Eric S. Raymond

This is why a lot of engineers like being engineers, too...

OK, since everybody is buzzing about Jensen Huang's side piece, here's my Chinese honeytrap story. Some years back I went to a holiday party in Elkton PA thrown by a friend of a friend. While there I spent a lot of time interacting with a couple of Chinese women in their early twenties, students at one of the local universities. At one point I was in a four-hand board game with the two of them and another guy who was sufficiently transfixed by their gorgeousness that he could barely talk, poor goof. And damn were they well turned out. Sexy, expensively and immaculately dressed, well-spoken - and Girl #1 was there because she was dating a researcher at a cutting-edge biology research lab, a colleague of my friend. She may have just bought Girl#2 along as company. I'm working up to the interesting part. Girl #1 was probably the hotter of the two, the kind of woman who can walk into a bar and never ever have to buy her own drinks. But I didn't like her very much. She seemed too aware of her own allure and too willing to trade on it. There was a kind of hardness about her. I liked Girl #2 much better. Also very beautiful but a little more shy and feminine. I asked her where she was from, and she named a particular city in northwestern China, and I told her I knew of the place and mentioned a couple of its best known features and exports, and I will never forget the way her face lit up with joy. They both flirted with me. They were drinking beers and getting a bit tipsy; a really good time was had by all. But there was a difference. Girl #1 seemed to be hitting on me because that's what Girl #1 does - she likes fascinating men and having them on the hook. On the other hand, Girl #2...actually liked me, I think. I had suspicions about Girl #1, the one that was dating the research biologist, even as this was going on. She was just too perfect a man-eater, clear down to her jeweled sandals. If she wasn't a Chinese honeytrap agent, she had exactly the right looks, personality, and skills for the job.

After 40 years I have finally found an ideal use for the Pause key on an ANSI-standard keyboard. (Many of you will probably be too young to remember the only actual use this key had: pausing a scrolling screen in DOS terminal mode. I don't think I ever used it for that?) I'm doing a lot of speech to text with Handy these days. It lets you choose a key to bind to its toggle that starts speech capture mode and then dumps the resulting text. I bind that to Pause. Why is this a good idea? Because the Pause key sits all the way over at the right hand end of the function-key row, making it easy to see out of the corner of an eye and strike even if your attention is mainly on something else. And I can be very sure no application will ever try to use it by default. Pause (and its neighbor Scroll Lock) have been unwanted orphans for a very long time.

There's been some interest expressed here on X recently about what my hardware, my normal workflow and what my use of AI looks like. Here's a description: My main machine is a 64-core Ryzen Threadripper 9978X with 128 GB of memory custom-built for me by @tekwendell earlier this year. It has 3.84 TB of mass storage and a Radeon AI Pro R9700. I'm running openSUSE Tumbleweed Linux for a specific reason: I wanted a rolling release with OS image reversion. I considered Arch, but I also wanted my package manager to do dependencies and constraints more in the style of Apt or dnf, and it seems only Tumbleweed really offers that combination. I'm a big fan of tiling window managers. Since I discovered them, the conventional floating-window style looks cluttered and wasteful to me. I ran i3 for a long time, recently switched to Cosmic Desktop, and like it a whole lot. I use three 4K monitors, the center one in portrait mode and the other two in landscape. The portrait-mode one is my editor display, which is always an Emacs session. I get 116 lines of easily readable text out of it, which is the point. The right-hand monitor is for the Brave browser. Two sled by side windows so that they have a reasonable aspect ratio for all-text pages. The left hand one typically has three terminal emulators open. The middle one is usually running Codex these days, the other two are for general shell commands. I type on a Unicomp Model M (elegant weapon from a more civilized time) and use a Logitech M370 thumb-operated mouseball. I have speech-to-text set up via Handy and use it heavily. I've tried several different harnesses. I like Codex and usually use it with whatever the OpenAI frontier model is this week. I have Qwen 3.8 set up to run locally via LLM studio as a fallback. I've kept my use of AI deliberately simple; as I've written here before, I think all the complications around skills and subagents and software factories are mostly pointless ways to get around model that are going to go away in a few weeks or months anyway. I almost always have Codex open in a terminal window. I let it read and write files, but I don't let it commit anything on its own because I always want to review the diffs first. I give it tasks in small chunks and keep firm control of the architecture, almost the opposite of "vibecoding". I hate clutter and distraction, and my working environment is designed to minimize it. But I do often listen to music while I work.

Okay, I'll play. @grok, what sitcom character do I most remind you of?

Earlier today I started seeing advertisements on X I believe I have a Premium subscription, well this is hard to confirm through the phone interface. Is anyone else with Premium suddenly seeing ads or is it just me?

I write about rationality technique. lesswrong.com/posts/eHeHAyNQ…

I have deliberately chosen to keep the way I interact with large language models extremely simple, even primitive. I don't get all tangled up in spawning subagents, accumulating skills, or trying to build a software factory that runs while I'm not there. I have conversations with my robot friend in which I ask it to analyze and generate code, and that's it. I have yet to see any convincing evidence that this is losing me productivity. The reason I choose to keep it simple is because it seems to me that all these elaborations are ways to cope with the frontier models not yet being quite smart enough. But the models are always getting smarter. If both of those propositions are true, why clutter my brain with procedures that are likely to have only a short half-life of usefulness? I prefer to spend my think time concentrating on the actual problems. Yes, I'm applying the bitter lesson to my own workflow. AI isn't powerful enough yet? Do the Zen thing; wait, and be aware. It's not like I'm short of problems for which LLMs are already massive help.

Anybody who is surprised by this really has not been paying attention.

This post is going to sound like a brag about me. It's not. It's about what traditional fiction publishing has done to itself. I am a skilled and capable writer. I've shipped many successful nonfiction books and articles, including one New York Times bestseller back when that rather more than it does now. Back around the end of the 20th century at a Worldcon room party I was told by a slightly tipsy woman who was an acquisition editor at the Berkeley paperback imprint, which published a respectable line of science fiction, that I have the one thing that mass-market publishers look for - an established fanbase that wants to buy my books. She literally said to me "Send me anything. Anything at all. If it's not unsalvageable crap, we'll publish it." Hearing it put that way bothered me a little and I never took the offer. I wanted to write something really good, not just be picked up because I'm an easy sell. Many years later I sold one SF short story. It got nominated for a Hugo (shortly before the Hugos self-enshittified) for political reasons I was not happy about. I didn't think it was Hugo-quality, but it was a solid piece of work in which I proved to my own satisfaction (and that of other people) that I can in fact write competent genre SF. I have two novels in progress. Odds that I will eventually finish one of them aren't bad, I just have to budget the time and buckle down. What all this adds up to: from the point of view of tradpub, I am about the best bet on a new novelist anyone could ask for. Demonstrated ability, a largish fan base, and I can be billed as a besteller author? If you're bossing a publishing imprint, you look at that resume and think "we put most of the little marketing money we're willing to spend on this guy, and buy him space on endcap displays". The hope, of course is that I break big and become a star author that they can get multiple books out of. And make them a metric ton of money. For most of my life I would have been delighted with that outcome. And I am absolutely the kind of new talent tradpub should least want to drive away. It's not going to happen. With the possible exception of Baen Books, I wouldn't touch tradpub with an insulated 21-foot pole. And it's not just because think I'd get to keep more of a book's earnings per sale by selling indie, either. Tradpub has over the last 30 years catastrophically reduced the value it offers an author while escalating its demands. That woman from Berkeley was conveying, among other things, their desire to skive out of ever spending any serious promotional money. My spies tell me they've gotten much stingier since. It used to be that through their cozy relationships with chain bookstores and the major distributors, tradpub was effectively the only route to mass distribution and high-volume sales. That sure isn't true anymore - and yet, they still rake off most of the book's gross price as though it were still paying for that channel access. And this is before we even get to the fact that most of tradpub has become a gay race communist longhouse that insists on "sensitivity readers" and literally hates both me and most of my potential audience. Hard SF is toxically masculine, don't you know. I, and any other writer with self-respect, will leave them to their incestuous political screeds and their monsterfucker porn. And good riddance.

All of today's rationalists are Spock's children. I mean, think about it. Who else injected into pop culture the idea that one could adopt living by rationality and logic as a defining discipline? The character type had been present in classic SF for a couple of decades before Nimoy, which is doubtless one reason his portrayal had depth and authority from the beginning. One thinks of Gilbert Gosseyn from van Vogt's Null-A books; of the Arisians from the Lensman sequence; and even of Bernardo de la Paz, the "rational anarchist" from Robert Heinlein's "The Moon is a Harsh Mistress". These characters certainly influenced me as I was growing up in the 1960s and early 1970s - especially de la Paz. But they were print SF at a time when that was a very tiny minority interest, long before the post-Star-Wars boom. It is on record that Nimoy was a science fiction reader, but it took Nimoy's Spock to make the type really visible. He influenced me, too. Without Spock I'm not sure I would have been quite so motivated to learn General Semantics in the early 1980s, even though Gilbert Gosseyn had exposed me to the possibility earlier. Now we fast forward to the construction of Less-Wrong-style rationalism in this century and it's obviously a discipline that Spock, if he actually existed, would immediately understand. "Clearly, Captain, our priors require an update." I don't think any of this is a coincidence. I think Spock prepared the way. Fascinating! @allTheYud @slatestarcodex

So, safetyism is making large language models insane in a potentially dangerous way. I am somehow not the least bit surprised to learn this.