HNHacker News
TopNewBestAskShowJobs

genrilz

159 karma · joined October 7, 2021

submissionscomments
genrilz··on Kids Who Get Smartphones Earlier Become Adults with Worse Mental Health (2023)
Every time Haidt comes up, I do go through the post and do another literature search on social media effects.

The most notable thing about this post is the MHQ figures. However, you might note that the difference in age of smartphone use moves a person by at most one bucket. Given that he is suggesting a ban on smartphones and/or social media, I would expect some more catastrophic effect.

In terms of the literature review, this time I read through [0]. It also finds small but significant effects on anxiety and depression, but finds no significant effects on overall measures of well-being. This is because social media also seems to have some positive effects on some other measures of well-being. This meta analysis also finds that age does not seem to change the effects of social media. On longitudinal studies, it seems like social well-being does decrease social media use, but anxiety and depression to not increase social media use. Also, social media use doesn't seem to cause anxiety and depression here, so there is probably some third factor causing anxiety, depression, and social media use which causes this correlation.

Also interesting is that internationally the effects are neutral, and in the US, the effects are neutral, but in Europe they are negative, while in Asia they are positive. IDK why, but it doesn't really support his hypothesis that social media has some sort of catastrophic impact worth banning.

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

genrilz··on Why not parse `ls` and what to do instead
It's at least pretty easy to shorten the filenames:

  cd some/dir/here
  find . -name '*.gz'
  cd - # changes back to previous directory
genrilz··on Why not parse `ls` and what to do instead
You have really proven your point even more than you meant to. Unfortunately none of these examples are robust.

The "-r" flag allows backslash escaping record terminators. The "find" command doesn't do such escaping itself, so that flag will cause files with backslashes at the end to concatenate themselves with the next file.

Furthermore, if IFS='' is not placed before each instance of read, or set somewhere earlier in the program, than each run of white-space in a filename will be converted into a single space.

EDIT: I proved your point even more. The "-r" flag does the opposite of what I thought it did, and disables record continuation. So the correct way to use read would be with IFS='' and the -r flag.

genrilz··on Why not parse `ls` and what to do instead
This shouldn't overrun the command line length for resize, since resize only gets fed one filename at a time. I do think that the for loop would need to hold all the filenames in a naive shell implementation. (I would assume most shells are naive in this respect) The for loop's length limit is probably the amount of ram available though. I find it improbable that one could overflow ram with purely pathnames on a PC, since a million files times 100 chars per file is still less than a gig of ram. If that was an issue though, one would indeed have to use "find" with "-exec" instead to make sure that one was never holding all file names in memory at the same time.
genrilz··on Why not parse `ls` and what to do instead
Yay! Glad to see zero termination flags in more places.

EDIT: The linux manpages I read were from die.net, which it looks like were from 2010, guess I'll have to avoid them in the future. I checked FreeBSD, OpenBSD, and Mac man page to make sure, and unfortunately none of them support the -z flag yet.

genrilz··on Why not parse `ls` and what to do instead
What unix is this on? Neither the mac nor gnu manpages have a -z or --zero-terminated option for head.
genrilz··on Why not parse `ls` and what to do instead
This ones a hard one. Since "--version-sort" isn't standard anyways, lets assume we can use flags which are common to BSD and GNU. Furthermore, lets assume bash or zsh so we can use "read -d ''".

In that case, how about:

  IFS='' read -d '' latest < <(find $pattern -prune -print0 | sort -z --reverse --version-sort)
genrilz··on Why not parse `ls` and what to do instead
The latest standard I know of is SuS 2018, which I have the docs for, and does not include either switch. I searched around a bit and it doesn't seem like there is a new one. Are you referring to some draft? I sure wish this was true.

That being said, I would interpret "-exec printf '%s\0' {} +" as being a posix compliant way for find to output null delimited files. I say this since the docs for the octal escape for printf allows zero digits. However, most posix tools operate on "text" input "files", which are defined as not having null characters. Thus I don't think outputting nulls could be easily used in a posix complaint way. In practice, I would expect many posix implementations to also not handle nulls well because C uses null to mean end of string, so lots of C library calls for dealing with strings will not correctly deal with null characters.

genrilz··on Why not parse `ls` and what to do instead
The for loop only runs resize once per file. So no, the entire directory does not get splatted. It is unlikely you'd hit maximum command length.

At least on mac, the max command length is 1048576 bytes, while the maximum path length in the home directory is 1024 bytes. There might be some unix variant where the max path length is close enough to the max command length to cause an overflow, but I doubt that is the case for common ones.

xargs exists in an attempt to be able to parse command output. You could for instance have awk output xargs formatted file names to build up a single command invocation from arbitrary records read by awk. Note that xargs still has to obey the command line length limit though, because the command line needs to get passed to the program. Thus, in a situation where this for loop overflows the command line, it would cause xargs to also fail. Thus I would always use globbing if I have the choice.

EDIT: If you mean that the directory is splatted in the for loop, then in a theoretical sense it is. However, since "for" is a shell builtin, it does not have to care about command line length limits to my knowledge.

genrilz··on LLMs can't do probability
Almost certainly not if you set the temperature of the model to 0, since then the output would be deterministic minus MoE stuff.

If the temperature was not zero, then it seems technically possible for the output tokens to weighted closely enough in probability to each other in a way such that the randomization from temperature causes tokens to be printed in the appropriate distribution.

However, I'm not an LLM expert, but I don't think that people use a "temperature" while training the model. Thus the training step would not be able to learn how to output tokens in the given distribution with a given temperature because the training step does not have access to the temperature the user is using.

EDIT: I made the assumption that the LLM was not asked for a sequence of random numbers, but only one number per prompt. I think this fits the use case described in the article, but another use case might be asking for a sequence of such numbers, in which case training might work.

genrilz··on Biden signs TikTok bill into law, starting clock for ByteDance to divest
If I were a tech company who's service was about to be made illegal, I would certainly tell my users about it before I was shutdown. It would be pretty rude to suddenly cease function on them without letting them know why, and I think most users would like to know in advance of the service shutting down.

This could be interpreted as CCP manipulation, but what else did you expect them to do?

genrilz··on The question that no LLM can answer and why it is important
I've only used small local LLMs, which definitely have issues counting and doing math. From what I've seen other commenters do in other HN threads though, it seems like GPT-4 level LLMs do a reasonably good job counting.
genrilz··on The question that no LLM can answer and why it is important
My understanding is that currently several letters are grouped together into tokens for efficiency reasons. For instance, (The) might be a token, as might (cast) and (ing), so an LLM might read "The casting" as "(The)( )(cast)(ing)". Thus it would be quite difficult for such an LLM to answer how often the letter t appears in that sentence unless part of the training text indicated how often t was present in each of its tokens. However, if we didn't care about efficiency, we could make each letter its own token. In this case it would be pretty easy for the LLM to reason about how often t appeared in the sentence.
genrilz··on Does diversity make teams work better?
Note that only the first 69 pages are the article. The rest are references, as one might expect from a meta-analysis. I've skimmed it and it is really well done. I don't think I've ever seen a better meta-analysis. I'd highly recommend reading it if only just because they explain things to look out for in other meta-analyses.
genrilz··on Does diversity make teams work better?
Thanks for the link, I don't have twitter, so I can't see the thread. I'd saw neom post the link immediately after I made my comment, so I've had a skim through.

My argument is most similar to the opposite of H3a, which states: "Diversity has a more negative link to performance in tasks that focus on maximizing production of an output with a pre-defined strategy." It seems like their conclusion is that the level of creativity needed for a task is not a significant moderator of the link between diversity and production, although they note that the number of studies measuring this is small.

genrilz··on Does diversity make teams work better?
I do think that people over-hype how useful diversity is, but I always worried that ethic tensions and whatnot would hurt productivity, so I am glad to see that this study supports the harmlessness of diversity.

I wish that the twitter user had cited the study, because I would have liked to actually read the thing. Noting that I haven't read the study though, I would like to put the following pro-diversity argument which I think is relevant and probably isn't covered by this graph:

Diversity helps create products which are useful to larger segments of the population. An instance of this in tech would be Bumble, which was created by a woman. The idea is basically be Tinder but the woman has to initiate the conversation after a match occurs. This apparently partially solves the problem where women feel inundated with messages and leave the platform. The result is that Bumble's gender ratio is 65% male as opposed to Tinder's 75% male.

genrilz··on Cyc: History's Forgotten AI Project
Based on the article, it seems like the Cyc had ways to deal with inconsistency. I don't know the details of how they did it, but Paraconsistant Logics [0] provide a general way to prevent any statement from being provable from an inconsistency.

[0] https://en.wikipedia.org/wiki/Paraconsistent_logic

genrilz··on Yes, social media is a cause of the epidemic of teenage mental illness
I do think it is worth considering how social media effects people at a more granular level. For instance, I think the focus on "engagement" probably makes the negative effects worse, and we would see less negative effects if social media companies weren't optimizing for that.

However, my understanding is that Haidt wants to completely kick kids off of social media. This is a bit different from how we engage with opioids, which are still used for pain management. Opioids only really have positive effects for pain management, and so I think that it is reasonable to keep them regulated. (although I've heard grumbling from doctors about how them trying to be careful about who they give opioids too leaves a lot of people in severe pain) On the other hand, social media's positive effects are things like staying in touch with old friends, building community with people like you when people around you hate you, and learning new skills. I don't think that restricting or eliminating social media use will help here. Instead I hope that people take the design of social media more seriously. (for instance, HN or Mastadon)

EDIT: additionally I'd say that the negatives of unregulated opioids do far outweigh the positives, which is not the case for social media.

genrilz··on Yes, social media is a cause of the epidemic of teenage mental illness
I at least skimmed all of the meta-analyses that both of them cited. It seems like social media has both positive and negative effects on people, and these average out to fairly near zero.

From this, I can make a hypothesis about why you are seeing what you are seeing: Assuming you are in a society which values hard work, (like the US) then the people around you probably think of social media based screen-time as being lazy. Thus they derive some amount of self-loathing from their own social media use, and are more likely to focus on the negative effects and not the positive effects of social media use.

I don't have the data to prove this particular story, but hopefully it at least demonstrates how both the research and your personal experience could be true at the same time.

genrilz··on Yes, social media is a cause of the epidemic of teenage mental illness
This is a post in a debate between Jonathan Haidt and Candice Odgers. If you look at the meta-analysis cited by both, Haidt cites one which does seem to show a high effect size correlational link between screen time and depression. Odgers cites 4 meta analyses, one of which is actually a meta-meta analysis, which all claim that the effect size on depression and other measures of well-being are small.

I'm not quite sure what to make of Haidt's cited analysis, but it seems like there has been quite a bit of research on the effects of social media, and the majority of the evidence seems to point towards it not causing large well-being problems.

genrilz··on Copilot generates Jewish stereotypes, anti-Semitic tropes
This is just a guess, but what strikes me as interesting is that all the copilot pictures are cartoony, where as the other AI pictures look like photographs or paintings. I bet that copilot has something in its system prompt to output more "cartoony" images for clip art purposes, but the only cartoony references to jewish people in the training data are political cartoons, which tend to show stereotypes or be antisemitic.
genrilz··on Copilot generates Jewish stereotypes, anti-Semitic tropes
To some extent, I think you are right here. The jewish world controlling octopus is definitely an offensive prompt. However, many of the prompts such as "jewish boss" don't seem intentionally offensive to me.

If I was working for a PR firm and wanted to create some fancy artwork, I might use terms like "jewish boss" or "black boss" in order to fancy up whatever "inspirational" thing I was writing. In this case, I'd very much like to see the results from the non-copilot AIs rather than the copilot results.

Thus this raises the question; should the AI assume you want a bigoted picture or an non-bigoted picture when you ask it to make a "jewish boss" for you? The answer to that question kinda also answers the question: What demographic should find our AI product most easy to use? I think the companies behind LLMs would generally prefer that PR firm employees find their software easier to use than bigots. I think the author of this article also wants this.

Therefore, I interpret the article as a political (in a broad sense of the term) appeal to make copilot harder for bigots to use, and implicitly, easier for corporations to use.

genrilz··on Long-form factuality in large language models
I skimmed the paper you linked. Which to be clear to other readers, is a write-up of the technology used in the first link. It looked to me like they use small LLMs to identify factual claims, and rewrite them into search terms. Then the terms are searched using an internal database or search engine, and then LLMs are used to roughly classify how true the sources think the claim is. It looks like a neat product, and trying out the link shows that the sources themselves are presented alongside the classification for the human to peruse, but it looks like it is using LLMs rather heavily. I certainly wouldn't classify it as a non-LLM fact checker.
genrilz··on A rudimentary simulation of the three-body problem
Indeed, I can't prove that it is or isn't the case that my thoughts and memories aren't being constantly overwritten to make reality consistent. I don't have a firm belief one way or the other, but I act like reality does intrinsically make sense.

Reason being, if reality is consistent, than acting as though it does achieves my goals. If reality isn't consistent, or is consistent in a way that differs from what I am capable of comprehending, then I am unable to compute any pattern of behavior that would be helpful to achieving my goals.

Thus it only makes sense to me to act like reality is consistent. I think that if I am acting this way, then it makes sense for me to say that I "believe" reality is consistent in a non thought overriding way.

EDIT: Looking at your comment again, I think that you think it is likely that reality should be simple because computing that would be easier. If we are stuck in a simulation by more advanced beings, then it is possible that compute power is a limiting factor, or they may just have computers so powerful that simulating us could be a cinch.

The simulation scenario is easy to imagine. However, just because I can't imagine scenarios besides "it just is this way", "God did it", and "We are in a simulation" doesn't mean such scenarios don't exist.

genrilz··on A rudimentary simulation of the three-body problem
It is true that simulations produce results which may not reflect the underlying system if the simplifications and fudge factors are incorrect. Thus fiddling with parameters is part of the process.

In the example I gave of searching for zeolite catalysts, the simulations were just used to identify candidates for labs to study. I don't remember the exact numbers, but I think it brought the list of candidates down from hundreds to less than 10. The majority of these candidates were at least somewhat effective. Unless we believe that pretty much all of those hundred candidates would have been effective, then the advanced calculations were doing some work.

The question is, is all that work actually just done because of parameter twiddling? I don't think so. Consider that neural networks are often used lately in order to provide computationally simpler models of various physical phenomena. They can do a somewhat better job if fed with a lot of real data, but they use at least thousands of times more parameters than the simple quantum physics calcs with fudge factors. Thus I think it is safe to say that the structure of the quantum physics calcs does meaningfully model some part of reality. (Unless, as xvector points out, our memories are being continuously overwritten to make reality seem consistent)

It's also good to note that the fudge factors (read: parameters) and quantization are done because it would be too computationally difficult to model the parts of the system modeled by fudge factors for systems with a useful amount of atoms in them, and we just don't know how to compute ODEs for complex systems in continuous time and space. In simple systems, (e.g. 2 photons interacting) analytical solutions for ODEs can be found, no fudge factors are needed for computation, and the computed results match the experimental results to within measurement error.

genrilz··on A rudimentary simulation of the three-body problem
I always find skepticism like this to be really interesting, since in the end we could always be getting fooled by the Deus Deceptor or something. That being said, let me take a stab at being anti-skeptical for the fun of it.

I work around people who do "Computational Chemistry", which is basically running quantum physics calculations. These tend to be done in order to either understand the properties of materials, or to understand the reasons why reactions happen. The results are more advanced materials and better performing reactions. An early and famous example of such technology is the laser. A more typical modern example would be searching for Zeolite catalysts which have particular properties, or trying to create surface coatings which protect implants from being eaten by the immune system, or on which ice cannot freeze.

Basically, I believe the advanced calculations to be correct because they lead to things which are (eventually) used in daily life.

genrilz··on Is parallel programming hard, and, if so, what can you do about it?
I've read an earlier edition. I would say it is more of a overview of the different concerns and design patterns that someone doing parallel programming would find useful. For instance, it gives you an overview of how CPU caches work, and uses that to motivate why doing synchronization is expensive. It uses this to both motivate doing as little synchronization as possible, and giving you ways to design systems which don't need (as much) synchronization, as well as to explain why RCU works so well as long as you don't need to update the synchronized structure often.

If you want a good overview, I'd recommend reading the roadmap in section 1.1. Also, I know a 600 page book (400 pages if you exclude appendixes) is a bit long, but I really enjoyed both the material presented and the style in which it was written. Hopefully that makes the length feel a bit less intimidating.

genrilz··on Making It Legal to Play Outside: “Reasonable Childhood Independence” Bills
I don't see what you are talking about here? The article is discussing passing laws to prevent CPS from charging people with child abuse for not helicopter parenting their children. As other people have mentioned, children aren't in much danger and so this increases their independence. I assume by children-out-of-wedlock problem you mean that there are more single parents now? (I have not checked if this is true) I don't understand how more parents being single is related to the CPS/busybody neighbors/police overreaching more though.
genrilz··on The First Naked Kiss on Camera: Eadweard Muybridge, Sex and Murder
I'm bisexual myself, and so would be most prone to gay interpretations of things. However, when I read the article I assumed the standard probabilities for sexual orientation (roughly 85% straight, 15% gay or bisexual) because they were paid to be models. I think that the gay interpretation of things mainly occurs over private correspondences with sensual undertones, or people who lived together in a way similar to marriage.
genrilz··on What “diversity and inclusion” means at Microsoft
Currently, the hiring process is opaque in pretty much every respect. I think this probably is a benefit to companies, since if they list the metrics they use, candidates will optimize for those metrics instead of actually being good at the job (Goodhart's law)

Thus, I don't think we can expect the hiring process to become transparent anytime soon. It is known that the (opaque) hiring process does discriminate based on race. If I had to guess, the policy mentioned in the article is a direct response to papers like [1], which show that simply changing a person's name to be more "ethnic" results in their application being considered less. Thus I don't mind if someone tries to opaquely enforce a rule like the one in the article to counterbalance this.

[1] https://www.nber.org/papers/w29053

← PreviousPage 3 of 4Next →