Greg's recent activity

  1. Comment on Academic authors 'shocked' after Taylor & Francis sells access to their research to Microsoft AI in ~tech

    Greg
    Link Parent
    I’d say “theft of the commons” is an excellent description of the academic publishing business model in general. They take research performed by academics, often funded by public bodies, then...

    I’d say “theft of the commons” is an excellent description of the academic publishing business model in general. They take research performed by academics, often funded by public bodies, then combine it with unpaid review work from other academics and charge the universities who did the research in the first place hundreds of thousands to access it (or a few thousand per paper up front for the privilege of actually letting people read it for free).

    21 votes
  2. Comment on Nvidia RTX 50 graphics card family TDPs 'leaked' by Seasonic in ~tech

    Greg
    Link Parent
    The difference really is mind blowing sometimes! As soon as you edge into “there’s some level of budget justified for this” it’s like the suppliers hear “money no object, add a nice big multiplier”.

    The difference really is mind blowing sometimes! As soon as you edge into “there’s some level of budget justified for this” it’s like the suppliers hear “money no object, add a nice big multiplier”.

    1 vote
  3. Comment on Nvidia RTX 50 graphics card family TDPs 'leaked' by Seasonic in ~tech

    Greg
    Link Parent
    Got you, and I really appreciate the info - it sounds like pretty much exactly the use case I’m looking at: enough to run a few different dev jobs in parallel, or one somewhat larger job more...

    Got you, and I really appreciate the info - it sounds like pretty much exactly the use case I’m looking at: enough to run a few different dev jobs in parallel, or one somewhat larger job more quickly, with anything significantly bigger going off to the actual servers. Probably means that worrying too much about bus bandwidth would be overkill, which is very much what I was hoping to hear.

    1 vote
  4. Comment on Nvidia RTX 50 graphics card family TDPs 'leaked' by Seasonic in ~tech

    Greg
    Link Parent
    Very good to know, thank you! I’m guessing no FSDP or similar in your scenario, in that case? If it’s just a case of ruling out the fancier kinds of parallelism that could definitely work for us.

    Very good to know, thank you! I’m guessing no FSDP or similar in your scenario, in that case? If it’s just a case of ruling out the fancier kinds of parallelism that could definitely work for us.

    1 vote
  5. Comment on Nvidia RTX 50 graphics card family TDPs 'leaked' by Seasonic in ~tech

    Greg
    Link Parent
    Bit of a tangent, but have you guys noticed any issues with PCIe bottlenecks on that setup? I’m speccing a similar machine and it’d be great to stick with AM5 rather than Threadripper if possible...

    Bit of a tangent, but have you guys noticed any issues with PCIe bottlenecks on that setup?

    I’m speccing a similar machine and it’d be great to stick with AM5 rather than Threadripper if possible (aside from the big cost advantage, there’s always something that ends up benefitting from extra single-core performance), but I’m having trouble finding info on how much difference running the cards at x8/x8/x4 makes in reality.

    1 vote
  6. Comment on Nvidia RTX 50 graphics card family TDPs 'leaked' by Seasonic in ~tech

    Greg
    Link Parent
    Given the 10x price gap for the A and L series, I’m more than happy for them to keep wedging the xx90 cards into the “consumer” lineup regardless of how much sense they might make there! I’d been...

    Given the 10x price gap for the A and L series, I’m more than happy for them to keep wedging the xx90 cards into the “consumer” lineup regardless of how much sense they might make there! I’d been hoping the Titan branding would come back with 36 or 48GB VRAM, but I guess they decided that’d too heavily cannibalise A6000 sales.

    7 votes
  7. Comment on Computer monitors that are good for watching videos? in ~tech

    Greg
    Link Parent
    Small additional note on the GPU side: it has no impact on the image quality you’re getting on the existing monitor, and physically can’t affect the viewing angle, but a decade old card might have...

    Small additional note on the GPU side: it has no impact on the image quality you’re getting on the existing monitor, and physically can’t affect the viewing angle, but a decade old card might have limited or missing compatibility with newer features on current monitors, and that could limit the panel to below what it’s capable of.

    HDR and refresh rate are the two major ones, with the former being the most likely to be missing. Monitor tech has moved a lot in the last few years, so any high quality modern IPS or OLED will be a big upgrade regardless - and viewing angle is likely to be a non-issue - but you won’t necessarily get the full brightness and contrast if you’re limited to SDR by the interface on your GPU.

    I really would recommend taking a look at 2023-24 generation QD-OLEDs in person, by the way. They aren’t cheap, so whether they’re worth it is extremely subjective, but they are one of the biggest jumps in colour reproduction we’ve seen in a long time. For the kind of person who keeps their tech for 10+ years, the investment could be a good one.

    1 vote
  8. Comment on A handful of US grocery stores now have ammo vending machines in ~news

    Greg
    Link Parent
    I find the concept pretty uncomfortable, as a reflection of US gun issues as a whole and for some of the practical reasons people have mentioned here, but putting that aside and treating is as...

    I find the concept pretty uncomfortable, as a reflection of US gun issues as a whole and for some of the practical reasons people have mentioned here, but putting that aside and treating is as more abstract I can see a reasonable case for the why.

    I’d imagine it’s solving a problem for the vendor more than the customer: the product can’t just go on the normal shelves, and there are presumably specialist supply chains to deal with too, so wrapping it in a vending machine makes it viable to serve a lot more locations where it wouldn’t be reasonable to have dedicated kiosks and staff. Best (and IMO fairly realistic) case, it’s about capturing planned purchases from customers who otherwise would’ve needed to drive somewhere else and now don’t have to.

    But yeah, if ever there were a product class where tighter regulation could justify reduced convenience…

    5 votes
  9. Comment on USB and the myth of 500 milliamps in ~engineering

    Greg
    Link Parent
    I feel this one! I think the USB-IF’s utterly insane approach to branding, combined with Amazon et al’s Wild West of product claims, means there’s not a lot of hope for most users if they need...

    I feel this one! I think the USB-IF’s utterly insane approach to branding, combined with Amazon et al’s Wild West of product claims, means there’s not a lot of hope for most users if they need something specific.

    The one saving grace for me is that we’ve also passed the point that even mid-range is now way overkill for 95% of actual use, so it somewhat balances out. Does that cable fully comply with 100W PD like the listing says? Probably not, but it was £7 for a three pack and my phone only supports 30W anyway. Is it 3.2 2x4 or 3.1 4x2 or some other essentially arbitrary combination of numbers that denotes >10Gbps speed? Well it’s going into a portable hard drive that’ll do 2Gbps on a good day, so it’s going to be fine regardless.

    Frustrating to have to fall back on that, but at least I can generally put it out of my mind. Display cables (either USB-C or HDMI) are the notable exception: pushing 48Gbps over one of those is pretty common even with an upper mid range modern display, and whether that’ll work without dropouts is an absolute crapshoot in my experience. That and tunnelling PCIe gen 4 over USB4, but that’s one I accept is at the edge of the spec and I’m one of like six people actually using.

    7 votes
  10. Comment on The spectacular failure of the Star Wars hotel in ~movies

    Greg
    (edited )
    Link Parent
    Secret Cinema are the absolute masters of this IMO, although primarily aimed at adults rather than kids - they even managed to license The Empire Strikes Back for one of their early productions...

    Secret Cinema are the absolute masters of this IMO, although primarily aimed at adults rather than kids - they even managed to license The Empire Strikes Back for one of their early productions back in 2015, pre-Disney buyout. They’ve actually worked with Disney since (they did Guardians of the Galaxy), so I guess anything’s still on the table - my only concern would be that Secret Cinema themselves were hit very hard by the pandemic and then bought out in 2022, so perhaps they’re on that inevitable slow decline in quality for short term profit too nowadays.

    The “we were robbed” bit in the video hit me surprisingly hard. Like @phoenixrises said further up, enshittification is a tired term, but holy crap is it apt for the situation. It feels like everything is going that way, and for some reason seeing it in an ultra-premium offering from a brand built on “magic” brings an even sharper focus to the million more mundane ways it’s happening that we can’t escape so easily.

    [Edit] And I just watched another little clip of old Secret Cinema and realised that show was at Printworks. Which we’ve also lost entirely. Sigh.

    5 votes
  11. Comment on Google Cloud accidentally deletes UniSuper’s online account due to ‘unprecedented misconfiguration’ in ~tech

    Greg
    Link Parent
    They’ll definitely have a named account manager and customer engineer, so I imagine it got picked up pretty quickly and then immediately turned into a very bad day for everyone in the chain it got...

    They’ll definitely have a named account manager and customer engineer, so I imagine it got picked up pretty quickly and then immediately turned into a very bad day for everyone in the chain it got passed up to. That tipping point where your organisation spends enough for interactions with a provider to flip from faceless ticketing system to actual specific human with a face and a direct email address and phone number is a game changer.

    For me, that makes it all the worse being stuck in the lower circles of first line support hell the rest of the time: there’s incontrovertible proof that these companies can do support properly, they just either don’t care about average customers at all or they actively want to make the support experience frustrating enough that we’ll go away and stop bothering them.

    15 votes
  12. Comment on There used to be a people’s bank at the US Post Office in ~finance

    Greg
    Link Parent
    For a huge number of customers, a bank’s primary and often only purpose is a safer, more comfortable alternative to stuffing cash under the mattress. You might be comfortable installing and...

    I don't understand the first two, why do I need fraud protection or FDIC if I self-custody?

    For a huge number of customers, a bank’s primary and often only purpose is a safer, more comfortable alternative to stuffing cash under the mattress.

    You might be comfortable installing and securing a proper safe and keeping your assets in bullion (literal or metaphorical) - you might even consider it preferable - but the vast majority of people absolutely do not want to deal with that, or with its digital equivalent.

    Even at the most basic practical level, I’d expect to pay a lot more in disaster insurance if I’m looking after my own valuables rather than leaving it to the professionals with centralised security.

    11 votes
  13. Comment on I was the poster girl for OCD. Then I began to question everything I’d been told about mental illness. in ~health.mental

    Greg
    Link
    This seems closer to home than a lot of articles I’ve read, but it still swings across at the end to a satisfying conclusion, and I think that’s a problem. I understand why it does - the very act...

    This seems closer to home than a lot of articles I’ve read, but it still swings across at the end to a satisfying conclusion, and I think that’s a problem.

    I understand why it does - the very act of thinking experiences through leads us to want a neatly wrapped up lesson and the comfort of closure, the act of writing them down makes that doubly true, and the act of writing for an audience practically requires it. But if we want to have a more meaningful conversation around mental health we need to admit that for now, we (i.e. humanity) are still way out of our depth.

    That’s no reason to give up, or to undervalue the massive scientific and social progress in the field over the last few decades, but it’s important to acknowledge how much we don’t know. Give it 50 years, maybe 100, and I expect things to be radically different. Hell, 50 years ago there were no MRIs, no gene sequencing, no SSRIs, certainly no real spaces to openly talk about mental health issues - we’ve really only just developed the tools to even attempt to figure out and fix what the hell is going on inside our heads, so it’s amazing we’re already as far along as we are. But for me, at least, hearing “we haven’t figured it out yet, but we’re with you for the long haul if you’re willing to keep trying” would have brought far more comfort than being told time and again that this new round of treatment would be different, the last 15 has just been mistakes but this is correct, only for everything to collapse again a few weeks or months later.

    I did, eventually, hit on a medication that seems to work extremely well for me. I’m immensely grateful to the patient, caring, extremely intelligent medical professionals who tried to help me along the way. But the fact remains that most of them didn’t actually manage to help. It still took over a decade, a huge chunk of good luck, and ultimately one or two pretty significant coincidences for it to work out. I’d rather we collectively admit that than trying to convince ourselves that this time, surely, we’ve got it figured out.

    27 votes
  14. Comment on Have we reached peak AI? in ~tech

    Greg
    Link
    I'll lay out my bias up front here, I'm the guy in the second panel on this one - so definitely influenced by how goddamn cool I think the tech is, but I was also working with it a good decade...
    • Exemplary

    I'll lay out my bias up front here, I'm the guy in the second panel on this one - so definitely influenced by how goddamn cool I think the tech is, but I was also working with it a good decade before anyone (including me) saw it as much more than an obscure academic tool and I'm largely disdainful of the hype cycle here.

    If the main point of the article is intended to be that "AI" is overhyped, companies are throwing giant buckets of money at it with no clear plan, and the media are fuelling the whole thing with nonspecific and emotive reporting, then yeah, I'd say pretty much the same. But the author is flat out wrong on enough specifics along the way that it undermines the credibility of the whole, and incoherent enough on other points that I'm genuinely not sure whether that's even what he's intending to say.

    A few key snippets for now:

    ...one problem that has poisoned reporting on generative AI is an unwillingness to clearly describe how these things work, instead referring to its actions as some sort of magical process done on a big, scary computer.

    Agreed on this in isolation, but it seems to form a broader undercurrent to the rest of the piece that's conflating "I don't know" or "reporters don't explain" with "nobody knows". You can go read the research papers, the LLaMA or Mistral or Stable Diffusion source code, the mathematical underpinnings of the whole field - most people can't or don't want to do that, and frankly I don't blame them, but the information is there. The big players have a definite edge, but we've seen open source catching up within a few months of most major breakthroughs, leading the way on some others, and the theoretical academic papers are often published from a university lab somewhere before the techniques make it into commercial models anyway.

    Science reporting has always sucked, this field isn't unique in that. If anything it's more open than the average - the vast majority of what's important goes through arXiv and GitHub rather than commercial journals and closed labs.

    ...OpenAI's models continually prove themselves unable to match even the dumbest human beings alive

    Seriously? Text model output is far from perfect, but we're already way past Turing test territory here. AI detection is a hot topic because people can no longer be certain whether they're seeing human or machine output. And that's just text - how many average humans can create images comparable to DALL-E or 3D animation comparable to Sora? And that's just considering models from a single largely closed source organisation.

    There are sometimes telltale signs of machine generated content because the failure cases are often distinct from the way a human would fail, but that doesn't make even the bad outcomes inherently worse than human: the fact we're even comparing the models to the best of humanity's work, to the famous artists and authors rather than to the mediocrity that most of us would manage when working outside our own fields, is a sign that it's already well above the average human in many verticals.

    These models do not "know" anything. They are mathematical behemoths generating a best guess based on training data and labeling, and thus do not "know" what you are asking it to do. You simply cannot fix them. Hallucinations are not going away.

    Large language models are by far the best data parsers we've ever come up with. You need factual accuracy, you point it at the database and tell it to sift through and link back to the primary information when it's got an answer - exactly as you would with a human. We don't expect a person, even a subject matter expert, to have every fact and figure memorised; we expect them to filter information and cite sources.

    While Joanna Stern may have said that Sora's generative video clips "freaked her out," much of what makes them scary is the assumption that OpenAI will fix hallucinations, something that the company has categorically failed to do

    This is just a weird statement - hallucinations are an issue for data accuracy, that's important if you're doing something like using a text model to compile statistics, but conflating them with creative output and then calling it a blocker doesn't make sense. Images and videos don't need perfection, and when glitches are visible or distracting you'll by definition notice that and hit regenerate.

    If you stop saying things like "AI could do" or "AI will do," you have to start asking what AI can do, and the answer is...not that much, and not much more in the future.

    I can't draw, I'm a shit photographer, the last time I did any 3D modelling was in the late 2000s, the list goes on. Those are all things I can now get a model to do for me, in high quality, in the space of seconds, at a cost of pennies. AI gives me capabilities I simply didn't have before.

    Is that a good thing? Does it have implications for the value of the skilled professionals who would otherwise do those things for me (if I needed them enough and had the budget to pay, neither of which is a given)? How much will that shift the commercial or economic landscape, if at all? What does it mean for understanding of truth and misinformation? Will this all shake out to be profitable for anyone? Is copyright law going to implode? Those are interesting and meaningful questions. But to ask what the systems can actually do, and then claim that generating content on demand is "not that much", just seems absurd to me.

    39 votes
  15. Comment on Hackers can read private AI-assistant chats even though they’re encrypted in ~tech

    Greg
    Link
    I love this kind of research: it’s impressive that they managed it at all, it exposes a flaw that a lot of people just wouldn’t have thought of, and as far as I can see it gives a very...

    I love this kind of research: it’s impressive that they managed it at all, it exposes a flaw that a lot of people just wouldn’t have thought of, and as far as I can see it gives a very straightforward mitigation in just padding or chunking tokens to disguise their length.

    Haven’t had a chance to read in detail yet, but it’s interesting that Google’s platform is already immune - they do love their protobufs, so I’m thinking maybe that’s inherently obscuring the individual tokens?

    9 votes
  16. Comment on Fun programming challenge: figure out which sets of passports grant visa-free access to the whole world in ~comp

  17. Comment on Unique things to do in Las Vegas? (and Los Angeles) in ~travel

    Greg
    Link Parent
    I wandered into the Neon Museum pretty much just on the basis that I'd seen one or two cool and vaguely apocalyptic looking photos from there, and ended up having an amazing hour long chat with...

    I wandered into the Neon Museum pretty much just on the basis that I'd seen one or two cool and vaguely apocalyptic looking photos from there, and ended up having an amazing hour long chat with one of the staff. My answer to "let me know if you have any questions" was something like "I honestly don't even have enough context to know what to ask - what do you think is most interesting here?" and she just lit up, talked me through the historical context, the engineering, the weird social circumstances that led to huge fucking neon signs being a key part of the local culture, everything. Pure good luck to catch the right person in the right mood on a quiet afternoon, but it was a great experience!

    1 vote
  18. Comment on Bitcoin tops $72,000 for the first time in ~finance

    Greg
    Link Parent
    I guess it depends on what you’re counting as an advocate; the loudest and most numerous generally fall into one or more of those buckets, but there are plenty of serious financial types gambling...

    I guess it depends on what you’re counting as an advocate; the loudest and most numerous generally fall into one or more of those buckets, but there are plenty of serious financial types gambling on crypto just as they do on any other asset class.

    I see it as interesting tech that could’ve found a niche, and instead became a microcosm of our insane financial system as a whole. The numbers are made up, but not substantially more so than the ones underlying some similarly volatile derivatives.

    9 votes
  19. Comment on Recommendations for wireless earbuds for extended PC use? in ~tech

    Greg
    Link Parent
    I use mine for gaming sometimes because apparently the mic in them is much clearer for everyone else than the one in my bigger headphones, and they’ve been pretty much fine with Windows once...

    I use mine for gaming sometimes because apparently the mic in them is much clearer for everyone else than the one in my bigger headphones, and they’ve been pretty much fine with Windows once they’re connected.

    There seems to be some kind of sync issue about half the time when they power on - took a while to work out but it seems like they connect to Windows individually as you take each one out of the case, and then the audio comes through with each ear a few hundred milliseconds off from the other, which is enough to be unusably distracting. Good news is that just hitting disconnect/reconnect from the Windows Bluetooth settings once they’re in your ears seems to reliably sync them up again. I’ve had a million issues with Windows audio over the years in general, across a whole range of hardware, so that counts as fine based on the expectations I’ve got here!

    4 votes
  20. Comment on Recommendations for wireless earbuds for extended PC use? in ~tech

    Greg
    Link
    Possibly a stupid question, but if latency, charging, and cost are all important would wired be an option? That’ll likely get you the best combination of those three things if it’s a possibility.

    Possibly a stupid question, but if latency, charging, and cost are all important would wired be an option? That’ll likely get you the best combination of those three things if it’s a possibility.

    2 votes