Johz's recent activity

  1. Comment on Bluesky melts down over Jesse Singal in ~tech

    Johz
    Link Parent
    Sure, that's why I'm saying that an algorithm could do the sorting for you, and figure out which posts from a person you want to see, and which posts you don't, without them having to tag them in...

    Sure, that's why I'm saying that an algorithm could do the sorting for you, and figure out which posts from a person you want to see, and which posts you don't, without them having to tag them in any way.

    4 votes
  2. Comment on Bluesky melts down over Jesse Singal in ~tech

    Johz
    Link Parent
    In theory, if one person is posting on two different topics, you'd expect those posts to be appreciated by different groups in different ways. So for someone with a large enough number of posts...

    In theory, if one person is posting on two different topics, you'd expect those posts to be appreciated by different groups in different ways. So for someone with a large enough number of posts and distinct enough groups of readers, you could automatically divide their posts up based on how many likes they get and who likes them. I assume YouTube does this to a certain extent, because there are certain channels where I only see the half of their output that I'm interested in, and the rest doesn't get shown to me at all.

    But like I say, that's something you can only do with a recommendation algorithm, and part of what I like about the Following feed is that there is no complicated algorithm.

    4 votes
  3. Comment on Bluesky melts down over Jesse Singal in ~tech

    Johz
    Link Parent
    There are multiple feeds and the "Discover" one uses a Twitter-style algorithm, but the "Following" one is just a straight chronological timeline of people you follow. I mainly use that, and then...

    There are multiple feeds and the "Discover" one uses a Twitter-style algorithm, but the "Following" one is just a straight chronological timeline of people you follow. I mainly use that, and then add people to it if I've seen an interesting post from them or prune them out if there's too much noise that I don't want to see. That works very well for me - I get a bit of US politics because it's just painfully difficult to completely shut that out these days, but most of what I see is interesting tech articles with a mix of board game and RPG news.

    There are other third-party feeds that you can subscribe to, each with their own algorithms, but I've not had much success with those.

    3 votes
  4. Comment on Bluesky melts down over Jesse Singal in ~tech

    Johz
    Link Parent
    It's a bit irritating that there's no easy way to say "I'm interested in this person's X posts, but not their Y posts". I followed a board game designer for a while, and it was interesting seeing...

    It's a bit irritating that there's no easy way to say "I'm interested in this person's X posts, but not their Y posts". I followed a board game designer for a while, and it was interesting seeing what she was working on and thinking about, except that she was also constantly posting about her sex life and I'm far too prudish and British to read about that.

    I am quite impressed that YouTube can figure out which kind of videos I'm interested in if a channel produces multiple different series or kinds of content, but I guess that's back to algorithmically driven feeds again.

    4 votes
  5. Comment on Bluesky melts down over Jesse Singal in ~tech

    Johz
    Link Parent
    I think it's similar to that. I have a list of people I follow, and those are the only posts I see on Bluesky. I normally find new people by them being reposted by someone I follow, and if someone...

    I think it's similar to that. I have a list of people I follow, and those are the only posts I see on Bluesky. I normally find new people by them being reposted by someone I follow, and if someone I follow is making too much noise I unfollow them again. My resulting feed has a bit of noise, but it's mostly stuff I'm interested in.

    I know there's also the discover feed, but that's been pretty rubbish in my experience. There are some alternative feeds that you can subscribe to, but I find they're also a bit hit-and-miss, so I mostly stick to the people I've specifically chosen to follow.

    5 votes
  6. Comment on JK Rowling dismisses Emma Watson as 'ignorant' over trans rights row in ~lgbt

    Johz
    Link
    I ended up looking up a person I used to know who came out as trans and has since made quite a name for herself as a trans anti-trans activist. I knew her vaguely through those early years of her...

    I ended up looking up a person I used to know who came out as trans and has since made quite a name for herself as a trans anti-trans activist. I knew her vaguely through those early years of her transition, and I wanted to see what she was up to now.

    Quite high up in the search results are two articles, both with titles along the lines of "why I disagree with Xyz". The first is from a trans woman saying basically "you can speak for yourself, but you can't speak for all trans people, many of us have very different experiences to you". The other is from shortly after my acquaintance wrote an article saying "JK Rowling, I get it and agree with you, but woah, you can calm down a bit on the rhetoric", and the tone of this response is roughly "see, look at this creepy MAN, we all knew he was WEIRD but now he's clearly a ULTRATRANS ACTIVIST, let's all BURN HIM AT THE STAKE".

    The difference between the two responses really strikes home that this basically feels like a way of legitimising hate for so many of the TERF activists. You see the same thing in this article from Rowling - there's no sense of "we don't agree, but that's okay", it's just pure antagonism from start to finish. It isn't about protecting women's spaces or about protecting children from harm or whatever, it's just a kind of twisted, gleeful hate.

    17 votes
  7. Comment on How many valid JSON strings are there? in ~comp

    Johz
    Link Parent
    I would think of a JSON document as being a document (i.e. file) in JSON format, in which case anything goes according to the spec, but you make a good point about document stores typically having...

    I would think of a JSON document as being a document (i.e. file) in JSON format, in which case anything goes according to the spec, but you make a good point about document stores typically having more restrictions on what's allowed.

    2 votes
  8. Comment on How many valid JSON strings are there? in ~comp

    Johz
    Link Parent
    I think some parsers only accept objects or arrays (because 99% of the time that's what you want), and some formats like Toml that are essentially based on the JSON data structure might only allow...

    I think some parsers only accept objects or arrays (because 99% of the time that's what you want), and some formats like Toml that are essentially based on the JSON data structure might only allow objects at the top level, but spec JSON is very flexible.

    2 votes
  9. Comment on How many valid JSON strings are there? in ~comp

    Johz
    Link Parent
    I'm not sure that's true. According to the spec, a JSON document is an element, which is a value surrounded by optional whitespace, and value can be a primitive, an array, or an object. Unless you...

    I'm not sure that's true. According to the spec, a JSON document is an element, which is a value surrounded by optional whitespace, and value can be a primitive, an array, or an object.

    Unless you are using "JSON document" in a very specific way here, but I've not come across it before like that.

    9 votes
  10. Comment on Taco Bell rethinks AI drive-through after man orders 18,000 waters in ~food

    Johz
    Link Parent
    Fuzzing is about putting a big variety of different inputs into a system and identifying the inputs that produce weird or unexpected outputs. In theory, it's really good at finding bugs of the...

    Fuzzing is about putting a big variety of different inputs into a system and identifying the inputs that produce weird or unexpected outputs. In theory, it's really good at finding bugs of the "wall jumping" kind, where one particular unexpected input at the wrong moment produces bad results.

    The difficulty is figuring out how to hook up the output of the fuzzer (basically a stream of random bytes) to the inputs of the system you're trying to test. Fuzzers mostly work best when the subject under test has a clear, reproducible "input -> output" structure, whereas games have a more complex "input + state + time -> output". But for complex software like compilers or databases, fuzzers are great at finding the sorts of bugs that a QA team might occasionally stumble onto.

    9 votes
  11. Comment on What are your favorite and least favorite airports? in ~transport

    Johz
    Link
    I've been enjoying flying from BER recently. There's a train station directly in the airport that's very accessible, and they've got the new scanners that make security very comfortable and...

    I've been enjoying flying from BER recently. There's a train station directly in the airport that's very accessible, and they've got the new scanners that make security very comfortable and relaxed. There are drinking fountains (never enough, but I know where they are now), and it's fairly easy to find your gate and fairly comfortable once you're there. It gets the basics right, and I think that should be more appreciated, because I've been through a lot of airports that can't even manage that...

    1 vote
  12. Comment on What are your favorite and least favorite airports? in ~transport

    Johz
    Link Parent
    Manchester Airports Group - they own Manchester and Stansted (and I thought Birmingham was well, but apparently only East Midlands), and they are all awful. I used to have to fly through a...

    Manchester Airports Group - they own Manchester and Stansted (and I thought Birmingham was well, but apparently only East Midlands), and they are all awful. I used to have to fly through a Manchester a lot, and Terminal 3 there is basically like you describe at Stansted, but without even the shopping mall - just high security labyrinths where you hope a plane will appear at some point.

    5 votes
  13. Comment on What are your favorite and least favorite airports? in ~transport

    Johz
    Link Parent
    Tegel was like this for a long time as well, until it got replaced by the new airport. Each gate had its own waiting area and security setup. In some ways it was more convenient because you only...

    Tegel was like this for a long time as well, until it got replaced by the new airport. Each gate had its own waiting area and security setup. In some ways it was more convenient because you only need to go through security for your gate, which means fewer people, but also each gate was much less efficient and the costs were surely enormous to staff each gate separately like that.

    4 votes
  14. Comment on What are your favorite and least favorite airports? in ~transport

    Johz
    Link Parent
    Them's fighting words! Every time I've come through Frankfurt, I've got lost getting in or out of the building, or wandering between locations. The last time, I was looking for a particular...

    FRA

    Them's fighting words!

    Every time I've come through Frankfurt, I've got lost getting in or out of the building, or wandering between locations. The last time, I was looking for a particular station, and I, along with multiple other, independent groups of people, ended up missing it because we needed to go through a closed, unmarked service door to get to the right place.

    The rest of the airport is fine - I'm not a fan of the food, but for airport food it isn't the worst, and like you say, you get some good views of the planes - but I just don't understand their complete inability to put basic signage up to show you where you're going!

    2 votes
  15. Comment on Forums are still alive, active, and a treasure trove of information in ~tech

    Johz
    Link Parent
    The forums I used to use often had strict rules about signatures, where images were completely disabled and there was a max height of four lines. This gave you enough space to create a unique...

    The forums I used to use often had strict rules about signatures, where images were completely disabled and there was a max height of four lines. This gave you enough space to create a unique identity out of the signature, but still avoided too much chaos.

    That said, I can't imagine signatures really working in a threaded environment, partly because the width of the message changes so much, and partly because threaded discussions are often compressed into much smaller space than flat discussions.

    3 votes
  16. Comment on <deleted topic> in ~music

    Johz
    Link
    Try looking in second hand shops or charity shops, perhaps? When I was a kid (~30 years ago), my parents had a record player with various children's records that I'm told we used to listen to (as...

    Try looking in second hand shops or charity shops, perhaps? When I was a kid (~30 years ago), my parents had a record player with various children's records that I'm told we used to listen to (as well as other records as well). Most of those have ended up in various charity shops, or possibly online.

    1 vote
  17. Comment on How social media shortens your life in ~tech

    Johz
    Link Parent
    I was also a voracious reader as a kid, and much like YouTube videos, I feel like there's both the edifying kind and the unedifying ones that just eat away your time without adding anything of...

    I was also a voracious reader as a kid, and much like YouTube videos, I feel like there's both the edifying kind and the unedifying ones that just eat away your time without adding anything of value. I do think it is harder to read books than it is to watch a YouTube video, which means that I'm less likely to get trapped in a "book hole" like I might get trapped in a YouTube hole, but I think in both cases what's needed is discernment.

    For YouTube, I simply blocked YouTube Shorts entirely (I think via a ublock query I found somewhere). The edifying/draining ratio is just too low that I do not want to see those videos ever. For the rest of YouTube, I try and notice how I feel after having watched a video. If I'm not really making it to the end, or at the end of it I just feel a bit dull, then I know not to watch more of those sorts of videos. In some cases I've unsubscribed from channels or even blocked them on YouTube to keep my feed a bit cleaner with videos I'm actually enjoying.

    7 votes
  18. Comment on The hater's guide to the AI bubble in ~finance

    Johz
    Link Parent
    Sure, but most of the time you don't need to maintain or expand your code, at least in the cases I'm talking about. In fact, I would guess that most code outside of professional projects is...

    The problem, just as with the LLM code, is once you needed to dig in and maintain or expand an existing thing, it falls apart rapidly and now you need a specialist to dive deep and reverse-engineer a codebase with 0 commments.

    Sure, but most of the time you don't need to maintain or expand your code, at least in the cases I'm talking about. In fact, I would guess that most code outside of professional projects is written once, and then either appended to or rewritten entirely if changes are needed. I'm thinking about research software, bash scripts to manage your local environment, IFTTT-style glue scripts, hell even my mechanical keyboard needs to be programmed to get the most out of it.

    And you're right, low-code solutions have existed for a long time, but they tend to be very limited because they can only do what has been programmed in by the designer. The advantage of LLMs here is that their output is standard code, which means if it can be programmed, you can (in theory) get an LLM to write it. This is also why tools like Scratch aren't really the answer: they're too basic if you're someone who enjoys programming, and they're too limiting to be useful if you're not interested in programming and just want to get stuff done. And someone needs to have preprogrammed all the behaviours you're interested in: there's no Scratch module for programming your keyboard, for example.

    I agree that in theory, anyone can learn how to program. Most people, however, don't want to. So the question becomes: how can someone who has no desire to program at all still be fully in control of their own computer? And not just in the "you own all the software (but you've got no idea how it works and you can't change it)" FOSS sense, but in the "you own a device that can truly do anything you think of, and you can come up with more things for it to do" sense. I think LLMs represent a serious step forwards in the regard.

    Fundamentally, I think this is an HCI thing. Go back to the 1970s and '80s and show researchers back then a way of telling your computer to do stuff that used natural language, was iterative, and used your own workspace as its context. They'd truly have thought you'd solved personal computing. I think that's the most underappreciated part of LLMs, and I suspect once the hype dies down, it's going to be one of the parts that most changes how we use computers long-term.

  19. Comment on The hater's guide to the AI bubble in ~finance

    Johz
    Link Parent
    I work in web development, and building proofs of concept to show how a particular feature might work or to explore whether something is feasible is definitely part of that. It's not something I...

    I work in web development, and building proofs of concept to show how a particular feature might work or to explore whether something is feasible is definitely part of that. It's not something I do every day, but it something that happens regularly.

    That said, it's not something I'm good at: like you, I like doing things "properly", taking my time, and learning something in the process. I see this particularly in little side projects: I'll spend a couple of hours setting up type-checking and proper minification for a twenty line JS script for my blog. It's interesting to me, but beyond that it's rarely productive, especially as I'm going to write that script once and then almost certainly never change it for the rest of its life.

    This sort of stuff seems to be one of the places that LLM coding systems shine: hacky solutions that don't need to be maintained, because they wouldn't be maintained in the first place. And that's a lot of code. I think those of us with careers in software development further just how much code can get way with being bad or mediocre code.

    That's what I mean when I talk about democratisation. People cannot run their own code unless they can write their own code, and most people (a) would benefit from being able to write little bits of hacky code to link things in their computer, and (b) cannot write code. There have been a bunch of solutions to that (visual coding, YAML configuration files, etc), but the benefit of LLMs is that you can now essentially program in natural language: tell your computer exactly what you want it to do, and let it figure out how to make that happen.

    FWIW, I don't use LLMs to program at all myself. I use them occasionally as a replacement for Google for things that I know Google won't be good at ("what word am I thinking of", etc), and for copy editing, and I find it very useful for these sorts of things. On the other hand, the rise of agentic programming has all happened while I've been on parental leave, so I've not really had much chance to try that out. I know some developers that I respect a lot have gotten a lot out of their LLM usage, but I know others who have found it more hindrance than help, so I've not really made up my mind on how useful it is for professional programming. (In other words: I certainly don't think I've drunk any kool-aid!)

    But I do strongly believe that for lay computer users, natural-language has been an HCI holy grail for decades, and it finally feels like something realistic, as opposed to a feature of science fiction. That's not to say that everything's perfect yet, but stories like Steven Mould's really demonstrate what the potential of LLMs can be for non-developers.

    1 vote
  20. Comment on The hater's guide to the AI bubble in ~finance

    Johz
    Link Parent
    Professionally, I can't really speak to the other things you mentioned, but in regards to this: I can't disagree with you more here. A lot of software engineering is knowing when you make software...

    LLMs can definitely help people hastily create PoC code, unit tests, crappy text, bloated prose, and a whole lot more. But I don't want to consume any of that. If you give a single flying fuck about what you're making and you put it together by hand, I'd rather consume that.

    Professionally, I can't really speak to the other things you mentioned, but in regards to this: I can't disagree with you more here. A lot of software engineering is knowing when you make software that's maintainable in the long term, and when to make crappy PoCs that will only run a few times, and a lot of code is the latter. Getting something that works and demonstrates that what you want is achievable is a really useful skill, and being able to automate that is an incredibly powerful tool.

    More importantly, it can represent the democratisation of computing. I watched a Steven Mould video the other day where he got hold of a device that could only be used by programming a microcontroller - a skill that he did not have at all. With the help of some coding AI, he was able to get it working just fine, and deliberate a handful of really cool ways the device could be used. He didn't need high-quality, artisanal code. He needed something that worked and made his computer do what he wanted. That is incredible! That has been the basic goal of HCI since Douglas Englebert sat down and gave the mother of all demos.

    I agree that there are all sorts of ethical and social issues that need to be dealt with. But I don't get this argument that LLMs are not incredibly useful.

    6 votes