At this point I can see parallels with Intel in the 2000s.
If you remember then, they went all in on the Pentium 4/"Netburst" architecture. It was sort of a dog from day 1. It ran hot, it wasn't very powerful, you needed purpose built power supplies and weird RAM at first. But it was the future because they planned to scale it to 10GHz which would make up for all its limitations. Sounds a lot like AI-- it's expensive and still dodgy, but the planned future design solves it all.
Narrator Voice: they never made it to 10GHz. Even at 3.8 it still sucked, and the thermals and power needs made it unviable to go further.
The LLM industry is flogging their metaphoric 3.6GHz Pentium 4 now. Maybe you can get one or two more sales cycles out of the current paradigm, but not much more. If a costlier trillion parameter model is only marginally better than a 200B model (that was only marginally better than a 25B one...), how do the economics justify the hundred-trillion parameter model? Training costs likely rise nonlinearly, too. It becomes harder to find fresh content for the model. If we need to scan vintage out of print books, we must have exhausted the easier sources.
Whether you're a true believer really chasing AGI or just surfing the grift, you're going to have to pivot hard soon if you have any business strategy that avoids commoditization. We're reaching the largest economically feasible versions of the current design. Intel survived by having such a pivot in their pocket; the replacement Core series CPUs were originally the "B-team" designs for laptop chips quickly scaled for a new use case. Does any big-dollar AI platform have something that paradigm-breaking to drop?
I know every so often there will be a small fad for making low-powere desktops from laptop CPUs.
Wouldn't soldering a mobile GPU on an x16 card be a business model if thry're more available/affordable?
I recall during the crypto fiasco they tried to sell crippled cards (not sure how much was hardware vs drivers) that wouldn't hash as fast to make it look like they were throwing a bone to gamers, but here the mask is totally off that thr core constituency is gone.
There are quite a few shops like that for reptiles, likely because of the special care requirements.
On the one hand, you don't want a minimum-wage no-training staffer who cam barely keep a hamster fed killing a $4000 lizard by not setting the heat and humidity right.
On another, it's sort of weird to sell pet and feeder rodents together.
If they said "requires a beefy dGPU or one of the handful of high-bandwidth APUs" that feels like it would make more sense than just 64G RAM. I thought most models weren't optimized to run on a CPU even with gobs of RAM.
And plenty of firms don't ship their devs 64G machines, especially in ghe current RAM crunch. Mine sent me a 16G machine in like 2023 and only replaced it with 32 on request because Win11+Docker were heavier than the old Win10/external server workflow.
Up to 1990, PRC banknotes had different portraits. I suspect many of them were "representatives" for different types of labourer or ethnic groups rather than idolizing a single figure.
The Fisher brand was named after Avery Fisher who founded the company. They were very well regarded in the tube era, but after solid-state took over, they got sold a few times and ended up in Sanyo's portfolio where they dragged around as a low-end brand until the 1990s.
I have one of their early 1980s amplifiers which is fairly heavy, but not nearly as well put together as most '70s kit.
I've been using the CoW compositor lately and it has a neat, but possibly accidental, feature:
You get that "virtual desktop with additional screen space" feature, as was common on older X11 setups, with a little minimap that shows the real screens within the context of the virtual space. What's unusual is that it's possible to move the individual screens on a multi-monitor rig independently, rathrr than as a group.
The state of the science for reptile care is advancing so fast that we sometimes still don't know what a reasonable lifespan in captivity is for some species. She might have gotten 10 years in 1980 but now 25 is feasible.
Bitch, you are so far back in the "ultimate expression of keyboard customization" that it's hilarious. This isn't Leif Erikson stepping ashore in Newfoundland and reporting that he's discovered a new world. It's a tourist landing at Toronto Pearson airport and trying to pass himself off as having discovered America.
I use a keyboard where I designed the PCB and case myself, soldered it myself, wrote the bloody firmware, and laser-etched a bunch of custom keycaps to go with my particular bindings. That's custom. And there are people who go beyond that into premium handcrafting and metalworking rather than cheap Chinese resin printing.
The "artisan keycap" industry consists of small makers doing short run novelty FOMO stuff at 50 bucks or so a throw, or you can buy something fun on AliExpress for a fiver. Is there a space for 90 dollar corporate caps?
Weather.gov has an API. Hit three endpoints- one to convert lat/long to their grid system, one for current conditions, one for forecast. Box the JSON into a pretty layout. <100 lines of TCL/Tk if you want to do it the lazy way and it still takes less RAM than an Electron Hello World.
At this point I can see parallels with Intel in the 2000s.
If you remember then, they went all in on the Pentium 4/"Netburst" architecture. It was sort of a dog from day 1. It ran hot, it wasn't very powerful, you needed purpose built power supplies and weird RAM at first. But it was the future because they planned to scale it to 10GHz which would make up for all its limitations. Sounds a lot like AI-- it's expensive and still dodgy, but the planned future design solves it all.
Narrator Voice: they never made it to 10GHz. Even at 3.8 it still sucked, and the thermals and power needs made it unviable to go further.
The LLM industry is flogging their metaphoric 3.6GHz Pentium 4 now. Maybe you can get one or two more sales cycles out of the current paradigm, but not much more. If a costlier trillion parameter model is only marginally better than a 200B model (that was only marginally better than a 25B one...), how do the economics justify the hundred-trillion parameter model? Training costs likely rise nonlinearly, too. It becomes harder to find fresh content for the model. If we need to scan vintage out of print books, we must have exhausted the easier sources.
Whether you're a true believer really chasing AGI or just surfing the grift, you're going to have to pivot hard soon if you have any business strategy that avoids commoditization. We're reaching the largest economically feasible versions of the current design. Intel survived by having such a pivot in their pocket; the replacement Core series CPUs were originally the "B-team" designs for laptop chips quickly scaled for a new use case. Does any big-dollar AI platform have something that paradigm-breaking to drop?
I know every so often there will be a small fad for making low-powere desktops from laptop CPUs.
Wouldn't soldering a mobile GPU on an x16 card be a business model if thry're more available/affordable?
I recall during the crypto fiasco they tried to sell crippled cards (not sure how much was hardware vs drivers) that wouldn't hash as fast to make it look like they were throwing a bone to gamers, but here the mask is totally off that thr core constituency is gone.
There are quite a few shops like that for reptiles, likely because of the special care requirements.
On the one hand, you don't want a minimum-wage no-training staffer who cam barely keep a hamster fed killing a $4000 lizard by not setting the heat and humidity right.
On another, it's sort of weird to sell pet and feeder rodents together.
I don't get it either.
If they said "requires a beefy dGPU or one of the handful of high-bandwidth APUs" that feels like it would make more sense than just 64G RAM. I thought most models weren't optimized to run on a CPU even with gobs of RAM.
And plenty of firms don't ship their devs 64G machines, especially in ghe current RAM crunch. Mine sent me a 16G machine in like 2023 and only replaced it with 32 on request because Win11+Docker were heavier than the old Win10/external server workflow.
Up to 1990, PRC banknotes had different portraits. I suspect many of them were "representatives" for different types of labourer or ethnic groups rather than idolizing a single figure.
For example https://en.numista.com/211574
The 100-yuan of the series features Mao with three other early leaders of the republic, but the lower notes are representatives.
Didn't they figure this out the first time two kids had to cut a cake?
Make sure the "out" group has to approve the gerrymander the "in" group drew.
If the committee that draws the maps is 75% red, the committee that approves it has to be 75% blue.
It just came out.
There was an outcry a few years back where they prepared the whole movie and intended to scrap it rather than releasing it for tax purposes.
It's not half bad, given the material they're working with.
The Fisher brand was named after Avery Fisher who founded the company. They were very well regarded in the tube era, but after solid-state took over, they got sold a few times and ended up in Sanyo's portfolio where they dragged around as a low-end brand until the 1990s.
I have one of their early 1980s amplifiers which is fairly heavy, but not nearly as well put together as most '70s kit.
I've been using the CoW compositor lately and it has a neat, but possibly accidental, feature:
You get that "virtual desktop with additional screen space" feature, as was common on older X11 setups, with a little minimap that shows the real screens within the context of the virtual space. What's unusual is that it's possible to move the individual screens on a multi-monitor rig independently, rathrr than as a group.
Mitch Mcconnel spoiled the conceot for everyone.
I'm more interested in his zoning agenda for mounting thousand-year-old castles atop Midtown skyscrapers.
Classic MacOS on clearly an x86 desktop. Absurd.
Oh, noble darkness, I'm turning into my father watching WWII movies!
The state of the science for reptile care is advancing so fast that we sometimes still don't know what a reasonable lifespan in captivity is for some species. She might have gotten 10 years in 1980 but now 25 is feasible.
Just siuth of Vienetta.
This is complete gacha thinking. "Every 10 pulls guarantees a four-star
characterintellect".How many people must be crammed on an overdrawn planet to pull Raiden Shogun, Jeff?
Oh, Heavenly Principles, please rid me of Paimon!
Bitch, you are so far back in the "ultimate expression of keyboard customization" that it's hilarious. This isn't Leif Erikson stepping ashore in Newfoundland and reporting that he's discovered a new world. It's a tourist landing at Toronto Pearson airport and trying to pass himself off as having discovered America.
I use a keyboard where I designed the PCB and case myself, soldered it myself, wrote the bloody firmware, and laser-etched a bunch of custom keycaps to go with my particular bindings. That's custom. And there are people who go beyond that into premium handcrafting and metalworking rather than cheap Chinese resin printing.
The "artisan keycap" industry consists of small makers doing short run novelty FOMO stuff at 50 bucks or so a throw, or you can buy something fun on AliExpress for a fiver. Is there a space for 90 dollar corporate caps?
K7S5A?
There was a week when it was both the cheapest and best performing mainboard you could buy.
Going next month.
The Phoenix show is the same weekend as an anime con 800 metres away. The crossover should be intense.
Yeah.
Weather.gov has an API. Hit three endpoints- one to convert lat/long to their grid system, one for current conditions, one for forecast. Box the JSON into a pretty layout. <100 lines of TCL/Tk if you want to do it the lazy way and it still takes less RAM than an Electron Hello World.