Greg's recent activity
-
Comment on I wonder if years from now hand written code will be antique in ~tech
-
Comment on Help me choose a HiDPI monitor for work in ~tech
Greg Link ParentDidn’t even think of Thunderbolt potentially being disabled, that’s another really good point in favour of multiple inputs! I don’t have the new laptop in hand to check yet, bureaucracy is still...Didn’t even think of Thunderbolt potentially being disabled, that’s another really good point in favour of multiple inputs! I don’t have the new laptop in hand to check yet, bureaucracy is still doing its thing, but the possibility of a security policy change killing Thunderbolt later even if it works here and now is a meaningful one to consider if I went the Apple route, so thanks for that.
The signalling on the old LGs is actually kinda interesting, my understanding is they split the monitor into two halves for signalling purposes (even though it uses a single physical TFT layer, not two bonded ones like some ultrawides do) and use two DP 1.2 streams tunneled over TB3 directly without using DP alt mode. It’s a technically compliant workaround to drive that many pixels even though DP 1.4 wasn’t common yet when they were designed, basically, but it’s also one that almost no other monitors used so it’s not well supported in practice - there’s a deep dive on the Framework forums, but the conclusion is almost definitely don’t expect them to work properly on modern non-Mac machines. No harm in trying when I have the laptop, but hopes aren’t high!
And cheers for mentioning Samsung, I’d somehow missed those. For anyone finding this later, it looks like the G80HF is comparable to the ROG, the S80HF is comparable to the ProArt, and the S90PC was in competition with Apple’s 2022 generation Studio Display but is reaching end of life. Reviews on all three vary, similarly to the equivalents from LG and Asus.
I’m leaning towards the BenQ at this point. It’s a shame to be getting a premium monitor in current year at less than 120Hz, it really does make a difference to the overall feel of using them even for productivity work, but the faster ones seem like an even less premium overall package. Plus the BenQ has more than one input, some of which are HDMI, and the total cost for a pair of those monitors is literally half that of two Studio Displays + Thunderbolt KVM to make up for the single input on each. Which would also only be 60Hz 🤷
-
Comment on Help me choose a HiDPI monitor for work in ~tech
Greg (edited )LinkSome updates, after going down a bit of a search rabbit hole: ProArt is the cheapest (as low as £500 right now), but 60Hz only and a matte display. Mixed reports on uniformity and overall quality,...Some updates, after going down a bit of a search rabbit hole:
- ProArt is the cheapest (as low as £500 right now), but 60Hz only and a matte display. Mixed reports on uniformity and overall quality, no local dimming - but at this price it’s easier to overlook those things
- ROG is next up, around £700, still matte but up to 180Hz (or 300Hz at reduced resolution, but that’s not especially relevant to me). Similarly mixed reports, again no local dimming
- BenQ is about £900, and placed as a direct competitor to the Apple Studio Display. Glossy display, Thunderbolt, generally good feedback on uniformity and quality - but only 70Hz, still no local dimming
- LG is £1100, matte, 165Hz, full array local dimming. But a lot of reports of quality control and uniformity issues, and potentially flaky firmware
- Apple’s “basic” (lol) model is £1,500 and everyone loves it, but it’s also 60Hz only which I think is kind of a deal breaker at that price
- Apple XDR is the undisputed king in basically every way, but near enough three fucking grand. I don’t really need the local dimming, and I might well have paid the premium for the non-XDR over the others if it were 120Hz, but it’s really hard to justify doubling the price of the next most expensive option just to get a decent refresh rate
- The Apple displays do apparently play well enough with Windows as long as you have a Mac to configure them with, but you’ll also need a Thunderbolt KVM because they only have one input, so that adds ~£300-400 to the overall cost
- DP alt mode (if available) or a DP-USBC cable apparently does the job for most Windows machines on the Apple monitors as long as you don’t need to change any monitor settings [edit: specifically a bidirectional cable, and you’ll still lose camera and audio I think, so using an alt mode capable port seems to be preferred if possible]
- If you do need to use an HDMI-only machine on an Apple display, it looks like choices are either an active HDMI-DP converter (which mostly seem a bit janky, and loses the camera and audio feed) or a £650 THK401-X4-V3 KVM. Which only supports one monitor…
-
Help me choose a HiDPI monitor for work
It looks like we’re finally getting some real options in the space, so I’m interested in people’s thoughts and experiences! I’ve been running a pair of LG 27MD5KL monitors forever, and they’ve...
It looks like we’re finally getting some real options in the space, so I’m interested in people’s thoughts and experiences!
I’ve been running a pair of LG 27MD5KL monitors forever, and they’ve been largely great - but they’re really showing their age at this point, and I’m being forced to upgrade either way because my new job requires using a Windows laptop (likely a Dell Pro 14, but TBC), and those old LGs use an esoteric dual link DP variant that only Apple supports. I’ll also be keeping my personal MacBook Pro, so smooth compatibility with both is a must. Use will largely be software work: lots of documentation, lots of terminal windows, lots of dark mode, but some design review work as well where decent colour accuracy helps.
Options I’m seeing at the moment are:
- Asus ProArt PA27JCV
- Asus ROG Strix XG27JCG
- LG 27GM950B
- BenQ MA270S
- Apple Studio Display
- Apple Studio Display XDR
I get the impression that Apple’s the better option where possible, but I’m concerned about compatibility, especially when I don’t directly control the choice of laptop. If anyone has real world experience on that one I’d be interested, especially the difference between “technically works on non-Apple laptops” vs “is an equivalently good experience on non-Apple laptops”.
I’m guessing just using separate inputs for each laptop will be fine on whatever I go for, but I’m open to getting a KVM if needed (L1Techs? I haven’t actually checked how hard it is to find one that supports 2x 5K yet…). Couldn’t immediately see anything that’ll take Thunderbolt input from the Mac and HDMI input from the Windows laptop and push both to Thunderbolt output, which might otherwise have been a way to bridge the Apple display question.
Probably going for 27” 5K rather than 32” 6K, unless there’s a compelling argument for the latter. I briefly tried a Dell 32” 6K a few years back and other than horrible backlight bleed, sitting that close to a 32” display ended up feeling like I was craning my neck rather than moving my eyes to see edge to edge.
So, what do people think would be best here?
Sending up a bat signal for @gary and @ButteredToast - you guys both know your stuff on this one!
8 votes -
Comment on Mesh LLM: distributed AI computing on iroh in ~comp
Greg Link ParentI’d be interested to see concrete numbers on performance vs link speed on this, figure out how much of a bottleneck it really is if running on a local network. I wouldn’t want to run it with...I’d be interested to see concrete numbers on performance vs link speed on this, figure out how much of a bottleneck it really is if running on a local network. I wouldn’t want to run it with internet-level latency, but for inference the data flow between layers is predictable enough that a couple of ms at two or three sensible breakpoints might not be a killer. Maybe.
Seems like a relatively straightforward experiment to run for someone with access to 2-3 DGX Sparks or similar, at least: just write a script to loop the same test and record the tok/s each time with the network cards in every driver supported mode from baseline 1GbE to full 200GbE with RDMA, PFC, and chocolate sprinkles.
-
Comment on What are the new EU border checks and how will they affect your summer holiday? in ~travel
Greg Link ParentAlmost definitely. Same way evacuating Fukushima was worse than telling people to stay would have been. But you also can’t trust large groups of people to be rational about these things - I can’t...Almost definitely. Same way evacuating Fukushima was worse than telling people to stay would have been. But you also can’t trust large groups of people to be rational about these things - I can’t even trust myself to be rational 100% of the time, hard as I try - so I don’t really know where that leaves us…
-
Comment on What are the new EU border checks and how will they affect your summer holiday? in ~travel
Greg (edited )Link ParentI’ve been through Lisbon airport a few times in the last few months (thankfully didn’t hit delays of the magnitude mentioned in the article, at least) and they’ve gone a step further by designing...I’ve been through Lisbon airport a few times in the last few months (thankfully didn’t hit delays of the magnitude mentioned in the article, at least) and they’ve gone a step further by designing the whole place as check in->security->food / lounges / shopping / seating->immigration->gates.
If you know it’s coming and if you get through check in and security with time to spare, you’re left with the option of getting a snack and somewhere to sit but risking a missed flight if the immigration queue balloons while you’re sitting there, or going through immigration immediately and then being stuck hungry and hot with nowhere to sit at the gate for a few hours if it’s quick that day.
If you don’t know it’s coming, which is pretty easy to miss given that exit immigration isn’t globally standard and the vast majority of other airports in the world don’t spring a surprise multi-hour queue on you in the corridor to the gates, you probably stopped to chill in the central area just like you would at every other airport and then missed your flight when the gate was called. As my friend did recently.
Aaaaand if the airport doesn’t put the gate number for your flight up until even closer to departure, and you don’t have a detailed map of this specific airport confirming whether all possible gates are connected on the post-immigration side or not (I think they are, but I actually couldn’t tell you that with certainty…) the timeline gets even tighter.
It’s such bad design that it’s honestly almost funny. It looks like it’s gonna be this way for the long term, because they were installing a whole bunch more e-gates in the same place when I passed through last week (which should at least help with the congestion). And it could so easily be fixed by putting immigration immediately after security, with a simple boarding pass check for Schengen-destination flights to skip the immigration bit. It’s genuinely weird they’ve done it this way, because it’s gonna piss off all those food and retail places paying top dollar for space on the wrong side of immigration, and it’s otherwise actually quite a pleasant airport to be in, as far as airports go.
[Edit] Clarity
-
Comment on Finland's last landline call has been made as the Nordic country becomes the latest to retire copper-wire phones in favour of fibre in ~tech
Greg Link ParentYou definitely can, it’s a function of how the tech works - but I’ll admit I’d originally thought Elisa was providing a bit more handholding with that process based on what I’ve seen in other...You definitely can, it’s a function of how the tech works - but I’ll admit I’d originally thought Elisa was providing a bit more handholding with that process based on what I’ve seen in other countries doing similar transitions, and I think you’re right that they actually aren’t!
You can either use an analog telephone adapter (ATA) to bridge the old handset to a VoIP connection on the fiber network, or a fixed wireless terminal (FWT) to bridge it to a 4G/5G connection. You’re looking at €50-70 for either option, so nothing crazy.
The bit I’d jumped over was the assumption they’d offer an ATA and the associated VoIP service as part of the transition - meaning that you don’t need to know these fairly niche terms for yourself to get set up! Turns out they’re just recommending GSM desk phones, which honestly is also pretty reasonable, but since the affected market is gonna be a small handful of older customers (and since they’re going cellular for the number transfers rather than VoIP) I feel like an FWT with the SIM installed and the number transferred so you can just unplug your existing phone and plug it into the new box would’ve been a nice option. Maybe they even do offer that if you speak to them 🤷
Also TIL the Finnish phone network used a really weird plug for analog phones for a long time, so that’s fun!
-
Comment on Finland's last landline call has been made as the Nordic country becomes the latest to retire copper-wire phones in favour of fibre in ~tech
Greg Link ParentIt’s a business decision, but also a big symbolic change to a technology that’s persisted since the industrial revolution. There had to be some last call on the copper network, and having the guy...It’s a business decision, but also a big symbolic change to a technology that’s persisted since the industrial revolution. There had to be some last call on the copper network, and having the guy in charge of the country’s communication policy on it seems as apt as anyone I can think of!
The headline (and article text, actually) is a bit misleading though - it’s by no means the end of landlines in Finland, it’s just the end of copper to the home landline-only service. And it’s not like many (any?) circuits were running full analogue copper end to end in the last few decades anyway, the only difference here is that the conversion to IP traffic now happens in a little box at the user’s house rather than a mile down the road at the telecom company side. You can even keep using that same physical landline phone your parents have had since the 80s if you like.
Like @MimicSquid said, telecom services are pretty heavily regulated and they’ll still have coordinated with the government either way - you don’t get tech in use for 150 years without building up some gnarly dependencies for weird embedded systems that make use of esoteric properties of the infrastructure that they should probably never have relied on in the first place - but for pretty much any end user, including the 107 year old who thinks computers are the tool of the devil, this is a billing change and a 20 minute engineer visit to switch over the wiring and then you go back to doing things exactly the same as you’ve always done.
-
Comment on US Supreme Court upholds birthright citizenship, ruling against Donald Trump's order in ~society
Greg Link ParentEvery aircraft you fly on has many layers of redundant systems in place that allow it to land safely even if a huge number of things go wrong. You should still be very worried about the situation...Every aircraft you fly on has many layers of redundant systems in place that allow it to land safely even if a huge number of things go wrong. You should still be very worried about the situation if you just landed a plane where several key systems failed in ways that should not be possible, and only that very last redundant backup prevented total catastrophic failure. You should absolutely not under any circumstances fly that same plane again until the failed systems are repaired and the root cause of the near miss is identified and remedied.
-
Comment on US Supreme Court upholds birthright citizenship, ruling against Donald Trump's order in ~society
Greg Link ParentSeriously, it’s fucking insane. I know I shouldn’t be surprised by anything anymore, but I read the headline and literally thought “yeah, even this lot aren’t corrupt enough to go against...Seriously, it’s fucking insane. I know I shouldn’t be surprised by anything anymore, but I read the headline and literally thought “yeah, even this lot aren’t corrupt enough to go against something this obvious”. Turns out 45% of them are…
-
Comment on $22,000 per hour: assistants use a legislative loophole to outearn US surgeons in ~health
Greg LinkPerhaps a stupid question, but why are the arbitrators approving these payouts? Conflict of interest? Corruption? Apathy? Ignorance? Hatred of insurance companies? I can actually see arguments for...Perhaps a stupid question, but why are the arbitrators approving these payouts? Conflict of interest? Corruption? Apathy? Ignorance? Hatred of insurance companies?
I can actually see arguments for the assistant’s payout not necessarily being hard capped as a percentage of the surgeon’s simply because there’s always a chance the surgeon got screwed on the billing somehow, but I don’t understand how there isn’t a flag against “more than 2x what the surgeon got” or “more than $300/hr” or “more than 3x the average for an assistant on this class of procedure” or whatever else that would very easily catch only wildly large payouts and trigger some kind of review panel to presumably rein it in a bit.
-
Comment on How to make the world's best black shirt in ~life.style
Greg (edited )Link ParentYeah I’d echo that, “alright” is the word. Perfectly adequate, sometimes less consistent than I’d hope, could use a bit more attention to detail, but serving the functional purpose on an item I...b+c shirts are alright
Yeah I’d echo that, “alright” is the word. Perfectly adequate, sometimes less consistent than I’d hope, could use a bit more attention to detail, but serving the functional purpose on an item I haven’t had time to do an obsessive deep dive on - that’s why I was quite excited to see this, actually, because it seems to be the end result from someone who already has done that obsessive deep dive!
[Edit] I used to like American Apparel a lot too back in the dim and distant past, then they were hard to find (particularly outside the US) when whatever blow up caused the switch to Los Angeles Apparel, then I heard a lot of varying and sometimes conflicting stories about the CEO being creepy and weird to a degree it meaningfully affected the employees, and by that point I’d needed to find a different brand for basics anyway because of availability and I never got around to looking for them again and figuring out if they were actually a company I wanted to support or not and if so whether they were easy to find in EU/UK. As you can probably tell, clothing is one of those things that I care about to a degree, but not so much it often manages to hit the top of my overly-full to do list if I’ve found something that kinda works…
-
Comment on How to make the world's best black shirt in ~life.style
Greg LinkI live in plain black tshirts, and I’m a big proponent in general of things being done well even if it costs more. Just ordered, looks like it’ll ship in a month or so, will report back with my...I live in plain black tshirts, and I’m a big proponent in general of things being done well even if it costs more. Just ordered, looks like it’ll ship in a month or so, will report back with my totally unscientific and arbitrary opinions on whether these are actually worth 6x as much as the bella+canvas blanks I have to replace far more often than I’d like.
-
Comment on No, artificial intelligence is not conscious in ~tech
Greg Link ParentYeah, I was mostly responding to the direct question of “where would a hypothetically conscious Claude exist, if it were conscious?” with the answer “there would be many thousands of separately...Yeah, I was mostly responding to the direct question of “where would a hypothetically conscious Claude exist, if it were conscious?” with the answer “there would be many thousands of separately conscious instances, each localised to its own server (or single-digit cluster of servers depending on VRAM demands)”.
I went into a bit more detail on the consciousness question as a whole the other day (the 404 media article there also links to this one). The tl;dr on that is I don’t think current gen LLMs are conscious (and wouldn’t expect them to get anywhere close without more work into self-modification at a bare minimum), but I also don’t find the MS Word or AoE 2 arguments especially compelling because I don’t think they really capture complexity as a key part of the question.
-
Comment on No, artificial intelligence is not conscious in ~tech
Greg Link ParentIt’s just a call center: each instance of Claude on each server (or small 2-5 node cluster) would be independently conscious if they were conscious at all, and your query will be routed to one of...It’s just a call center: each instance of Claude on each server (or small 2-5 node cluster) would be independently conscious if they were conscious at all, and your query will be routed to one of them but not necessarily the same one. The different data centres are different offices that might contain the agent handling your query, and the server racks are rows of cubicles. Claude as a whole in this context is more like TaskRabbit or Uber than a single entity, with lots of separate Claudes all operating under one coordinating piece of chat software, if you were to fully anthropomorphise it. Honestly I could even imagine one of those “virtual PA” services in the pre-LLM days giving the company a single person’s name and having every employee use that same name for a sense of continuity.
You can get into really interesting territory around swarm intelligence and diverging clones and other philosophically weird goings on that are enabled by how we can copy, modify, and distribute LLM weights but can’t do the same with biological life - but almost none of that practically applies to current LLMs as deployed. They’re just lots of independent instances on lots of separate servers.
-
Comment on Power consumption of LLM's in ~tech
Greg LinkThe scaling from 120B to 1T+ parameters likely isn’t as bad as it might look if you just extrapolate from that post, at least based on my understanding of their methodology. They’re using the same...The scaling from 120B to 1T+ parameters likely isn’t as bad as it might look if you just extrapolate from that post, at least based on my understanding of their methodology. They’re using the same 8x A100 machine for all tests - sensibly so, to get a like for like comparison - but that’s an older machine and kinda overkill for the model sizes they’re testing, so I’d guess that it’s a good baseline for comparing the models tested but less so for getting an absolute W/token value that applies to bigger models on newer GPUs with the batching and VRAM allocations tuned perfectly.
-
Comment on Nvidia announces liquid cooling system that promises to reduce electricity consumption and cut water use by up to 100% in ~tech
Greg Link ParentCustom loops for end users are always gonna be very niche, though, and the people building them tend to do a single build or upgrade every… 3-5 years, maybe? It’s deep into amateur enthusiast...Custom loops for end users are always gonna be very niche, though, and the people building them tend to do a single build or upgrade every… 3-5 years, maybe? It’s deep into amateur enthusiast territory for sure.
This kind of system is going to be more the domain and scale of commercial building plumbers, HVAC techs, maybe people who’ve done similar installations in labs or hospitals (or people trained across from those specialties to this particular type of build), so I’d imagine it’ll be pretty substantially more reliable just from specifically trained installers that do thousands per year.
-
Comment on Does generative AI have a natural limit without a major innovation? in ~comp
Greg Link ParentSorry for the very late reply, I appreciate the discussion and I missed your response on this one! I guess what I’d say is that your line of thinking reminds me strongly of biological evolution on...Sorry for the very late reply, I appreciate the discussion and I missed your response on this one!
I guess what I’d say is that your line of thinking reminds me strongly of biological evolution on long timescales, and of human societal and technological development on shorter ones. The vast majority of random genetic mutations go nowhere, and almost none will ever be a step change for a species on their own. The vast majority of people can live long, productive, fulfilling lives without ever really hitting on an idea so new that humanity as a whole hasn’t already considered it somewhere along the way - and again, of those tiny number of truly new ideas that do happen and are valuable, the number that make a large impact on their own rather than being incremental improvements are vanishingly small.
So if that’s the case for us as people, I see the occasional model making the occasional leap beyond what was known at the time it was trained to be pretty absurdly impressive, really. I see at least conceptual scope for those models being able to stack those incremental improvements autonomously if we set them on a path to doing so, especially since they can be copied and fine tuned at a scale and speed vastly higher than biology allows.
It’s all far too abstract to say with any confidence that letting loose a few million self-modifying models interacting with each other in a “society” would or wouldn’t lead to true development of the “individuals” and the system as a whole - and like I said in another comment a week or two ago, I certainly don’t have access to the billions of dollars in hardware that it’d take to test that at meaningful scale with no real pitch beyond “it’s gonna be so interesting”. But seeing that they’re capable of creating new knowledge at all is enough to prevent me from confidently betting against those capabilities…
-
Comment on Blog post: 'AI stole my face and made me a digital flesh puppet' - should I publish my life's work when extractive AI is rampant all over the internet? in ~life
Greg Link ParentI’ve been meaning to come back and give this a real reply for days, because I think you’ve hit on by far the best take I’ve seen to date on the whole messy bundle of worries and conflicts of...I’ve been meaning to come back and give this a real reply for days, because I think you’ve hit on by far the best take I’ve seen to date on the whole messy bundle of worries and conflicts of interest and frightening changes and justified concerns and unjustified concerns and genuine exploitation.
I utterly despise the basic principle of the currently leading AI companies: that you can just extract other people's work and even identity (an artist's identity is necessarily present in their work to some degree or they are not really an artist), without acquiring permission to do so and profit from it, distort it, mangle it, tarnish it, basically do with it whatever you feel like. Their products and practices reflect this deeply problematic, colonialist attitude: violate vulnerable people without care, apologise later when the damage is already done and the profits are in your pocket. Unsurprisingly, these products have enabled like-minded individuals all over the world to assault anyone with an online presence in hopes of extracting value without having to learn how to produce valuable content themselves. People like that have always existed but as the author says, AI has made their activities frictionless.
I’ve seen so, so many people - most of them genuinely worried about either real or perceived threats to the place in the world they’ve carved out for themselves so far - advocating for regressive and destructive approaches that would have consequences far beyond what they realise because they start from an understanding that what these companies are doing is bad, and then try to fit that into the tangible structure of rules and regulations they only sort of understand already.
It seems like I far too frequently find myself trying to thread the needle between “yes, these companies are evil; no, the underlying technological concept is not evil and I actually find it fascinating and exciting, even if those companies’ implementations are tainted; no, the suggestion of modifying IP law in xyz way is not viable because it would make computers themselves illegal or require selective enforcement that we know will just favour a different group of wealthy sociopaths”.
Framing it as a question of identity, accepting that the wrongness is less tangible and more philosophical but still no less real, closes that gap perfectly.
What these companies are doing is wrong, but most of it probably isn’t illegal and arguably there might not even be a sensible way to make it illegal. I don’t necessarily share your feeling that the biggest changes are likely to fizzle out, I think we’ll see a lot individual companies culled just as we did in the dot com era, but the sweeping shift remains just as the internet itself did back then. But I do still think and hope that culture as a whole can adapt around these changes and understand a bit better why certain actions are a problem, rather than drawing battle lines that only bear a cursory relationship to the outcomes most people actually want.
This has always been how I’ve expected things to go too. I love the craftsmanship and sheer dedication that made Rollercoaster Tycoon happen, I love older stories about physically mapping drum memory for efficiency, and I love seeing people find techniques to squeeze ever more complex feats out of vintage hardware*.
But I can’t pretend that compilers and garbage collectors and even electron apps were a bad idea when it comes to getting shit done, even if they can be easy to misuse and facilitate a lot of slop. Same goes for coding agents, but as with many things in our thin-end-of-the-singularity cyberpunk dystopia, it goes faster and more extreme now, for better and for worse.
*Side note: why do so few people seem to notice the insane quality of ultra-specialist hardware and software work going on in the retro gaming community? These people have the skills to run rings around some of the most highly paid engineers on the planet, and they’re investing thousands of hours into random niche projects for the literal love of the game. I want every tech executive looking at a huge flashing neon sign that just says “where’s your capitalism now?” and points to some dude in a basement engineering an entire motherboard single-handedly. Anyway…