HNHacker News
TopNewBestAskShowJobs

genrilz

159 karma · joined October 7, 2021

submissionscomments
genrilz··on LLMs know more than they show: On the intrinsic representation of hallucinations
I think what you are saying here is that because it has no "concept" (I'll assume that means internal model) of truth, then there is no possible way of improving the truthiness of an LLMs outputs.

However, we do know that LLMs posses viable internal models, as I linked to in the post you are responding to. The OP paper notes that the probes it uses find the strongest signal of truth, where truth is defined by whatever the correct answer on each benchmark is, on the middle layers of the model during the activation of these "exact answer" tokens. That is, we have something which statistically correlates with whether the LLM's output matches "benchmark truth" inside the LLM. Assuming that you are willing to grant that "concept" and "internal model" are pretty much the same, this sure sounds like a concept of "benchmark truth" at work. If you aren't willing to grant that, I have no idea of what you mean by concept.

If you mean to say that humans have some model of Objective Truth which is inherently superior, I'd argue that isn't really the case. Human philosophers have been arguing for centuries over how to define truth, and don't seem to have come to any conclusion on the matter. In practice, people have wildly diverging definitions of truth, which depend on things like how religious or skeptical they are, what the standards for truth are in their culture, and various specific quirks from their own personality and life experience.

This paper only measured "benchmark truth" because that is easy to measure, but it seems reasonable to assume that other models of truth exist within them. Given that LLMs are supposed to replicate the words that humans wrote, I suspect that their internal models of truth work out to be some agglomeration (plus some noise) of what various humans think of as truth.

genrilz··on LLMs know more than they show: On the intrinsic representation of hallucinations
I'm not sure why you think hallucinations can't be "fixed". If we define hallucinations as falsehoods introduced between the training data and LLM output, then it seems obvious that the hallucination rate could at least be reduced significantly. Are you defining hallucinations as falsehood introduced at any point in the process?

Alternatively, are you saying that they can never be entirely fixed because LLMs are an approximate method? I'm in agreement here, but I don't think the researchers are claiming that they solved hallucinations completely.

Do you think LLMs don't have an internal model of the world? Many people seem to think that, but it is possible to find an internal model of the world in small LLMs trained on specific tasks (See [0] for a nice write-up of someone doing that with an LLM trained on Othello moves). Presumably larger general LLMs have various models inside of them too, but those would be more difficult to locate. That being said, I haven't been keeping up with the literature on LLM interpretation, so someone might have managed it by now.

[0] https://thegradient.pub/othello

genrilz··on LLMs know more than they show: On the intrinsic representation of hallucinations
Overall, excellent points! I would like to add on to that though. RLHF actually does effectively have one LLM educating another. Specifically, human trainer's time is valuable, so they train an AI to express the same opinion about some response as a human trainer would, and then have that trainer AI train the LLM under consideration.

It's both interesting and sensible that we have this education in the training phase but not the usage phase. Currently we don't tend do any training once the usage phase is reached. This may be at least partially because over-training models for any special purpose task (including RLHF) seems to decrease performance.

I wonder how far you could get by learning from retraining from some checkpoint each time with some way to gradually increase the quality of the limited quantity training data being feed. The newer data could come from tasks the model completed, along with feedback on performance from a human or other software system.

Someone's probably already done this though. I'm just sitting in my armchair here!

genrilz··on Internal representations of LLMs encode information about truthfulness
I mean, sure. OP implied that "capital of Connecticut is Moscow" is the sort of thing that a human "model of truthfulness" would encode. I'm pointing out that the human model of that particular fact isn't inherently any more truthy than the LLM model.

I am saying that humans can have a "truther" way of knowing some facts through direct experience. However there are a lot of facts where we don't have that kind of truth, and aren't really on any better ground than an LLM.

genrilz··on Internal representations of LLMs encode information about truthfulness
We kinda do have LLMs in a swarm configuration though. Currently LLMs training data, which includes all of the non RAG facts they know, come from the swarm that is humans. As LLM outputs seep into the internet, older generations effectively start communicating with newer generations.

This last bit is not a great thing though, as LLMs don't have the direct experience needed to correct factual errors about the external world. Unfortunately we care about the external world, and want them to make accurate statements about it.

It would be possible for LLMs to see inconsistencies across or within sources, and try to resolve those. If perfect, then this would result in a self-consistent description of some world, it just wouldn't necessarily be ours.

genrilz··on LLMs know more than they show: On the intrinsic representation of hallucinations
There is some model of truthfulness encoded in our heads, and we don't draw all of that from our direct experience. For instance, I have never been to Connecticut or Moscow, but I still think that it is false that the capital of Connecticut is Moscow. LLMs of course don't have the benefit of direct experience, which would probably help them to at least some extent hallucination wise.

I think research about hallucination is actually pretty valuable though. Consider that humans make mistakes, and yet we employ a lot of them for various tasks. LLMs can't do physical labor, but an LLM with a low enough hallucination rate could probably take over many non social desk jobs.

Although in saying that, it seems like it also might need to be able to learn from the tasks it completes, and probably a couple of other things too to be useful. I still think the highish level of hallucination we have right now is a major reason why they haven't replaced a bunch of desk jobs though.

genrilz··on What happens when people with acute psychosis meet the voices in their heads?
I acknowledge that all lies are going to make things more difficult in the long run while making things easier in the short run. I would like to point out that sometimes things need to be easier in the short run though. In code terms this is something like technical debt.

Getting back to the example. If there is some combination of not being close enough, (in the example, lets say just started dating) and insecurities running too high, then they won't necessarily trust your reasoning. They might just feel like you are attacking them, or that you are trying to cover for not reacting fast enough. The end result is that they now feel ugly and like their date hates them. This won't make their body image issues any better.

While I've not dated anyone with that level of insecurity, I've been friends with a some, though none of them had body image issues specifically. In my experience they have to both trust you quite a bit, and also be ready to confront their own insecurities. I've tried to speak my mind clearly to one friend who evidently weren't ready to process their insecurities, and in the end it just seemed to make them feel bad and also hurt our friendship. They seemed to feel like I was attacking them and making excuses in a manner similar to what I talked about in the example. I've had some level of success with another friend, but they had actually expressed self-awareness and a desire to change. My current policy is to keep away from people's insecurities unless they express a desire to change or ask me to be brutally honest.

I could choose to only let people into my life that don't need me to route around their insecurities. However, I do want to be available to people who need emotional support, which means occasionally placing myself in a position where I need to be careful with people. My actual weapon of choice here is to re-frame the situation, because I'm pretty terrible at lying convincingly and also don't enjoy doing it. An example of a re-frame is the "you look beautiful" in my first post. However, re-frames are also a stop-gap measure as other people won't do such re-frames on their own. What I am doing is not technically deception, but I don't feel like the sort of deception that white lies are composed of has significantly worse side effects. They are both just a way to push the issue down the road. Furthermore, coming up with re-frames is a moderately difficult skill that I have specifically cultivated, and I think white lies are easier to do "untrained". Thus, if someone is put on the spot, I'd prefer if they didn't write off white lies just because they fit the definition of lie.

As far as your comments on doing myself dirty. Guilty as charged, I certainly have work to do on that front.

genrilz··on What happens when people with acute psychosis meet the voices in their heads?
I think I spelled it out in the post you are referencing, but I see a white lie as the equivalent of "not saying something because it is rude", except for when you are forced to say something and not doing so would be meaningful.

I'm pretty curious about what you think white lies do in a social dynamic though. I'd appreciate it if you could elaborate.

EDIT: My post is literally your GP, but based on your stance earlier in this thread, I think you were talking about aleph_minus_one's post I was responding to. In which case I think our opinions are similar?

genrilz··on What happens when people with acute psychosis meet the voices in their heads?
I agree that in professional settings, it's usually best for the efficiency of the company to be direct about everything. I also generally prefer that people provide accurate feedback about how I am doing in interpersonal relationships. I do think that "sparing" someone from an emotional reaction is sometimes a reasonable thing to do in an interpersonal relationship though, depending on the relationship.

Back to the dress example. Let's say the dress wearer is my girlfriend, and I have been asked the question "does this dress make me look fat?". Let's also assume I do think this particular dress makes her look fat, but generally find her attractive, and know that she has body image issues. If I answer "yes", or even "that dress isn't particularly flattering on you", she might interpret that as "He thinks the dress makes me look fat" -> "I am fat" -> "I am ugly", which is not actually what I think and not what I want her to think of herself. Even if I try to reassure her that I don't think she's ugly, she might think that my reassurance is a lie. People with insecurities don't always think logically about them. I think "You look beautiful" would probably be the most ideal response, (in that is both true and doesn't cause her to spiral) but if I didn't have time to think about my response, "No, it doesn't make you look fat" feels like a better response than something else which would cause her to spiral.

If I feel the need to actually help her choose the right dress, I can back that up by pointing out exactly how some other dress that looks better on her actually looks good.

Of course, if I feel like she could take it well, then pointing out exactly how the bad dress is bad might be helpful to her. However, I feel like keeping her from spiraling is more important then informing her of every detail of my taste in dresses.

Obviously, I made this scenario very concrete, but I feel like you can't really decide if it is a good idea without that very concrete knowledge. Which is one reason why white lies might be a bad idea in professional settings, where you don't necessarily know how people will respond to things in the first place. (Because you might not know them well)

genrilz··on What happens when people with acute psychosis meet the voices in their heads?
I believe you when you say that white lies violate your trust. I'm saying that that experience isn't universal though. Different people define violations of their trust differently. Thus my suggestion at the end to make sure that the people around you know that this is part of your definition. (Though you might have already done this)
genrilz··on What happens when people with acute psychosis meet the voices in their heads?
I think what GP is saying is that white lies are not necessarily betrayal. Betrayal is when someone violates your trust somehow. [0] You can see some examples of lying in the wiktionary definition I linked. (see the last two definitions) However, both these definitions involve the lie causing some negative consequence. A white lie is specifically a lie done to spare someones feelings. It's possible that the white lie can cause negative consequences, but that is not always the case.

To take a concrete example, consider the classic white lie "no, that dress doesn't make you look fat". It is possible that this could cause someone to wear an ugly dress, but the person probably looked at themselves in the mirror too, so they probably will end up choosing a dress that doesn't look awful. In this case there is no negative consequence associated with the white lie unless person wearing the dress is unable to accurately self assess. (which is something the speaker hopefully would know about the dress wearer)

You could define being lied to as a violation of your trust. In particular, I think autistic people often can't pick up on social cues, and so rely more on people speaking the truth to them. However, that doesn't mean white lies are betrayal for everyone. If you in particular feel like white lies are betrayal, you might want to tell people in your life that so they know it's important to you. They won't automatically know and it isn't automatically important to them.

[0]: https://en.wiktionary.org/wiki/betray#English

genrilz··on Microsoft and OpenAI's close partnership shows signs of fraying
I think we are speaking past each other here. I'm talking in consequentialist terms and you are (I think, feel free to correct me if I am wrong) talking in virtue ethics terms.

I'm assuming you aren't familiar with these terms and so am defining them. Forgive me if you already were familiar.

Consequentialists think that the purpose of morality is to prevent "bad" consequences from happening. From a consequentialist perspective, one can very much argue about what makes a consequence "bad", and it makes a lot of sense to do so if we are trying to improve the human condition. Furthermore, I think consequentialists tend to care more about making their systems consistent, mainly so they are fair. As a side effect though, no principles have to be sacrificed when making a concrete decision, since none of them conflict. (That's what it means for a system to be consistent)

Virtue ethicists think that the purpose of morality is to be a "good" person. I think you are correct that it's pretty hard to define what a "good" person is. There are also many different types of "good" people. Even if you had such a person with consistent principles, if you try and stuff everyones "good" principles into them, they would become inconsistent. It's hard for me to tell exactly what the point of being "good" is supposed to be if it is not connected to the consequences of one's actions, in which case one would just be a consequentialist, However, if the point was to improve the human condition, then I think it would take a lot of different types of "good" people, so it doesn't try and make sense to argue our way into choosing one of them.

This isn't really an argument for a position as much as me trying to figure out where we disagree. Does that all sound correct to you?

genrilz··on Microsoft and OpenAI's close partnership shows signs of fraying
Aren't the broad principles how we make the finer grained judgements in a fair and consistent manner though? Judges are at least supposed to use the outcomes of previous cases and the letter of the law to do this. The rest of us need something concrete too if we want to be consistent. I think that's worth arguing over.
genrilz··on Microsoft and OpenAI's close partnership shows signs of fraying
On the utilitarian front, I would prioritize my mother, but it would be because I care about her, not because I would be acting morally. I accept that I can't ever be a perfect utilitarian. Perfect is the enemy of good though.

I don't have kids, it does make a lot of sense that that would affect a person's psyche. The bit about having to define what behavior is good or bad seems to me like you are working out your beliefs through others, which seems like a reasonable way to do things since you get to have an outside perspective on the effects of what you are internalizing.

About debating morality though. That's exactly where principles become needed. It's great to say that we should be kind, but who are we kind to? It can't always be everyone at the same time. To bring things back to the trolley problem, I may save my mom, but it really is super unfair to the 20 people on the other track. This sort of thing is exactly why people consider nepotism to be wrong

genrilz··on Microsoft and OpenAI's close partnership shows signs of fraying
Sorry about the delay. I got sidetracked.

That... is actually a pretty interesting argument. I have to admit that if an objective morality existed floating in the Aether, there would be no way to logically prove or disprove that one's beliefs matched it.

Since I can't argue it logically, let me make an emotional appeal by explaining how my beliefs are tied to my life:

I chose to be a utilitarian when I was 12 or so, though I didn't know it had that name yet. The reason I chose this is that I wanted my beliefs to be consistent and kind. Utilitarianism has only one basic rule, so it can't really conflict with itself. Kindness wise, you can technically weigh others however you like, but I think most utilitarians just assume that all people have equal worth.

This choice means that I doubted that my emotions captured any truth about morality. Over the years, my emotions did further effect my beliefs. For instance, I tweaked the rules to avoid "Tyranny of the Majority" type things. However, my beliefs also changed my emotions. One fruit of this is that I started to mediate conflicts more often instead of choosing a side. Sometimes it does make more sense to choose a side, but often people will all behave well if you just hear them out. Another fruit of these beliefs is that rather than thinking of things in terms of "good" or "bad", I now tend to compare states of the world as being better or worse than each other. This means that no matter how little capacity I have, I can still get myself to make things a little better for others.

All this to say, I feel like deciding to doubt my own feelings very much did what young me wanted it to do. I wouldn't be able to grow as a person if I thought I was right in the beginning.

I'd be interested to hear how you came to your beliefs. Given how firmly you've argued in this thread, it sounds like you probably have a story behind your beliefs too.

genrilz··on Microsoft and OpenAI's close partnership shows signs of fraying
While I personally also consider my relationships to be more important than my earnings, I am still going to argue that it's subjective. Case in point, both you and I disagree with Altman about what success means. We are all human beings, and I don't see any objective way to argue one definition is better than another.

In case you are going to make an argument about how happiness or some related factor objectively determines success, let me head that off. Altman thinks that power rather than happiness determines success, and is also a human being. Why objectively is his opinion wrong and yours right? Both of your definitions just look like people's opinions to me.

genrilz··on Eye Contact Correction: Redirecting the eyes to look at the camera
EDIT: I got arguments mixed up. It's master-lincoln who was making the health arguments, not you. You were talking about not holding women to unreasonable beauty standards. I mostly agree, although how to actually stop doing that is very much an unresolved question. I'm leaving this comment here because I think it is interesting, I may explore how to break down beauty standards in a future comment.

First off, I know this is long. It is long because I am having to deconstruct the way we think about showering and make-up. I do think the deconstruction is interesting, but I wouldn't blame you for deciding not to read this.

Though ndndjdjdn's comment was phrased pretty snarkily, but I do think it is a good point. I'll try to explain why, addressing the first and last points, followed by the middle point:

I grew up on a farm. It can smell "bad", especially right after a rain storm. However, since I lived there my entire childhood, it mostly just smelled different rather than bad. I wasn't bothered by it like visitors were, though if I had to pick, I'd prefer the non-after-rainstorm way it smelled to the after-rainstorm smell.

If everybody stopped showering, then everybody would start smelling "bad". I'm sure people would adapt quickly enough, though they would still probably prefer the smell of people who regularly shower to some extent. Thus people tend towards the "showered" state to be more appealing to other people, even though people would probably get used to it if we all eliminated regular showering from our habits.

Similarly, I have seen both IRL and online where men think women who aren't wearing make-up are sickly or ugly, and that women who are wearing natural make-up aren't actually wearing make-up. If a woman stopped wearing make-up, then men would suddenly find that woman to be less appealing, though men would probably get somewhat used to it if all women stopped using make-up. Thus women are pressured to be in the "make-up" state, even though men would probably get used to it if all women eliminated make-up use from their habits.

I would like to note that the logic of the last two paragraphs is the same. Thus, my rebuttal to your point that the body would stink "uncomfortably" is that it probably wouldn't be uncomfortable if everyone stopped showering.

Now to address the middle point. I don't think there is a meaningful difference between showering and showering + wearing perfume (which I will call "perfumed" from hear on). People being in the "showered" state is considered normal, and from that point of normalcy, being in the "unshowered" state is bad and the "perfumed" state is good. However, if we are trying to figure out the best way for the world to be, I don't think what is currently "normal" should matter at all.

Now let's lay out the states people can be in:

Smell wise, people can be in the "unshowered", "showered", and "perfumed" states. As laid out in the comment you are replying to, "unshowered", or at least not daily showered, is the healthiest of these states.

Sight wise, women can be in the "no make-up" or "make-up" states. As you point out, health wise the "no make-up" state is the healthiest of these states.

Thus, if we are prioritizing health above cosmetic appeal, everybody should be in the "unshowered" and "no make-up" state. As I have argued earlier, everyone would probably get used to this eventually, but there would always be pressure for people to shower and wear make-up. Thus I think it is inconsistent to want people to be in the "showered" and "no make-up" state if you are arguing health is the reason. My personal take is that we should just let people make whatever choices they want based on their own values, and not mine or yours.

Thank you for coming to my TED talk. I hope you at least found that interesting. If you did read all the way through, I'd be interested to hear your response.

genrilz··on Eye Contact Correction: Redirecting the eyes to look at the camera
I feel like we might have been reading the original arguments in different ways, so let me summarize how I see this thread:

I interpreted ndndjdjdn's comment as sarcasm. (due to the use of the phrase "next up") That is, I think he was saying that if you take sadcherry's logic to its limit, then people wouldn't shower or would shower less. sadcherry's logic is that people shouldn't wear make-up because it cosmetic is not beneficial to health. Thus I think ndndjdjdn was talking about the fact that people use showers for cosmetic reasons, and believes sadcherry probably doesn't actually want people to shower less, and so should probably rethink his views about make-up.

You then posted your comment, saying that the health benefits of showers justify them even if they do have cosmetic benefits.

I then comment, saying that I shower in a way that is bad for my health because of cosmetic reasons. I wanted to imply that a lot of people shower like this, and therefor the fact that moderate showers might have some health benefits is irrelevant, because the way many/most people shower is actually unhealthy. I probably should have been more explicit about the fact that I thought many/most other people shower in unhealthy ways.

As an aside, I don't actually know of any concrete benefits to health besides making sure open wounds don't get infected. I tried to search the web for other benefits, and the only additional ones I got are exfoliation (which is cosmetic) and relaxation. (but relaxing things aren't generally classified as "healthy") With that in mind, I tend to believe the health benefits of showers are probably pretty over-hyped, (though not non-existent) and more like a cultural fiction to keep people showering than true knowledge.

I'd be interested to hear if you have a different take.

genrilz··on Eye Contact Correction: Redirecting the eyes to look at the camera
This is actually probably more fair of a comparison then you'd think. Daily showers are bad for health your health[0], but I absolutely do them because my body produces a lot of oil and odor. This is a cosmetic reason similar to make-up.

[0]: https://www.health.harvard.edu/blog/showering-daily-is-it-ne...

genrilz··on Understanding the Limitations of Mathematical Reasoning in LLMs
I feel like that is a pretty high bar to call "reasoning". I would like to think I am capable of reasoning, and yet I would not be able to write out by hand a binary file to be loaded on to the controller without using a compiler or looking at an assembly reference manual.

It seems like you want LLMs to be able to use tools (which some of them do. For instance, see the search engine chat bots, which can do searches) and make independent decisions (Search term here is "agent", I don't know how well they work, but I wouldn't personally let my computer do things like that unsupervised). However, I personally wouldn't consider these things to be a prerequisite to reasoning.

I would consider being able to solve a large range of problems that a human could solve with just pencil and paper to be reasoning. LLMs don't really seem to be as good as humans, but the certainly CAN solve these types of problems.

genrilz··on AAA Gaming on Asahi Linux
I think it is annoying to have to copy past the link in order to get around the essay too. I think its fine to say so. Its the people saying that HN isn't actually like how the essay represents HN to be that my original post and previous reply are talking about.

Given the comment thread you linked. I feel like photomatt has a very strong stomach. I'm glad I'm not him.

EDIT: I see where your comment is coming from, in that I did say that I thought it was fair to redirect to the essay. My opinion is that it was both fair to redirect to the essay, and fair for HN readers to find that redirect annoying. This all seems pretty civil to me, although some sibling comments do seem to be acting more hurt about it then I would expect.

genrilz··on AAA Gaming on Asahi Linux
The comment is not something someone would say to a person they intend to treat as a peer. I personally wouldn't post a comment like that unless I thought the person I was talking about was a danger to other people, regardless of how I felt.

Also, if I was Marcan or Lina^, then I would probably be the type of person who would like to read and participate in Hacker News, but would likely not because of that type of comment. (I suspect I could have found more toxic ones if I searched harder. This was in the first result) Since they have in effect been excluded from Hacker News, it feels fair to me that they put up an easily bypassable barrier to inform people visiting their site that they have been excluded, and what can be done to fix that.

^ I neither know nor care if the two are the same. I haven't reviewed the evidence, and it's bad internet educate to try and dig up someone's anonymous persona, so I won't try.

genrilz··on AAA Gaming on Asahi Linux
I do have some memories of not particularly trans positive comments in the comments section on earlier blogposts about their opengl/vulkan drivers, and also references to kiwi farms. I don't see such things commonly on most hacker news articles, but you'll usually see something like that anytime the article even incidentally has something to do with trans people.

Edit: I search "asahi lina site:news.ycombinator.com" in duckduckgo, and found "https://news.ycombinator.com/item?id=35237006" in the first result, for people who want proof.

genrilz··on Large Enough
For this model, it seems like the point is that it uses way less parameters than at least the large Llama model while having near identical performance. Given how large these models are getting, this is an important thing to do before making performance better again.
genrilz··on Unconditional Cash Study: first findings available
It seems like you are getting down-voted, but I think you are probably right.

One concern I have is that publicly announcing ways to break the law might result in the government cracking down on those particular ways. You could make a case for telling people how to break laws which don't really matter either way, and let them figure out how to break laws that are stupid and matter though.

Another, less pressing, concern I have is that if taken to its limit, no laws will matter anymore. I would hope human kindness would keep people from murdering and stealing from each other for the fun of it, but I'd expect some Tragedies of The Commons to occur. This could still be worth it though. I'm in a fortunate enough position to not really understand how limiting the stupid laws are, but I do understand that laws are hard to repeal.

genrilz··on Unconditional Cash Study: first findings available
I am certainly glad that you managed to escape. I don't think our society's response to the welfare cliff should be to tell people to break the law though. Surely we should redesign the law to lift everyone up instead instead?
genrilz··on Rust for Filesystems
Obviously, there are some cases where it makes sense to use an unsafe block. However, I think there might be fewer cases then people might think.

As an example, both the popular generic self-referencing crates ouroboros and self_cell have had memory safety bugs in them in the past. (links at the end) Both of them were carefully reviewed by experienced rust developers before their first public release, and yet they still ended up with such bugs. Admittedly, part of the issue is that both crates are trying to be more generic, so they have to be correct over a larger range of circumstances.

But still, these crates have one job, are both less than 1500 LOC, and they were carefully reviewed to ensure they did that one job before their public releases, and they still ended up having issues that were not caught. They might still have issues.

Thus, while it might be fine to use unsafe to state that your array of zeros is a valid utf-8 string without a runtime check, it's probably a good idea to twist yourself into a pretzel if the invariants are not trivial to prove and the overhead to maintenance/runtime isn't too high.

[0]: https://rustsec.org/advisories/RUSTSEC-2023-0042.html [1]: https://rustsec.org/advisories/RUSTSEC-2023-0070.html

genrilz··on Kids Who Get Smartphones Earlier Become Adults with Worse Mental Health (2023)
I would like to note that he has created/supported a variety of political causes in the past. His current cause does seem to be pointing to restricting kids internet access, but I can't really figure out what he wants.

One of these is his "moral foundations" theory, which claims that people have different underlying reasons for having various moral opinions. He claimed to have identified some but left the exact reasons open ended and changeable. He claimed that everyone realizing this would decrease political polarization. Neuroscientists pointed out that the "modules" he described can't map to actual brain structures, and he claimed that it didn't matter. I'm kind of unsure of if I would even call this science, as I don't really see how to make a predictable hypothesis out of it. However, it could be valuable as political speech if people do indeed lack awareness that other people value different things than them. (I always assumed everyone knew that, but maybe they don't?)

He also argued for "anti-fragility". The concept that kids these days are kept from experiencing failure and other unpleasant things, and that this will lead them to become worse off during adulthood. This seems more testable, but almost seems to be in direct contradiction to his current anti-smartphone/anti-social media cause. Perhaps this does come from a general "Kids these days are online too much" sort of sentiment, but anti-fragility by itself doesn't really directly suggest we should deny kids internet access.

I really don't know what to think of the man's political leanings, but I don't trust his scholarship much either way. Any time I try to fact-check him, I come up with rather inconclusive or small effects.

genrilz··on Kids Who Get Smartphones Earlier Become Adults with Worse Mental Health (2023)
I think I actually failed to respond properly in this comment. You were talking about smartphone use while the article was talking about social media use. I think Haidt and implicitly both of us were conflating the two, but they are separate topics.

I did a search for the effects of smartphones, trying to avoid meta-analysis describing only "overuse", and [0] came up. This seems to indicate somewhat larger effects of smartphones on anxiety. I can't access the article easily right now, but from the titles of other studies, it seems like sleep disruption might be a factor.

[0]: https://onlinelibrary.wiley.com/doi/abs/10.1002/smi.2805

genrilz··on Kids Who Get Smartphones Earlier Become Adults with Worse Mental Health (2023)
I've done literature reviews after seeing posts from Haidt come up here and elsewhere. If you go to semanticscholar.org for instance, type in "social media wellbeing meta-analysis", you will get something like [0]. (found a free link by searching the study title)

This and most other meta-analysis I have read on the subject seem to indicate that the effect of social media is pretty neutral. This one in particular does look at the effect on adolescents and depression specifically, and finds that it effects adolescents the same way it effects adults.

That being said, there is a small but statistically significant effect on anxiety and depression. This is cancelled out in overall wellbeing by other positive effects. However, given Haidt is saying we should implement age gates on social media, I would want the evidence to indicate something a bit more catastrophic and not cancelled out by positive effects.

So if you read the research, you will find there is good reason to be skeptical.

[0]: https://gwern.net/doc/sociology/technology/2022-hancock.pdf

← PreviousPage 2 of 4Next →