Examples of group madness in technology

One of the worst arguments I hear is that “thing X is inevitable because the smart people are doing it.” As I’ve extensively documented over the last 15 years on this blog, smart people in groups are not smart and are even more subject to crazes and mob behavior as everyone else. Noodle theory, nanotech, quantum computing: plenty of examples in history of large groups of smarty pants people marching off a cliff. The same holds true for actual technological trends. I think LLMs are overrated magic-8 ball word predictors, autonomous vehicles are mostly vaporware Potemkin technologies, and we’re never going to replace internal combustion engine cars with electric golf carts, no matter how many dumb imac screens you put in them because batteries contain 20-100x less energy density than guzzoline and the grid won’t support it. These are unpopular opinions which non-thinking “smart” people will scoff at due to the ideology of continual progress (which these dorks haven’t noticed has failed for 50 years, hypnotized by their fuggin screens). Yet, my track record is better than non-thinking “smart” people because I think about things rather than uncritically accept what “smart” people tell me. It’s useful to consider the record of non-thinking “smart” word-regurgitators on a few technological examples.

Itanium: this is one I watched developing with some amusement. In the late 90s there was a thriving ecosystem of competing server architectures: (SGI) MIPS, (DEC) Alpha, (Sun) Spark, (HP) PA-RISC. All of these architectures were superior to Intel chips: most of them were more or less 64 bit clean RISC architectures. At some point in the late 90s, they all decided, quite unprompted, to get out of the hardware business and use Intel’s pending Itanium chip. I assume the main reason was MBA reasoning: designing chips is capital intensive, complicated, risky, and you can’t put clean estimates into spreadsheets which makes bean-counters sad. Why not use somebody else’s silicon? Intel has more money than any of these server companies and all the other server companies were gonna do it. Itanium, of course was a massive dud. This was something completely predictable, as the last time they tried to innovate away from their antiquated 8086 architecture, the 32-bit iAPX 432 architecture: that was also a massive failure. Same reason for failure too: they did all the hip and trendy things, making an actual stack machine with hardware support for object oriented programming, multitasking and hardware support for Ada (which everyone at the time assumed was language of the future: it still is). It was 10x slower than its much less futuristic looking competitor the 68000 from Motorola which persisted in its atavistic use of registers, which were too backward looking for the intel iAPX432. Also the compiler was hot garbage, despite the fact that it was designed to make it easier to write compilers for it. Intel by all rights should have died from this, because they are fucking retarded, but the IBM PC was invented around that time and saved their bacon, more or less by IBM accidentally picking their shitty 8086 chip for inside baseball reasons (basically they hated Motorola). The Itanium used the trendy VLIW architecture which never really worked as well as predicted (again group of "experts" failure), and offloaded a lot of work to the compiler. What could go wrong: first time someone explained to me why VLIW was supposed to be great, it was obvious to me it wouldn’t work: “bro, compilers suck.” I guess the Russians have one VLIW chip called Elbrus, but they don’t sell them to outsiders, probably because they suck for anything other than being free of NSA spyware in the chipsets. VLIW does work OK for signal processing, which is why they thought it was a good idea at the time, but signal processing and general computing are only tangentially related: that’s why we didn’t use signal processing chips for general purpose computing. What do people use now? Basically the same thing DEC, Sun and HP were using: RISC. RISC was the right approach all along, with some additions for speculative execution. AMD did the obvious thing and produced a RISCy 64-bit extension to the Intel 32 bit architecture and emulated all the crufty 16/32 bit stuff. This was such a huge success, Intel now uses AMD’s architecture. Retards. All of these companies which literally had the right answer already (64 bit RISC chips) went out of business anyway because they listened to “experts” and MBAs.

Fifth generation computing project. I’ve written about this already, but for the sake of my listacle, it is a great example. For some reason people in the 80s, particularly Japanese people, thought that GUIs + VLSI + parallel computing + SQL + Prolog magic = Star Trek Brain in a Can computard <TM>. The hysteria around this was worse than the present hysteria around LLM “AI.” Governments got involved in an imaginary arms race with the Japanese, starting their own fifth generation computing projects. Tens of billions (of 80s/90s dollars) were spent on these boondoggles; careers ruined, and the whole thing went on for a decade and a half before people realized they were being retarded and had no idea how to build Star Trek Brain in a Can computard <TM>. There were also startups which everyone praising the OpenAI goons or quantum computard startups seem to have forgotten. Lisp Machines still inhabit the fever dreams of Lisp enthusiasts, but the startup was a massive failure, and was definitely part of the fifth generation computing project craze. Thinking Machines was another MIT related fifth generation startup. This one got biggest, fastest, perhaps because it was most explicitly fifth generation; 1000 employees to make a rube goldberg massively parallel computer (the Connection Machine) that literally nobody could make do anything interesting. It should have been obvious: the CPUs were radically interconnected, but lacked enough RAM to do much more than keep track of when some other CPU was asking it to do something. They also could only process 1 bit at a time, serially, which is completely insane. Oh yeah, they also made you program this atrocity in Lisp, which is also insane. Later they tried and failed to compete with Cray by switching to sane CPU architectures. Symbolics lasted the longest because they had an actual product people would pay for: the computer algebra system Macsyma, but they didn’t exactly set the world on fire, and their Lisp machine wannabe 5th generation computer was a giant failure. By the 90s they were basically just a Lisp and Macsyma that runs on their Lisp, and were a nerf ball for various private equity plays. I guess theoretically Cyc still exists, but nobody knows what they’re up to. This one is analogous to LLMs as “AI” and quantum computardism as nobody had the slightest idea how to do any of the intermediate steps between having VLSI and having Star Trek Brain in a Can computard <TM>. It seems unforgivably stupid from present perspectives, but at the time it probably seemed reasonable: Horn clauses were doing neat things comparable to present day LLMs, but were godawful slow and limited from small databases. But it is retarded, just like OpenAI’s approach to “general AI” is retarded.

Underwater Colonization. For some reason in the 50s through the 70s, people thought we might build giant underwater colonies, or at least single family housing. It’s difficult to imagine today why anybody might think this was a good idea, but if we think it through it makes some sense. Up until maybe 1900, Western Civilization had been hugely expansive; finishing off the colonization of the Americas, Siberia and having colonies on all the major continents. 1960 wasn’t so far off from those days and people still had the colonial spirit; pulling up roots and moving to California or Vladivostok or Australia for a better life. Aquariums had become a popular hobby in the 50s and 60s with the advent of electric pumps and ways of shipping live fish; why not live in one? The new technologies of SCUBA had been developed recently, as well as massive offshore drilling platforms. Jacques Cousteau was on the TeeVee showing his undersea explorations. Cousteau even built an undersea house for himself where he enjoyed smoking in the pure oxygen atmosphere. The CIA spread some disinfo about underwater mineral mining to cover up their retrieval of Russian nukes in a sunken submarine; people believed it. They believed it to the extent people still talk about seabed mining; seems like one of the worst ideas imaginable, but at the time it probably seemed inevitable. They even saw underwater colonization as a first step towards colonizing space; after all it’s a lot cheaper to get to the ocean. Old books are filled with the advantages: beautiful undersea views, get away from the overpopulation, you could raise underwater crops to feed the hungry (another nuts idea from the past, though I guess we do it now with groace fish farming), freedom and exploration for everyone! Of course this would never happen on any kind of scale. It’s much cheaper and safer to build concrete boxes with a door than waterproof metal chambers and airlocks. Some people like adventure, but nobody really likes living in a metal tube with no air surrounded by immanent death by suffocation. This didn’t reach the level of craze as something like Fifth generation computing or the Itanium, but the old books and documentaries are filled with breathy predictions that it was going to happen. As far as I know only Cousteau actually tried doing it. Cousteau was really cool dude. That said, it would be really cool if there were underwater colonies of aquanauts. Sort of like it would be really cool if there were space colonies.

Personal Digital Assistants and handwriting recognition. These achieved some market penetration before they got wiped out by phones, and it's kind of peculiar example because of this. I was an enthusiast, owning the excellent HP95,HP100,HP200LX systems which I used as PDA and a kind of mini-laptop with two weeks battery life on AA batteries. I got enormous amounts of work done on these things in grad school in the 90s; I had everything from Derive (mathematica like computer algebra system with a superior UI) to a Fortran compiler, and could upload work to the Cray from it using Kermit or whatever, while it had the ordinary notepad, alarm, contact and todo lists that all PDAs came with. It's still much superior to an ipotato for getting things done (and had a better keyboard than my new T14), and ... battery life of 2 weeks was cool. Nobody but me and some wall street types thought this was good. Everyone else thought little screenpad things with handwriting recognition were the future. Magic Cap, Apple Newton, various Geoworks devices, Palm Pilots. I was an early enough adopter I made a few bucks speculating on these weird little companies on short term trades. They were very heavily traded, so people figured they'd be awfully important. I guess in a way they were, but nobody at the time realized that teenage girls texting each other and sending pictures were the real market for having a computard in your pocket. Absolutely nobody realized that thumb typing on a screen was way more acceptable than memorizing a shorthand handwriting technique your computer could recognize. Probably Magic Cap came closest by realizing carrying around a small computer was useless without fancy GUIs and internet capabilities, but their market failure was epic (and made me a few thousand bucks in grad school, for which I am forever grateful). Smart people all agreed PDAs would be important and would use some kind of handwriting, because people back then were used to writing things in a little notebook. Nobody could foresee the computard part was unimportant: it was the communication and escapism function which people wanted. Nobody could foresee that handwriting actually became an obsolete skill. That's what you get for listening to "smart people" talking about inevitabilities.