tauon's recent activity

  1. Comment on What about having an LLM teach you to code? in ~comp

    tauon
    Link Parent
    You might be interested in Matt Pocock’s teach skill as well. I haven’t personally tried it yet, but it looks quite promising. There is also an accompanying video which introduces the skill a bit...

    You might be interested in Matt Pocock’s teach skill as well. I haven’t personally tried it yet, but it looks quite promising.

    There is also an accompanying video which introduces the skill a bit more.

    2 votes
  2. Comment on Does generative AI have a natural limit without a major innovation? in ~comp

    tauon
    (edited )
    Link Parent
    Without delivering concrete proof here, I am fairly certain most major American labs, and for sure (like, confirmed by them) some of the Chinese labs known for “distillation” work (Moonshot,...

    Without delivering concrete proof here, I am fairly certain most major American labs, and for sure (like, confirmed by them) some of the Chinese labs known for “distillation” work (Moonshot, Zhipu/Z.ai, DeepSeek, Alibaba’s Qwen), are already using synthetic training data, which is to mean data originally produced by an LLM (or a specific/deterministic code-driven process), and then (eventually) fact-checked and/or refined by a human.

    It’s worked pretty well so far in the cases that were published, for example Moonshot’s Kimi K-model series:

    [Step] 4. Simulate Usage of the Synthetic Agents: The team simulated multi-turn tool-use scenarios in order to generate “trajectories” – a fancy way of saying the detailed set of steps documenting the inputs and steps models take to accomplish their goals. Some of these scenarios simulated “users” – fake people with diverse communication styles – interacting with these agents, while others simulated autonomous usage.

    Edit: This is not to say I believe synthetic training data, for LLMs specifically, will get us to “AGI”/further-than-human intelligence. I’m sure there’s an inherent quality ceiling we’ll encounter somewhere.

    1 vote
  3. Comment on Does generative AI have a natural limit without a major innovation? in ~comp

    tauon
    (edited )
    Link Parent
    To pile onto that, ErdosBench(mark) is a thing now, comparing multiple models’ performance on (previously not published) adaptations of Erdős problems.

    To pile onto that, ErdosBench(mark) is a thing now, comparing multiple models’ performance on (previously not published) adaptations of Erdős problems.

    4 votes
  4. Comment on Arch User Repository compromised, 1500+ packages affected in ~tech

    tauon
    Link Parent
    Does the AUR provide a mechanism to determine package/PKGBUILD (if even applicable?) “orphanaged status” (or if the user providing PKGBUILD recently changed)? That way, in a next step users could...

    As a user, uninstalling any orphan packages is probably for the best.

    Does the AUR provide a mechanism to determine package/PKGBUILD (if even applicable?) “orphanaged status” (or if the user providing PKGBUILD recently changed)?

    That way, in a next step users could set up rules for e.g. warning, then blocked execution for orphans over a given number of months (or the reverse and requiring a cooldown-before-use for the case of submitter-not-maintainer user having changed).

    4 votes
  5. Comment on Access to Fable and Mythos 5 cut off after US government order in ~tech

    tauon
    Link Parent
    Context for the uninitiated, and in a bit more detail here.

    Context for the uninitiated, and in a bit more detail here.

    3 votes
  6. Comment on Any fellow software engineers using paid GitHub copilot? in ~comp

    tauon
    Link Parent
    This is the part I don’t understand about the current “AI push”: Isn’t awarding employees by most LLM (token) spend just “paid by LOC” all over again? Has management (in the aggregate) really...

    Coworkers were actively tracked down and applauded for being top AI spenders.

    This is the part I don’t understand about the current “AI push”: Isn’t awarding employees by most LLM (token) spend just “paid by LOC” all over again?
    Has management (in the aggregate) really learned nothing, or alternatively, forgotten every lesson from that time?

    I mean, even if you’re not familiar with the SWE side of this (hi)story, maybe you ought to have heard of Goodhart’s law instead if you’re in a position to green-light these massive budgets with absolutely no way of tracking, let alone predicting them?
    I know I’d be uncomfortable if my company could just sack my role as the responsible person once things went south.

    9 votes
  7. Comment on Claude Fable 5 and Claude Mythos 5 in ~tech

    tauon
    Link Parent
    Mea culpa, that’s a great point. If the goal was a very large, general-purpose model, I imagine research that’d come out of potentially resource-constrained environments [in comparison to the “AI”...

    Mea culpa, that’s a great point.
    If the goal was a very large, general-purpose model, I imagine research that’d come out of potentially resource-constrained environments [in comparison to the “AI” big players] would focus on algorithmic and size optimization, regardless. It’s still cutting-edge after all, just potentially at a smaller scale and/or helping achieve smaller hardware requirements.

    (As an aside, I’m a little embarrassed, but I just now realize the same is actually true for my alma mater: they have both the huge data/computing center as well as a nuclear reactor, too – although I’m not sure how much free computing capacity could be scheduled there towards an intensive task like (very) large language model training.)

    5 votes
  8. Comment on Claude Fable 5 and Claude Mythos 5 in ~tech

    tauon
    (edited )
    Link Parent
    It’s an interesting thought, but with what funding? Even excluding the knowledge and skills necessary, the hardware requirements alone probably already halt this idea. Edit: Not even necessarily...

    It’s an interesting thought, but with what funding? Even excluding the knowledge and skills necessary, the hardware requirements alone probably already halt this idea.

    Edit: Not even necessarily the GPUs alone are a bottleneck, just (ideally redundant) storage for the raw training data would already be prohibitively expensive for most entities or orgs entertaining this idea…

    2 votes
  9. Comment on "Teachers are going to hate it": How social media apps hooked teens at school in ~tech

    tauon
    Link Parent
    You are absolutely correct, and there’s plenty examples. AdTech has done so much harm on a societal and personal/psychological level to so many, and hopefully one day we’ll get to wonder how there...

    Fines are not enough. They need to be punished severely and never trusted again.

    You are absolutely correct, and there’s plenty examples.

    AdTech has done so much harm on a societal and personal/psychological level to so many, and hopefully one day we’ll get to wonder how there was a time it wasn’t banned for under-18 year olds. Maybe for everyone.

    16 votes
  10. Comment on What change would make you quit Tildes? in ~tildes

    tauon
    Link Parent
    Aside from “just” preventing spammy/bot signups (which as an aside is something we here IMO take for far too granted), probably the better part of Tildes being pleasant to use, unreasonably...

    Aside from “just” preventing spammy/bot signups (which as an aside is something we here IMO take for far too granted), probably the better part of Tildes being pleasant to use, unreasonably effective, etc. lies in being invite-only, and users generally being restricted to one account (throwaways/alts notwithstanding).

    On average, most people here seem to self-select towards being nice and generally chill anyways, but if you were to misbehave, you’re probably out for good pretty quickly, whereas elsewhere, you could just head straight back to the signup page after receiving a community/user/site-wide ban or block. Thus people might think twice about how strongly their posts are worded, or if yet another rebuttal-and-disagreement is actually needed in an argument that’d get heated if it were to take place on another platform. Here, we see something unfortunately rare: People can and will just “agree to disagree” in a civil discussion.

    The (relatively) smaller scale helps with this too, of course. You’d probably be more inclined to start a shouting match with a stranger, potential bot/paid actor, than someone who is clearly a recurring character, perhaps even recognizable by name, with their own interests and opinions.

    Same for inviting: Nobody is inviting assholes in large numbers, since the invitation tree makes it both clearly (admin-)visible where it all started, and trivial to prune.


    TL;DR: Keep invite-only accounts to ensure Tildes’ quality. Anyone can read along anyway, and there’s a multitude of ways to get invited.

    5 votes
  11. Comment on What are people's experiences with using Kagi? in ~tech

    tauon
    Link Parent
    Totally fair point, in the end much of it is going to come down to personal preference for sure. Two notes: Waiting on the request to go through DDG despite not actually needing their...

    Totally fair point, in the end much of it is going to come down to personal preference for sure.

    Two notes:

    • Waiting on the request to go through DDG despite not actually needing their servers/search engine for resolving the bang is a “known” issue, see https://unduck.link for an example that’s trying to alleviate that by caching the redirects logic.
    • As a practical example for bangs where it’s less effort than trying to configure all browsers, potentially on a device not my own; I land on more than just the English Wikipedia depending on device/browser locale setting, and !w.en <query> (or !w.de, etc.) guarantees I land where I intended to go.
    3 votes
  12. Comment on Tildes Survey #7: What is your gender identity? (Results) in ~talk

    tauon
    Link Parent
    Hm, that's a tough one. On the one hand I totally see your point, but I'd also personally love to see a result that's not skewed towards e.g. heteronormative respondents by default/by...

    Hm, that's a tough one. On the one hand I totally see your point, but I'd also personally love to see a result that's not skewed towards e.g. heteronormative respondents by default/by unintentional questionnaire design, lest we get a false impression of the average Tildes user :-)

    2 votes
  13. Comment on What are people's experiences with using Kagi? in ~tech

    tauon
    Link Parent
    Regarding ethics: you win some, you lose some, and nobody's perfect. Their transition into a Public Benefit Corporation two years ago was one of the more outstanding "winning" moments, however....

    Regarding ethics: you win some, you lose some, and nobody's perfect.

    Their transition into a Public Benefit Corporation two years ago was one of the more outstanding "winning" moments, however. Hearing that the almighty dollar (and by extension, company profits) isn't the be-all-end-all for a Palo Alto tech startup was so refreshing and reassuring, and them clearly demonstrating that such a style of company culture is very much possible, maybe even beneficial to growing likeminded customers, was a boon to the entire industry.

    9 votes
  14. Comment on What are people's experiences with using Kagi? in ~tech

    tauon
    Link Parent
    This just reminded me that most people have never even experienced the sheer power of bangs, since somehow they're only a feature in DuckDuckGo (who originated them), Kagi, and maybe some other...

    I get so angry when I have to search on Google on someone else's computer.

    This just reminded me that most people have never even experienced the sheer power of bangs, since somehow they're only a feature in DuckDuckGo (who originated them), Kagi, and maybe some other search engines (Brave too, I believe?), but not available in Google search.

    3 votes
  15. Comment on What are people's experiences with using Kagi? in ~tech

    tauon
    Link Parent
    And, just as an FYI to anyone reading this, as it was noted in one of the previous discussion threads on Kagi here: It’s not just default-off, you can actually fully disable even little things...

    And, just as an FYI to anyone reading this, as it was noted in one of the previous discussion threads on Kagi here: It’s not just default-off, you can actually fully disable even little things with the “AI” features, such as the question mark at the end bringing up Quick Answer, outright.

    (Side note, but I love their settings being both clearly labeled and sensibly categorized: kagi.com/settings/ai is both completely reasonable to expect as a settings category for any reasonable “tech” company, and nice to remember)

    6 votes
  16. Comment on My Accessibility Stack and the future on Wayland in ~comp

    tauon
    (edited )
    Link
    Thanks for posting. This has unfortunately been on the horizon, clearly visible for all, and I truly mean all, to see coming (I’m not even using a Linux desktop, only via command line on servers,...

    Thanks for posting. This has unfortunately been on the horizon, clearly visible for all, and I truly mean all, to see coming (I’m not even using a Linux desktop, only via command line on servers, and yet was aware about the lack of accessibility in Wayland vs. X11, although I only knew about issues on the “output” side, as the article calls it).

    Sadly, due to the way (F)OSS works, unless there are vocal maintainers pushing for something to be made (or huge sponsors donating money with a dedicated purpose), whether that be a new protocol/standard or implementation, I think it just won’t come into existence at all.

    8 votes
  17. Comment on Outsourcing plus local AI will soon become more economical vs frontier labs in ~tech

    tauon
    Link Parent
    Yeah, now I wanna see return-per-million-tokens (since that is what’s already being billed). I know it’s gonna be a positive outcome/ROI for most people using LLMs, but exceedingly likely not for...

    Yeah, now I wanna see return-per-million-tokens (since that is what’s already being billed). I know it’s gonna be a positive outcome/ROI for most people using LLMs, but exceedingly likely not for all users.
    The next question after that would be how to prevent your users from sliding into that second group, from a customer to the big labs company’s perspective, and how to reliably measure that in the first place.

    That’s probably a whole new business category: tracking and controlling (both in the accounting and the “setting a ceiling/limit” sense) LLM spending in an org.

  18. Comment on Donating 80% while it still counts in ~society

    tauon
    (edited )
    Link
    Great read! This is highly in- and aspirational behavior, although I am quite glad to read footnote 1, that despite the two even drawing on savings in order to be able to give more, due to their...

    Great read!

    This is highly in- and aspirational behavior, although I am quite glad to read footnote 1, that despite the two even drawing on savings in order to be able to give more, due to their high-performing stocks it’s still not bordering on what could be otherwise seen as almost self-destructive action.
    I was especially surprised to read childcare as an item in their expenses, I guess donating a share and absolute amount of income – and by extension, future wealth – this large is something I didn’t expect from “ordinary” (non-billionaire/uber rich) people who have children.

    Apart from that, the amazingly straightforward “three futures” consideration will be occupying my thoughts for the next while, too.

    6 votes
  19. Comment on I'm going on vacation in ~tildes

    tauon
    Link Parent
    In addition to the hierarchical tags as described in the Tildes docs, for my own posts, I try to ask myself: "If filtering for a topic, what would be good categories for this post to appear in so...

    In addition to the hierarchical tags as described in the Tildes docs, for my own posts, I try to ask myself: "If filtering for a topic, what would be good categories for this post to appear in so that I’d find it even if I didn’t know the title?"

    And that seems to work quite well, and most of the time my own existing tags needn’t be changed, only amended with new ones I didn’t think of adding :-)

    6 votes
  20. Comment on Excerpts from actual one-star Amazon.com reviews of books from Time’s list of the 100 best novels from 1923 to the present in ~books

    tauon
    Link
    These are absolutely hilarious, thanks for sharing. Not even three in and I’ve found a new quotable motto.

    These are absolutely hilarious, thanks for sharing.

    If I was the author of this book I’d tell myself to get a grip on the real world.

    Not even three in and I’ve found a new quotable motto.

    19 votes