It's hype to get Trump to regulate the import of competitive Chinese models. A regulatory capture play.
21,392 karma · joined July 27, 2007
When you're talking about business, the software itself doesn't really matter at all. Accountability is everything. There has to be a throat you can choke, an ass you can drag onto the carpet, when it fails. That is the reason why Oracle persists as an organization: they assume accountability for the data storage layer, freeing their clients to worry about other things.
This also has implications for developers grousing about "being an accountability sink" for AI-generated code. Being an accountability sink is the business and as Larry has demonstrated, quite lucrative indeed. Developing the code is a side effect that can now be done close to 100% by machine.
The PGC was perhaps the first GPU for PC systems. It had its own 8086 processor on board, which would accept graphics commands through a ring buffer shared with the host system, and the 8086 would translate those into frame buffer operations it needed to perform. A few years back some clever git found an exploit to allow arbitrary code to be run on the PGC's 8086.
The Tandy 2000's color graphics was not spectacularly great by any means, nowhere near comparable to high-end solutions like the PGC. It was a 96 KiB framebuffer divided into 3 bitplanes, only one of which was mapped into system memory at any one time. You changed which plane was accessible by poking one of the video registers. There were also some palette registers that let you change which 8 (out of the standard IRGB 16-color set) were visible on the screen at once. Since drawing a full-color display meant 3 plane selects followed by 3 writes to video RAM, it was slow to produce such a display and there were "temporal color fringe" artifacts as the writes to different planes were momentarily visible. But it was still great for CAD and DTP applications, especially for the money, and good enough that Microsoft used it as a development platform to build a color GUI for Windows, while Macintosh was still committed to producing the best looking, black and white display they could for the money.
Only a few expansion options were available, all from Radio Shack: a color graphics option (640x400x8 colors in 1983!) for which you had to procure the matching monitor, a memory expansion (up to two could be installed for a total of 768k, by stacking RAM chips some aftermarket vendors got this to 896k), a bus mouse adapter/real-time clock combo board, and I believe an external hard disk controller. One expansion, the "TV/Joystick expansion" which would give the 2000 composite out and access to joysticks for gaming, was planned but (to my knowledge) never released. As far as I know only one graphical game saw commercial release for the 2000: a special build of Microsoft Flight Simulator.
GNOME is a project whose manifesto is called "Let's Make Unix Not Suck", for a very specific value of not sucking roughly equivalent to "Let's make Unix into a shitty version of Windows". I'm picking on GNOME because its development psychology has infested the Linux ecosystem, yielding horrors like NetworkManager, dbus, frickin' systemd, et weary cetera. There's a sense that Unix-style composability, modularity, and (conceptual) simplicity is something that has to be hidden from the poor end user, but the people who are implementing the means to do so lack the organizational competence to pull it off. So the absolute state of things for Linux desktop use in 2026 is a tradeoff between jank and brokenness on the one hand and being exposed to Linux's technical underside on the other. I don't mind so much because I've never minded getting technical. I'm the kind of guy who still logs in at the VT console and types 'startx' to get a graphical session going because I came up on early Linux and appreciate being reminded of what underlies that UI convenience. And it sure beats being stuck on Windows or Mac, where it works until it doesn't, and when it doesn't you often can do little but shrug your shoulders and say "wellp, them's the breaks".[1] And that's before adding in AI cruft, advertisements, mandatory online accounts and vendor spyware.
The price to pay for having a hot rod is a lack of a luxury interior from the finest Swiss craftsmen. But a) you have a hot rod; b) it won't sell your driving data to insurance companies!
[1] In the 90s I heard stories from hackers about normies who just kind of assumed that computers were flaky, fragile machines that lost your work every now and then and needed to be restarted at least once a day in order to function. And when they heard of Linux or FreeBSD, their minds were blown because they didn't know computers could do that. But the computer wasn't flaky or fragile; Windows 9x was.
Back in the 2000s, one of my favorite programmers from the Before Times—Fabrice Bellard—came up with an algorithm for computing digits of pi that was so much better than the state of the art, he beat with a $3000 PC the record for speed of pi calculation, which had been established with a supercomputer.
Some bright spark—maybe they'll have Claude assist, maybe it'll trigger Claude's safety locks, I dunno—is going to derive a way to do transformer or other frontier AI inference in a really efficient manner on consumer hardware, making local models practical to run on the PCs people have today and nearly, if not actually, competitive with today's frontier models. "There is another theory which states that this has already happened"—perhaps several times already, and we just need a few more goes of the cycle. I find that for supposedly a crowd of the smartest dudes in the room, people in the Valley like to do things the stupid way around, by sheer fucking brute force: scale up massively, burn billions and trillions of VCbux on capex, aggressively conserve developer effort at the expense of everybody's CPU time and storage capacity—yours, your cloud provider's, your customers'... So I suspect, but at this time cannot prove, that there's a whole lot of headroom for optimization of AI inference, which has the potential to put some truly powerful models in the hands of ordinary people.
And that's very dangerous, because so much for pacing the frontier! If Joe Schmoe can build a mini-AI lab in his mancave, he can then gain access to a model that gives him all the big scary things Dario and Sam are worried about: instructions on how to build bombs or bioweapons, child porn generation, human sacrifice, cats living with dogs, mass hysteria! If you're concerned about an AI-mediated workplace panopticon, Joe Schmoe could, in principle, build one of those for his auto body shop with a local model. The only way to stop Joe from doing so is to stop Joe from acquiring the equipment it would take to run such a model.
Pacing the frontier is at least as much about pacing what you can compute as it is about what goes on in billion-dollar data centers. Any argument for limiting AI growth and development—and admittedly, there are some strong ones—is an argument against general purpose computing. So if we're really going to regulate AI, bend over for the laws that mandate OS/firmware/hardware-enforced restrictions on what binaries your computer may run. And so long open source. Oh well, those laws were coming anyway; that's what the age verification thing is all about. Society has determined that people can't be trusted with computers.
And unfortunately that motion clarity comes at a cost: flicker. Some people, like my sister, cannot bear that cost. She has photosensitive epilepsy, but even before that ripened into full-on seizures in her 20s she became profoundly uncomfortable playing video games, and never got truly into them until flat panels predominated. Even without photosensitive epilepsy, flicker can be annoyingly perceptible and cause eyestrain even if you don't consciously register it. Most of the motion blur mitigation strategies we see on panels, like black frame insertion, suffer from the same problems. The only way we'll come close to solving both problems at once is with very high speed panels, but until then I'll take a bit of blur, thank you.
What I've found—based on limited experience with a "modern" arcade cabinet (Fix-it Felix Jr., a tie-in promotion to the 2012 film Wreck-it Ralph)—is that if you have an LCD or OLED that can do true enough blacks, the effect is very similar to having a brand spanking new CRT, even in arcade applications.
But that goes only for raster games. Nothing we currently have can replace the razor-sharp, completely unaliased lines of an original Asteroids and its vector display.
All I know is that when CDs hit the market I savored the additional detail they provided. A good, non-loudness-crushed recording had so many sumptuous highs I used to reach my fingers out and try to "feel" the consonants and vowels of the vocals. And when LCD monitors began to supplant CRTs my eyes were relieved at the additional sharpness. I remember these transition points, so I can say definitively that there was a time when we thought the newer stuff was indeed an improvement.
"There shouldn't be any goddamned high places!!!"
I used to call that the "Sony-mandated camera spin" because I half-seriously thought that spinning the camera over the field and characters was a prerequisite for getting a PlayStation devkit in the first place. It's in so many JRPGs, it was in Bust-A-Move (Bust-A-Groove), the dancing game, when one character solos, etc.
Sony actually had policies like this; for example, sprite-based PlayStation games were popular in Japan but near-forbidden in North America, because SCEA's marketing strategy was to emphasize 3D polygon graphics and convey a sense that the world had moved on from sprites. Of course, much like modern mobile policies, if you were big enough the rules could be bent for you. Street Fighter, Mortal Kombat, and other tentpoles could thus release sprite-based games.