Greg's recent activity
-
Comment on Advice on new PC build in ~comp
-
Comment on Advice on new PC build in ~comp
Greg Link ParentIf you’re seeing QD-OLED for $75 let me know where because I’m buying four! Strong opinion from me is that the panel itself becomes the real differentiator nowadays in ways that just refresh and...If you’re seeing QD-OLED for $75 let me know where because I’m buying four! Strong opinion from me is that the panel itself becomes the real differentiator nowadays in ways that just refresh and resolution won’t tell you. The colours, contrast, and (perhaps the biggest impact that most people don’t expect) motion rendering thanks to instant pixel response on an OLED are significant for what it’s like to use day to day. I won’t claim to know MSI’s alphabet soup to judge that one in particular, but they make decent QD-OLED displays in general (a friend has one that’s very solid) and I’d call that a good price from what I know. I’m also on team QD-OLED over WOLED from what I’ve seen of both, although that is getting down into the finest of fine differences.
Cooler, it’s an AIO, everyone uses them nowadays (for better or worse!). I actually do kinda agree with you, I’d only bother liquid cooling a CPU if it’s part of a custom loop to cool the GPU anyway, but if you look at the 9800X3D it is getting towards the spec limits for an NH-D15 or similar. Extra fan surface area by putting them side by side on the AIO rather than one giant one that’s gotta still fit between the motherboard and side panel can win out at lower RPM if you do it right.
Oh and @Atvelonis: great build in my opinion, one thing I’d say if it were me is go Corsair for the PSU if you can. Nothing wrong with MSI PSUs that I know of, but Corsair are the guys with the “absolutely bulletproof, exceeds spec in almost every destructive test” track record that I’d say you want in the part that’s got the theoretical ability to fry all the others.
-
Comment on Origin of ECDSA curve seed provided by NSA to NIST in 2000 without explanation turns out to be simple PRNG output with UNIX timestamp as seed in ~comp
Greg Link ParentI’d guess it’s the same as why they often use the “cannot confirm or deny” answer to questions. If they started giving detailed explanations of things they do that aren’t sketchy, it makes the...I’d guess it’s the same as why they often use the “cannot confirm or deny” answer to questions. If they started giving detailed explanations of things they do that aren’t sketchy, it makes the lack of an explanation when they are doing something sketchy carry much more weight.
-
Comment on Anyone using a DGX Spark? in ~comp
Greg Link ParentYeah it sounds like you very much fall into the niche that’ll actually get value out of these, and more broadly I do think that most people who buy at least two of them and actually use the 200GbE...Yeah it sounds like you very much fall into the niche that’ll actually get value out of these, and more broadly I do think that most people who buy at least two of them and actually use the 200GbE are also mostly in that bucket! The rack full of eBay and AliExpress parts from datacenter teardowns plus the dedicated heat pump to cool the damn thing is a commitment all its own…
I was more spitballing for the “audience”, especially after the price rises, because I’ve actually had a few acquaintances ask me for advice on this recently (which has then been “try managed inference first, then a cloud VM, then buy a Strix Halo or Mac Studio once you know your use case”). But those aren’t the people building robots or worrying about microsecond-level RDMA timings!
I guess in a way it's both compute and a diversified portfolio (is that how watch people justify their expenses?)
I despise unnecessarily gendered memes, but if ever there were a good case for a “boy math” tag… (and I say that knowing 100% I’m no better!)
-
Comment on Anyone using a DGX Spark? in ~comp
Greg LinkThey’re only approaching worth it if you need to experiment with 200GbE and RDMA, or really need CUDA specifically, in my opinion. Which is no small thing, that’s an incredibly important part of...They’re only approaching worth it if you need to experiment with 200GbE and RDMA, or really need CUDA specifically, in my opinion. Which is no small thing, that’s an incredibly important part of the tech stack to have experience with if you’re going to be deploying across nodes in a datacenter, and being able to play with that hands on is pretty damn useful! But everything else the Spark does I see done better elsewhere.
If you’re going “true” homelab you can get better performance and the same 200GbE networking with used GPUs and ConnectX cards from eBay, so as I see it the Spark specifically is for companies/universities that want to give a few engineers some datacenter-like boxes to play with without allocating actual servers that could be doing real work, or scouring eBay to hack something together like us plebs.
If you want local models specifically and don’t care about the networking side, a 128GB or (soon) 192GB AMD or Apple device with unified memory is the way to go for a sane option, or a box of old 3090s is the price/performance king if you don’t mind figuring out power draw and PCIe contention and things.
To actually run LLMs efficiently, you won’t beat the 1/1000th of a server that you pay for a few hundred thousand tokens at a time from CoreWeave or Runpod or Nebius or whoever, those are optimised to within an inch of their lives and kept at capacity 24/7 to get the most out of the capital spend. Local running is for privacy or education, not cost. And privacy/education are both super important, you’ve just gotta know you’re paying a premium for them, sadly.
Robotics is the one thing I’ll say could change the equation if it’s still really NVIDIA focused: not my field, so I can’t tell you anything useful you don’t know there, but if there’s significant vendor lock in it could make sense. That said, the stuff I see in other parts of the industry is wayyyy more GPU-agnostic than it was even a year ago, so it’s less of a push at the single device side for me (they’re still the kings for raw performance if you’re speccing a big server though).
Source: have a rack of used GPUs on 100GbE network for my consulting business (scientific ML, not LLM; I need less VRAM and more fp32 performance, which is why that makes sense. Also I bought before the rampocalypse), chose a 128GB HP G1a (AMD Strix Halo) when the company I’m currently working for separately on LLM stuff asked what hardware I needed. I’m sure I could request a Spark (probably even two plus a network switch) from them if needed, but I wouldn’t spend my own money and I didn’t see the need over the Strix Halo laptop even when I was spending someone else’s 🤷
-
Comment on Given everything we know about AI so far, what do you think it can or cannot be good for? in ~tech
Greg Link ParentThis is one of those interesting nuances in the “AI” conversation: that’s true of the commercially provided models as configured, not necessarily of the technology itself. RLHF does naturally bias...This is one of those interesting nuances in the “AI” conversation: that’s true of the commercially provided models as configured, not necessarily of the technology itself. RLHF does naturally bias towards certain styles of response in a lot of cases, but even that can be overridden if there’s appetite to do so.
It’s another analogy to the internet itself: the capabilities that actually come from a planet-wide network of billions of interconnected devices are incredible, you can see the depth if you start considering all the ways society would collapse if “get information from A to B quickly without caring where A or B are” went away, but the dominant experience for most people is Reddit, Meta, Amazon, Google, TokTok, and maybe one or two major media conglomerates.
The internet (comms layer) and “the internet” (5-10 major web platforms) are both globally important in different ways, as is the fact that a handful of companies do control so much of it is another important factor in and of itself.
Same goes for AI (imprecise label for a rapidly advancing field of machine learning, usually covering LLMs and often covering other deep neural net developments too) and “AI” (3-4 major LLM providers). The massive providers are shaping the conversation and the economy and it’d be absurd to ignore that, but the tech itself has, is, and will be transformative (heh) independently of them too.
-
Comment on Our AI midwife in ~health
Greg (edited )Link ParentAlthough as an interesting tangent for this specific topic, a lot more negative scans, especially ones subtly close enough to positive scans that humans might misjudge them, is exactly what we...Although as an interesting tangent for this specific topic, a lot more negative scans, especially ones subtly close enough to positive scans that humans might misjudge them, is exactly what we need to feed into an image classifier to negate that problem. It’s obviously not ethical to scan people for that reason alone if it’d require more invasive follow ups to identify positive/negative definitively, but it is worth keeping in mind as an actively changing level of risk at the individual and population level right now.
Depending on how good the stats are right now (my knowledge on this specific subtopic is a couple of years stale at this point), the classifiers may already be good enough from existing data to reduce the false positive rate to an acceptable level and justify, say, 20% more scanning of uncertain cases than would be done now because the signal to noise ratio has already shifted in the right direction and the individual risk of missing a diagnosis is now higher than the risk of a false positive. Thus giving more scans to analyse, thus giving better classifiers, thus reducing the risk of doing scans!
-
Comment on Our AI midwife in ~health
Greg Link ParentWholeheartedly agree. But. There are known and widespread flaws in how medicine is practiced relating to women’s health. There are known and widespread flaws in how medicine is practiced in...Wholeheartedly agree. But. There are known and widespread flaws in how medicine is practiced relating to women’s health. There are known and widespread flaws in how medicine is practiced in general. And in how academic research is disseminated, and in how software is documented and tested, and in how business information is managed, and in how organisational knowledge is used, and in how…
And I really do get you, I’ve been the one hand wringing at obvious failures for most of my life because they bother me. But that’s the thing, I’ve been doing it for all those decades and these problems aren’t getting fixed because the fix is often “if people would just…”, and people generally don’t collectively “just <whatever>” even when the outcome would benefit them and everyone else.
So if we can use the machine to quickly and broadly patch human process gaps that’ve been embedding themselves for decades or centuries, I see it as a massive win. Perhaps not to the level the AI execs are promising, because I’m increasingly convinced most of them have lost their goddamn minds, but certainly to a level that’ll bring world-changing positives (will they outweigh the world-changing negatives those execs are pushing along the same path? Unclear, if I’m honest).
So yeah, it shouldn’t have happened, but if we’ve got an individual-level fix here and now and it can change things when all the plausible systematic fixes are just being ignored that’s still pretty exciting to me.
-
Comment on Do you think you're special? in ~talk
Greg Link ParentI really appreciate this, all of it, and I wanted to say a big thank you in particular for ”I try to focus less on how far I am from what I could be, and more on how far I am from where my worst...I really appreciate this, all of it, and I wanted to say a big thank you in particular for ”I try to focus less on how far I am from what I could be, and more on how far I am from where my worst traits would like to drag me”. It’s such a succinct framing of a concept that’ll do me a lot of good to hold on to, and that’s very meaningful to have.
-
Comment on Do you think you're special? in ~talk
Greg Link ParentThis is a beautiful take that I agree with wholeheartedly. And for what it’s worth I did always read the optimistic view on Arthur Dent and his sandwich making, I saw it as a glimmer of hope even...This is a beautiful take that I agree with wholeheartedly. And for what it’s worth I did always read the optimistic view on Arthur Dent and his sandwich making, I saw it as a glimmer of hope even in Adams’ clearly deeply depressed writing in that last book.
So, with the hope that people will understand the context I’m coming from when I say this and the deeply mixed feelings I have about it: I was the “gifted kid” who never hit that wall, who people kept seeing that way well into my adulthood and career. And I am proud of that, it does drive me in a lot of ways, I do have some of the material rewards to show for it. I’d be colossally ungrateful to really complain about it.
But it’s deeply isolating in a lot of ways too, it creates a distance that my decidedly unexceptional and frankly somewhat below average social skills often aren’t up to bridging. It comes with a host of mental illnesses and related struggles that I have to believe are the other side of the coin on whatever parts of my brain and mind make certain technical skills feel more natural to me than talking, like all my capacity was dumped into the latter at the cost of the former. And even then, I’ve never once felt I’m really living up to my potential, however hard I try and whatever other people tell me.
I guess what I’m saying is I wish more people shared your understanding and looked for the whole diverse array of niches and contexts that people can excel in. Not only would it be much needed recognition for the people who fit within those niches, it’d take some of the laser focused pressure off the areas that people do already think of too.
-
Comment on Off the Norwegian island of Fedje lies a German World War II submarine believed to contain a huge load of mercury – why salvaging it has become a race against time in ~humanities.history
Greg LinkI wonder why they can’t freeze the containers to prevent leaks before lifting them? I’m guessing they can’t, because they know what they’re doing and I very much don’t, but video game logic...I wonder why they can’t freeze the containers to prevent leaks before lifting them? I’m guessing they can’t, because they know what they’re doing and I very much don’t, but video game logic absolutely makes me want to see them pulling up giant mercury-filled ice cubes until the wreck is safe.
-
Comment on Do you iron your clothes? in ~life
Greg Link ParentTrousers category definitely covers leg wear that isn’t shorts or skirts in general, so yeah actually I guess that’s the broad equivalent and it just doesn’t have the confusingly more specific...Trousers category definitely covers leg wear that isn’t shorts or skirts in general, so yeah actually I guess that’s the broad equivalent and it just doesn’t have the confusingly more specific meaning attached as well like shirt does! Although I guess I probably would say “jeans” or whatever if I were referring to a specific pair someone was wearing, so maybe it’s the opposite in that it sounds natural for the category as a whole but not for a specific instance? I think… maybe? It’s also occurring to me that I don’t even remember when I last had a conversation about trousers and the semantic satiation is really setting in.
-
Comment on Do you iron your clothes? in ~life
Greg (edited )Link ParentI am indeed British, so that’s probably part of it, although I’m like 98% sure I’ve heard people use it similarly in the US (I work in tech where the assumed default is jeans and t-shirt and most...I am indeed British, so that’s probably part of it, although I’m like 98% sure I’ve heard people use it similarly in the US (I work in tech where the assumed default is jeans and t-shirt and most of the people are on the West coast, and am male, if those add context!).
And yeah linguistically fuzzy even then. “Shirt” as a category, especially when considering dress codes like the post above? Totally clear delineation in my mind it implies buttons and a collar, wouldn’t question it. Friend is wearing a cool band t-shirt? “That’s an awesome shirt, did you get it at the gig?” is also totally clear in my mind and doesn’t sound weird at all referring to something without buttons or a collar. Once again, yay language!
[Edit] And now you’ve got me thinking about it, polos are a wildcard because they do have buttons and a collar but I wouldn’t default them into the shirt category for that specific example. But I might in other situations where t-shirts would still be implicitly excluded 🤔
But then only upper-body-wear gets this privilege of flexibility! Pants exclusively refers to underwear in British English (except if you’re from Liverpool, I think?), there’s no “pants category means underwear but you might compliment someone’s pants meaning outerwear” option there, you’d just be assumed to be American, ESL, or just weird if you have a local accent!
Don’t even get me started on bread rolls or small connecting alleyways between streets…
-
Comment on Do you iron your clothes? in ~life
Greg Link ParentFor me shirt implies buttons, t-shirt is not within the category “shirt” (yay language!).For me shirt implies buttons, t-shirt is not within the category “shirt” (yay language!).
-
Comment on Where do we go from here? (Regarding AI) in ~tech
Greg Link ParentA wonderful quote to live by, from a faith I know less about than I’d like to, so thank you for that. Framing personal salvation as not the purpose of religion appeals to me in a meaningful way...A wonderful quote to live by, from a faith I know less about than I’d like to, so thank you for that.
Framing personal salvation as not the purpose of religion appeals to me in a meaningful way (not that I’m going to find myself becoming religious, but in terms of where the philosophical purpose can be found). It’s hard to say that most faiths are embodying that: I rarely to never hear religious people or leaders putting service over salvation, I see them “selling” salvation and if we’re lucky using that as a means to encourage service to earn the reward. If more religions (as practiced, not just “no true Scotsman” readings of scripture) framed earthly good as an end rather than a means, I’d be substantially more comfortable with them.
On the nation states side of things, I’m not just going on numbers, I’ve spent significant time over the course of a few decades travelling in places right across the spectrum of development and of religiosity. I’ve had incredibly warm welcomes almost universally, and even then some that truly stand out have indeed been in extremely religious countries. I also consciously chose not to name names on the countries I was painting with the less positive brush: I specifically wasn't trying to tie it to economic development alone and for what it’s worth I consider the US as a soft entrant in the “more religious” grouping.
So yeah, I’m absolutely not dismissing the close ties, the community, the togetherness that does genuinely seem to exist more strongly in places that out outside the Western individualist bubble. But. And it’s a big but. I’m also not willing to wear rose tinted glasses when a so called “good person” perpetuates stark sexism, or homophobia, or racism, or corruption, which I’ve also personally witnessed as extremely strong in the more religious parts of the world and, as a microcosm, even the more religious parts of the US since you mentioned it specifically. Those are all longstanding traditions too, explicitly or implicitly baked into many of the same religious structures that bring closeness and community. And bluntly: I’ll toss out the community bonds if the price to pay for them is abuse of an out group. I don’t think it is the price, I think we can have the good without the bad, but that needs a genuine shake up to happen.
I chose the countries and metrics I mentioned in that first post carefully. Not economic powerhouses, not measures of comfort, just measures of trust and happiness that my subjective experience says do exist in those places without the negatives that often come with them in more paternalistic strong communities.
cc @lostwax
-
Comment on Where do we go from here? (Regarding AI) in ~tech
Greg Link ParentI don’t think I’ve ever been more ambivalent about something I’ve read! In a lot of ways I agree, the concept of collective societal aspiration and a duty to an abstract sense of morality rather...I don’t think I’ve ever been more ambivalent about something I’ve read!
In a lot of ways I agree, the concept of collective societal aspiration and a duty to an abstract sense of morality rather than to blind and selfish wealth maximisation does (at least subjectively) seem to have declined sharply.
But I see plenty of worse examples historically, plenty of situations where that abstract morality was either rewritten to justify the exact same kind of wealth maximisation (just within a strict hierarchy rather than the more freelance shittiness we see now), or it was used directly as a cause célèbre for atrocities.
I also don’t see what your goal for an outcome is. I’m fascinated by religion, more so in my experience than a lot of people for whom it’s a daily or weekly observance, but I don’t believe any of it and I’d be lying to you and myself if I claimed to. Morality is of deep importance to me, but religion is largely orthogonal to that from what I see in myself and others.
Lastly, I’d be fascinated to know your thoughts on the negative correlation between religiosity and quality of life at a nation state scale. And I’m not talking about GDP or other materialist bullshit, I’m talking societal trust, equality, happiness. I’m talking Denmark or Iceland or New Zealand vs any highly religious country you’d care to mention.
-
Comment on Where do we go from here? (Regarding AI) in ~tech
Greg Link ParentAnd it’s the perfect analogy, whether either of you meant it to be or not! A movement triggered by the economic and political concerns you rightly point out, but one that responded by direct...And it’s the perfect analogy, whether either of you meant it to be or not! A movement triggered by the economic and political concerns you rightly point out, but one that responded by direct attack on the automation technology itself.
Then, as now, nuance is hard, and introspective nuance that hits one’s sense of self and sense of purpose is really really hard. I imagine some Luddites saw attacks on equipment as a means of direct action towards political and economic change and equitable distribution of the gains of automation. I also imagine a hell of a lot of them were justifiably scared and unbalanced by a change that shook their economic and social status, and absolutely wanted to hit pause and halt the technological change itself.
I see a lot of breathless corporate bullshit pushing AI in unjustifiable ways, and I see a lot of genuinely transformative power in the technology. I see a lot of very reasonable anti-AI backlash because of that corporate behaviour, and I see a fair amount of unexamined blind anger against the technology itself too.
90% of the time, when someone frames an issue as “taking jobs”, they’re contributing to the anti-technology rather than anti-corporate argument whether they intend to or not. The implication of “taking jobs” is that the jobs should remain, despite a machine being able to do them instead. Sometimes there’s a quality of work argument attached, which is where things get more nuanced again, but often it’s a genuine belief that jobs are a necessity and that blocking technological progress is preferable to eliminating them.
Wealth redistribution and work reform are legitimate anti-corporate arguments around properly allocating the gains of automation for the good of society as a whole, but both require an uncomfortable conversation about (a) the fact they probably won’t happen in a serious way without the kind of societal change that’s historically brought a whole lot of bloodshed as well, (b) they require a widespread severing of the “paid employment -> sense of self worth” link that many people hold dear and that needs to happen before a critical mass of people are willing to seriously push for reform (in a hopefully-nonviolent-but-let’s-be-realistic-about-history way).
-
Comment on Router recommendations? in ~comp
Greg Link ParentMy view might be skewed from only using their crazy >100GbE stuff, but is Microtik the way to go for someone who considers themselves not that techie? I didn’t think they had a particularly strong...My view might be skewed from only using their crazy >100GbE stuff, but is Microtik the way to go for someone who considers themselves not that techie? I didn’t think they had a particularly strong UX focus even on the more consumer grade stuff.
-
Comment on Router recommendations? in ~comp
Greg Link ParentFor what it’s worth the UCG series (their consumer/small office routers), and I think the UDM series as well (same market segment, WiFi integrated rather than separate, although I strongly prefer...Unless they've changed significantly, Ubiquiti makes unfortunate choices with their management software being arranged around needing a constant, and often very heavy, management service, arguably to sell cloud services and extra devices
For what it’s worth the UCG series (their consumer/small office routers), and I think the UDM series as well (same market segment, WiFi integrated rather than separate, although I strongly prefer separate), have pretty solid CPUs for the price and run the controller software locally nowadays.
I get the impression they’re moving away from the separate controller (either software or hardware) that used to be their default and just beefing up the gateway devices to be the controller as a way to either make for a cleaner user experience, or increase vendor lock in, depending on your viewpoint.
-
Comment on An AI researcher writes about his crisis of faith in ~tech
Greg Link ParentReally? Maybe it’s my bubble, but I’ve heard a ton of developers in the field discussing the various paths we see future dev work taking, with theories anywhere from “there will be even more human...Really? Maybe it’s my bubble, but I’ve heard a ton of developers in the field discussing the various paths we see future dev work taking, with theories anywhere from “there will be even more human dev work a couple of years when slop code works its way into critical systems and starts imploding”, through “we need architects, but half the jobs are going to be gone soon and nobody’s training juniors anymore”, and right up to “fuck it, I’m off to become an HVAC tech”.
And to put my cards on the table here: I want my job automated away, assuming a machine can do it well enough. We don’t need jobs, we need purpose and we need the means of acquiring food and shelter. The former I can find for myself, and if it really comes to it I’d rather fight tooth and nail for equitable distribution of the latter than waste my life emulating something a computer can (in this hypothetical) do just as well because I’m forced to facilitate some cargo cult justification of wealth distribution.
Yeah they’re genuinely impressive, the only LCD I’ve seen that I think meaningfully outperforms a good OLED is the Apple Studio XDR, and that costs $3,500, so… yeah. I’m three years in on an OLED TV with no signs of burn in, but obviously that’s a lot less likely to have a static image on it for 8+ hours/day!
My new work laptop also came with an OLED display and I do keep that turned off when it’s docked; it’d proooobably be fine based on the numbers just above, but for the sake of an extra bit of display space off to the side of my big monitors that I’d rarely actually use I just went with the minor inconvenience of disabling it when plugged in. On that basis an LCD would’ve actually done me better there, so it’s not totally clear cut, but I do think the newer panel tech broadly lives up to the hype.
I’ve been meaning to try a CRT-style line by line renderer on my TV actually, I’m interested to see if it gives a better perception of motion for cinematic content (CRT lines had a natural brightness falloff as the frame refreshed and it was good for naturally smoothing out the perceived transition between frames; we’ve come full circle now and can potentially replicate that by rendering multiple frames on a high refresh display with the brightness gradient calculated appropriately because it’s got the pixel response for it to make sense and we’re still watching sometimes as low as 24fps content on displays that can often do 10x that or more!)