This doesn't seem like a well-thought-out post. I mean, I probably could have told you that from the fact that it's either pro-AI or anti-AI (in this case the latter), but this is a particularly poorly thought out post. Others here have pointed out the water-wasting red herring, but there are several other pieces of nonsense here — the general environmental doom-mongering, and labeling recursive self-improvement as "mystical", for example.
Me, I pointed Claude Opus 5 at a C codebase I've been working on for years and it immediately found five serious bugs and told me how to fix them, as well as how to use Linux system calls I wasn't familiar with to solve some other problems. And I gave it a new language for PEGs I'd written up a couple of years ago, and it wrote me a working implementation in OCaml that afternoon, fixing several bugs in my example grammars along the way and resolving a conceptual problem I'd had a roadblock on. I asked Fable 5.1 to implement a minimal proof assistant, and it wrote a clone of J-Bob in Python, which I'm studying now. Although there is a certain firehose quality to this stuff, I sure don't intend to use AI to actively de-skill my brain.
That doesn't guarantee that AI will be a beneficial innovation overall, of course! As with most innovations, it probably depends on the balance between people being able to use the innovation to gain control over their own lives, and people being able to use it to gain control over others' lives. (Barring a FOOM scenario, of course, where our prediction ability is nonexistent.)
So, I don't think a pro-AI post can be well thought out, either. It's too early and chaotic to understand what's going to happen, and it may actually be uncertain. Pro and anti are both far too simplistic.
I am sure many smokers in the 1950s also saw the health benefits of tobacco, that they were very happy to experience the increased energy after a smoke and could get more done as a result. I‘m pretty sure none of them felt their lungs and throats deteriorating either.
Likewise, you may not feel like AI is de-skilling your brain. But it probably is. We don‘t have the research (yet) that AI is addictive and causes brain atrophy, but all the evidence is certainly pointing that way, the same way the evidence were pointing that way about tobacco in the 1950s. My only hope is that it doesn’t take us another 2 decades to prove the obvious.
One thing I've learned in life is to accept that I'm in many ways an outlier, which means that even if the research is done, it frequently doesn't apply to me anyway.
Like I wouldn't doubt that research could show that more time spent on youtube is inversely correlated with learning. But if you just use it to watch lectures or tutorials for some skill, obviously you will be an outlier.
Or when "research shows that dieting doesn't work," obviously the result there is "research shows that most people don't maintain a healthy diet," and it shouldn't dissuade you from doing so. Choosing to do so anyway is what makes you the outlier.
And so it will likely go with AI's effect on the brain. Most people will likely use it to avoid thinking, and the research will show it to be a disaster. A small group will use (or already is using) it to challenge them and expose them to new ideas. You can choose to be in the latter group if you'd like.
> One thing I've learned in life is to accept that I'm in many ways an outlier, which means that even if the research is done, it frequently doesn't apply to me anyway.
This is a complete logical fallacy. It is quite literally mathematically impossible for you to not be "typical" for at least some aspects of your life. If you weren't, you'd probably be dead, or immortal.
Sure, I expect that if you gave me the LD99 of some poison, I will die. But there are still many things (and my point is this is one of them) where all it takes to be an outlier is to decide to be one.
I am pretty sure that likewise in the 1950s many smokers did indeed decided to be the outlier who smokes occasionally, and for the right reasons, and using the “correct” method/brand of cigarettes. And that every one of them decided that they were not gonna get addicted to nicotine. I am also sure that every single one of them did indeed get addicted to nicotine, and many of them died from cancer.
The majority of smokers did get addicted to nicotine, but significant percentages did not. However, it is true that even many "non-daily" smokers died from cancer as a result of their smoking.
I know that research on tobacco is full of bad science and bad actors, and experts in different fields (such as social psychology) continued to publish bad science on the topic well into the 2000s. This includes your 2002 German paper (which I can’t read because of a paywall).
Your second source says nothing about addiction in the chapter you posted, however they do talk about addiction in chapter 6, and they claim that most people agree that smoking is addictive, but heavy smoking is more addictive.
Still, (chiming in), your claim that " I am also sure that every single one of them did indeed get addicted to nicotine, and many of them died from cancer" could use some evidence to back it up.
My understanding is that only about 35% of those who ever try nicotine become addicted.
If you are going to instead say that nearly everyone who smokes occasionally will become addicted, you need to back it up with evidence, because it's not apparent.
I am speaking general. I don‘t have historical sources either that people spoke like that in the 1950s, and that is a much more glaring omission of citation. If I am wrong about that my whole argument falls apart. But if say 65% of regular tobacco users who use some strategy or otherwise believe the addictive properties of nicotine won‘t apply to them for some reason, then my argument still stands.
I don‘t know the figure here, but if it is true that 35% of those who try tobacco get addicted, and given that we know nicotine is highly addictive, I would guess that most regular users do indeed get addicted (and so would most people).
I am specifically talking about this cognitive bias of believing that the mean behavior does not apply to you. Very few people believe this about cigarettes anymore because the evidence is simply overwhelming and public health experts along with anti-smoking activists were able to (despite incredible odds) shift the common rhetoric closer to what is true.
There is a similar phenomena going on with ads. Most people believe they are unaffected by ads, but in reality they absolutely are. And the advertising industry has vested interest in the general public believing this illusion (otherwise they risk being regulated).
We are seeing the exact same phenomena with AI usage as we did with smoking in the 1950s. All the evidence is pointing towards these products being bad for you and being addictive. Yet most users of AI believe they are not addicted, and that they have found a way to use AI in a way which is less harmful for them. This is bias, these users are wrong, so are you, and so is OP. This is addictive to you, and it is bad for you.
I know research is lacking, but eventually we will have the evidence to show the case as clearly as we do today with tobacco.
On the other hand there's the consensus bias where you believe that everyone else is like you, when in fact they are not (which is why I said this is something I've had to learn to accept).
Sure, for most distributions, most people will be statistically normal, by definition. On the other hand, I know that e.g. state standard test scores when I was a kid said I was 99th percentile academically (and there was some kind of IQ test for the gifted program? I remember I qualified in all 3 areas they tested me), so I have some evidence to hypothesize that I might be an outlier along other psychometrics too. And I'll go ahead and guess that many people here are also not the median in many ways.
I don't think that this makes me somehow immune to lung cancer from smoking, but I might suspect that when I read about others' use of AI and how it rots their minds, they're perhaps not using it to learn Lean (the process of learning has mostly been the AI spits out crap proofs that seem way more complicated than they ought to be, and then I spend some time trying to find ways to simplify it. We have a back-and-forth about different encodings, or ideas for syntax macros, etc. Then I can compare with e.g. Mathlib's solution) or checking their understanding as they work through some notes. I don't know what most people use it for or what they get addicted to, but I believe it might be different from what I do.
With regards to ads, I live somewhere where outdoor advertising is banned, I block ads on my computer, I don't watch TV or movies or "news," and I don't install whatever apps, so there aren't a whole lot of channels for ads to reach me. I suspect again that this is not the typical experience. That said I do view ads as fraud-adjacent and damaging to markets, so fully support strongly regulating or outright banning paid advertising.
This comes across as an athlete excusing their smoking behavior because their superior physical abilities makes them immune to the bad effects of smoking.
I assure you a 99th percentile in 100 m dash may have believed that smoking did not affect them the same way it does an average person, but they would be wrong. And so are you about your AI usage.
You keep assuming that LLM usage is automatically harmful with literally zero reason to believe that, and several anecdotes from people on how they use it positively.
There's simply no reason to even draw a weak analogy to smoking. You'd do much better to draw an analogy to e.g. methylphenidate or amphetamine, which certainly you can abuse and get addicted to, and might even be harmful for the average person (even carrying the possibility of psychosis!), but which carry cognitive benefits with minimal side effects for others.
I'll give another concrete use: the other day I scanned a storybook with my phone and had codex OCR it and compare with a list I've started of words and sound patterns my kid knows how to read so that it could identify irregular words or new sounds for us to go over before we start the story.
Please, concretely, let me know what the adverse effects were there.
I keep saying it because all the evidence (while still not conclusive) is pointing in that direction, just like they were with tobacco in the 1950s (It took another two decades for the evidence of tobacco to become conclusive; and two decades still for regulators to catch up).
We do know that LLMs can cause psychosis, we do know it can cause people to behave in an unhealthy manner, etc. We are seeing more and more evidence that AI usages in education is at least correlated with bad grades. We are seeing people exhibit addictive behavior around AI. There is plenty of reasons to suspect that as we gather more evidence that they will become conclusive about the addictiveness and the unhealthyness of these products. For me this is obvious.
Evidence doesn't point in that direction though, even if we establish correlations.
Like the education effect is pretty obvious: most people will use it to cheat on their homework and avoid learning. Some people will use it to learn new topics or with more depth, and give them additional targeted probing questions to answer (basically, give them more homework). Average achievement will probably drop. The high achievers will probably do even better.
Boiling it down to "it harms your brain" when people are already telling you how to use it to learn more effectively is just silly. Most people are happy to rot their brains. They'll take any opportunity to do it. If you're an outlier, you're probably aware of that fact on some level, and it doesn't make sense for you to make decisions looking at the median who behave completely differently. You have to look at the mechanisms behind the data and interpret under what situations they could plausibly apply to you.
Like I said, the parallels to the 1950s tobacco usage is glaring. People used to say the exact same about cigarettes, that the way they specifically use cigarettes, the way their body is build, the way they control the habit, etc. etc. all leads to them being unique in some way or another making the habit healthy actually.
The evidence were all pointing towards tobacco being both addictive and unhealthy, but it still took two decades for the evidence to become clear and public health experts to be proven right. Maybe in two decades I will be proven wrong and you will be proven right, but I doubt it. When the evidence are all pointing in one direction, it is usually because they are right.
> We do know that LLMs can cause psychosis, we do know it can cause people to behave in an unhealthy manner, etc. We are seeing more and more evidence that AI usages in education is at least correlated with bad grades. We are seeing people exhibit addictive behavior around AI.
Aren't most of these true of amphetamines/methylphenidate also, though? Except the "bad grades" part. Certainly Adderall is habit-forming for some people and has risks for abuse, just like it can contribute to mental breakdowns and psychosis.
Likewise, the psychosis-LLM link is incredibly rare.
It kinda looks like you're overstating the evidence. This is not a good comparison to 1950s tobacco -- by the 1950s, the scientific evidence for tobacco being unhealthy was already quite substantial. The problem was getting that information to the public in the midst of a disinformation campaign. Here, for contrast, the evidence that moderate LLM use in professional settings has overall negative impacts is still pretty light.
I only use LLMs rarely, and I don't have any kind of emotional ties to the issue. I just think your claims extend decently far beyond the evidence.
kragen · · focus · HN ↗
Me, I pointed Claude Opus 5 at a C codebase I've been working on for years and it immediately found five serious bugs and told me how to fix them, as well as how to use Linux system calls I wasn't familiar with to solve some other problems. And I gave it a new language for PEGs I'd written up a couple of years ago, and it wrote me a working implementation in OCaml that afternoon, fixing several bugs in my example grammars along the way and resolving a conceptual problem I'd had a roadblock on. I asked Fable 5.1 to implement a minimal proof assistant, and it wrote a clone of J-Bob in Python, which I'm studying now. Although there is a certain firehose quality to this stuff, I sure don't intend to use AI to actively de-skill my brain.
That doesn't guarantee that AI will be a beneficial innovation overall, of course! As with most innovations, it probably depends on the balance between people being able to use the innovation to gain control over their own lives, and people being able to use it to gain control over others' lives. (Barring a FOOM scenario, of course, where our prediction ability is nonexistent.)
So, I don't think a pro-AI post can be well thought out, either. It's too early and chaotic to understand what's going to happen, and it may actually be uncertain. Pro and anti are both far too simplistic.
runarberg · · focus · HN ↗
Likewise, you may not feel like AI is de-skilling your brain. But it probably is. We don‘t have the research (yet) that AI is addictive and causes brain atrophy, but all the evidence is certainly pointing that way, the same way the evidence were pointing that way about tobacco in the 1950s. My only hope is that it doesn’t take us another 2 decades to prove the obvious.
ndriscoll · · focus · HN ↗
Like I wouldn't doubt that research could show that more time spent on youtube is inversely correlated with learning. But if you just use it to watch lectures or tutorials for some skill, obviously you will be an outlier.
Or when "research shows that dieting doesn't work," obviously the result there is "research shows that most people don't maintain a healthy diet," and it shouldn't dissuade you from doing so. Choosing to do so anyway is what makes you the outlier.
And so it will likely go with AI's effect on the brain. Most people will likely use it to avoid thinking, and the research will show it to be a disaster. A small group will use (or already is using) it to challenge them and expose them to new ideas. You can choose to be in the latter group if you'd like.
voidhorse · · focus · HN ↗
This is a complete logical fallacy. It is quite literally mathematically impossible for you to not be "typical" for at least some aspects of your life. If you weren't, you'd probably be dead, or immortal.
ndriscoll · · focus · HN ↗
runarberg · · focus · HN ↗
kragen · · focus · HN ↗
<a href="https://journals.sagepub.com/doi/10.1177/002204260203200220" rel="nofollow">https://journals.sagepub.com/doi/10.1177/002204260203200220
<a href="https://www.tobaccoinaustralia.org.au/chapter-3-health-effects/3-36-health-effects-of-occasional-smoking" rel="nofollow">https://www.tobaccoinaustralia.org.au/chapter-3-health-effec...
The majority of smokers did get addicted to nicotine, but significant percentages did not. However, it is true that even many "non-daily" smokers died from cancer as a result of their smoking.
runarberg · · focus · HN ↗
Your second source says nothing about addiction in the chapter you posted, however they do talk about addiction in chapter 6, and they claim that most people agree that smoking is addictive, but heavy smoking is more addictive.
Windchaser · · focus · HN ↗
My understanding is that only about 35% of those who ever try nicotine become addicted.
If you are going to instead say that nearly everyone who smokes occasionally will become addicted, you need to back it up with evidence, because it's not apparent.
runarberg · · focus · HN ↗
I don‘t know the figure here, but if it is true that 35% of those who try tobacco get addicted, and given that we know nicotine is highly addictive, I would guess that most regular users do indeed get addicted (and so would most people).
ndriscoll · · focus · HN ↗
runarberg · · focus · HN ↗
There is a similar phenomena going on with ads. Most people believe they are unaffected by ads, but in reality they absolutely are. And the advertising industry has vested interest in the general public believing this illusion (otherwise they risk being regulated).
We are seeing the exact same phenomena with AI usage as we did with smoking in the 1950s. All the evidence is pointing towards these products being bad for you and being addictive. Yet most users of AI believe they are not addicted, and that they have found a way to use AI in a way which is less harmful for them. This is bias, these users are wrong, so are you, and so is OP. This is addictive to you, and it is bad for you.
I know research is lacking, but eventually we will have the evidence to show the case as clearly as we do today with tobacco.
ndriscoll · · focus · HN ↗
Sure, for most distributions, most people will be statistically normal, by definition. On the other hand, I know that e.g. state standard test scores when I was a kid said I was 99th percentile academically (and there was some kind of IQ test for the gifted program? I remember I qualified in all 3 areas they tested me), so I have some evidence to hypothesize that I might be an outlier along other psychometrics too. And I'll go ahead and guess that many people here are also not the median in many ways.
I don't think that this makes me somehow immune to lung cancer from smoking, but I might suspect that when I read about others' use of AI and how it rots their minds, they're perhaps not using it to learn Lean (the process of learning has mostly been the AI spits out crap proofs that seem way more complicated than they ought to be, and then I spend some time trying to find ways to simplify it. We have a back-and-forth about different encodings, or ideas for syntax macros, etc. Then I can compare with e.g. Mathlib's solution) or checking their understanding as they work through some notes. I don't know what most people use it for or what they get addicted to, but I believe it might be different from what I do.
With regards to ads, I live somewhere where outdoor advertising is banned, I block ads on my computer, I don't watch TV or movies or "news," and I don't install whatever apps, so there aren't a whole lot of channels for ads to reach me. I suspect again that this is not the typical experience. That said I do view ads as fraud-adjacent and damaging to markets, so fully support strongly regulating or outright banning paid advertising.
runarberg · · focus · HN ↗
I assure you a 99th percentile in 100 m dash may have believed that smoking did not affect them the same way it does an average person, but they would be wrong. And so are you about your AI usage.
ndriscoll · · focus · HN ↗
There's simply no reason to even draw a weak analogy to smoking. You'd do much better to draw an analogy to e.g. methylphenidate or amphetamine, which certainly you can abuse and get addicted to, and might even be harmful for the average person (even carrying the possibility of psychosis!), but which carry cognitive benefits with minimal side effects for others.
I'll give another concrete use: the other day I scanned a storybook with my phone and had codex OCR it and compare with a list I've started of words and sound patterns my kid knows how to read so that it could identify irregular words or new sounds for us to go over before we start the story.
Please, concretely, let me know what the adverse effects were there.
runarberg · · focus · HN ↗
We do know that LLMs can cause psychosis, we do know it can cause people to behave in an unhealthy manner, etc. We are seeing more and more evidence that AI usages in education is at least correlated with bad grades. We are seeing people exhibit addictive behavior around AI. There is plenty of reasons to suspect that as we gather more evidence that they will become conclusive about the addictiveness and the unhealthyness of these products. For me this is obvious.
ndriscoll · · focus · HN ↗
Like the education effect is pretty obvious: most people will use it to cheat on their homework and avoid learning. Some people will use it to learn new topics or with more depth, and give them additional targeted probing questions to answer (basically, give them more homework). Average achievement will probably drop. The high achievers will probably do even better.
Boiling it down to "it harms your brain" when people are already telling you how to use it to learn more effectively is just silly. Most people are happy to rot their brains. They'll take any opportunity to do it. If you're an outlier, you're probably aware of that fact on some level, and it doesn't make sense for you to make decisions looking at the median who behave completely differently. You have to look at the mechanisms behind the data and interpret under what situations they could plausibly apply to you.
runarberg · · focus · HN ↗
The evidence were all pointing towards tobacco being both addictive and unhealthy, but it still took two decades for the evidence to become clear and public health experts to be proven right. Maybe in two decades I will be proven wrong and you will be proven right, but I doubt it. When the evidence are all pointing in one direction, it is usually because they are right.
Windchaser · · focus · HN ↗
Aren't most of these true of amphetamines/methylphenidate also, though? Except the "bad grades" part. Certainly Adderall is habit-forming for some people and has risks for abuse, just like it can contribute to mental breakdowns and psychosis.
Likewise, the psychosis-LLM link is incredibly rare.
It kinda looks like you're overstating the evidence. This is not a good comparison to 1950s tobacco -- by the 1950s, the scientific evidence for tobacco being unhealthy was already quite substantial. The problem was getting that information to the public in the midst of a disinformation campaign. Here, for contrast, the evidence that moderate LLM use in professional settings has overall negative impacts is still pretty light.
I only use LLMs rarely, and I don't have any kind of emotional ties to the issue. I just think your claims extend decently far beyond the evidence.