would take me half an hour to figure out at least. It went from a poor junior to clever junior and half a senior in one year. It's not great at one shot architecting yet, but when told how to architect a thing it can move stuff around to conform with the desired architecture. It creates and modifies SVG's, understands colors and directions. It does all of that blind, without running the code. All it has is compiler (language server really) and linter.
You mention it's an autocomplete. I barely used auto-complete at all. I just told it what to do. I touched the code a little bit only because I wanted to, never because I had to. When I wanted a change to the code, 95% of the time I just told it what change I needed and it made it.
It basically replaces about 1/3 of software team already enabling (and pushing me towards) a role of (tech lead/architect/qa/product owner) and this fraction is rising.
It's not flawless. Sometimes it makes things that don't work on the first try. But you don't always need to revert. It's capable of fixing what it made when told about how the desired behavior differs. And all that without running the program.
I anticipate that further development is going to be letting it run and observe what it built so it's gonna move towards QA role. And the other thing might be slowly learning to differentiate between well architected code and badly architected code but that's probably harder as it requires careful preparation (and creation) of training data. While moving towards QA just requires figuring out how to let it run and inspect stuff. Maybe use the debugger.
I'm doing everything you're doing and probably more with Claude code (I 'let it run'). I'd say I too have been through a few of the different phases of the gartner hype cycle now.
As I said it's a good tool, it's far from being an "auto-pilot".
Both because in the US new construction is basically outlawed. But also because 1950s style houses and goods in general are so shoddy, you can no longer legally produce or buy them. In many cases, no one would even take them off your for free, even if they were still legal.)
On a global scale, the times between perhaps 1940s to 1970s had some of the harshest inequality ever. It's been since about the 1980s that inequality has gone down markedly.
Remember how India and China used to be on the verge of famine (or outright in famine). Nowadays obesity is the bigger problem.
I guess if anyone went into a cave and came out for 5 years things would be like science fiction. FPV drones on the battlefield would be another example...
As I said earlier, not downplaying it, really just warning you that if you lean to heavy on it without guiding it quite precisely, you will get burned, properly.
All good, we have different opinions.
Computers being able to keep the conversation going about complex topic like development of a computer program were 5 years ago farther away than fusion energy. Perpetually 30 to 100 years away.
Sure, for the best results you need to guide it because without it you'll get some program but not a good program. If you don't help it to control the complexity of the code with sensible architecture it will paint itself into a corner. But a junior developer would also do that if you just piled up requirements and never steered his output.
I agree with you that it's a tool. But more like a combine harvester, that you still need to drive, for now, than a hammer. Already more like a CNC machine than a power drill. And still getting better.
Everyone can have their own opinion on this. But I think next 5 years will change those opinions.
For me it is at least as magical as cell phones. When I was a kid the peek of wireless personal communication was walkie-talkie with a very limited range. The idea that you could cram enough electronics, RF stuff and a battery in a small enough package and route the calls globally with vast networks of two way communication stations seemed absolutely ludicrous. I haven't even seen SMD element before I had my first cellphone and I picked apart a lot of electronics. Then it happened. It was the kind of technology that got me completely surprised, like LLMs. I remember not being that shocked because I was young and was expecting new things. I was reading a lot of SF and there instant wireless communication was everywhere. But after 4 decades of basically no progress in the machines that do anything that resembles thinking this innovation caught me off guard completely. I would be less surprised if we already had fusion reactors providing most of our electricity.
On a more practical note, do you switch a lot the models that you use? I can't really tell them apart because I haven't worked with them all that much but I heard good things about GPT5 and when I used it it was very good (but others, like Claude or swt-1 were great too).
With regards to the models I use. I'm pretty much using Claude Sonnet 4 with Claude Code and Gemini 2.5. In my opinion Sonnet 4 is the most superior model I've used yet, although I've not used GPT-5. Where some of my attitude comes from is even in the way you get used to using differnt models, like you really kind of get a grasp for how to prompt them and kind of know what to expect. That's where I don't see them so much as an intelligence, but an approximation engine. I feel that what many "coders" fail to realize is how extremely common the patterns they're employing to solve a problem are. The approximation engine can see that as numbers, match it and give you a similar eyt slightly modified (to your context) version of what you want.
These models have been trained on probably hundreds of millions or billions of lines of code after all, yet there probably isn't all that much variation in what most code looks like. What you'll notice is that it's horrible at coding in anything it's not trained on. Recently I tried to use it with FastHTML, and it knows almost nothing about it, and the author of that project will even provide you with a context file you can give your "AI" to use so it can "understand" how to code with it.
We also follow like common architectural patterns to get them most out of "AI". We've found that to be very helpful to reduce the amount of random and unhelpful code we get out of them.