What assets? Like, yes they could acquire the database of user prompts and chats, but this is like saying that I could acquire a gently-used Playboy from the woods. And I presume that, just like with ABC and ESPN, Disney will start looking for synergy, but Character doesn’t have any character copyrights which can be extracted into theme-park attractions, nor would Character’s users appreciate a Disney-fication of their UX, nor is Character’s core service profitable on its own with no changes.

Oh, to have a second upvote. Literally the only two modes of interaction I’ve had with Torvalds, across multiple kernel summits, are either (1) he ignores me, maybe kind of grunting, while I’m trying to say something like “good morning” or “how are you?”, or (2) yelling at the person standing next to me, who I’m trying to collaborate with, and thus yelling at me by proxy. I hope it’s no wonder that I only hack kernel in private, for patches I need, or for close friends.

I haven’t really kept up, but it looks like Character’s business model was still focused on NSFW content. If so, then this is also a disaster for the industry at large, since NSFW content is one of the few genuine money-printers for chatbot and image-generation services; people really do want thousands of iterations of the same fantasy, slightly changed in order to stay engaging, and Character was highly recommended as a source of affordable slop. (Examples of unaffordability include the rampant free-tier abuse on ChatGPT, Gemini, and Grok; I’m assuming that customers are willing to open their wallets at least a little. In general, the porn industry turns a profit, so I do think that people will pay for porn.)

My prior estimate (previously, on Awful and on Lobsters) was that token generation was subsidized to the tune of an effective 80-90% discount; this looks to be more in the 90-95% range. For contrast, consider chatbot-booster Willison’s belief, in reply above, that Chinese labs will magically produce a five-fold improvement in token production; this still would be between two to four times slower than needed to turn a profit for Character.

For copyright purposes, all patches must be substantially human-written. However, it seems that most patches are now prepared with chatbot assistance, so that effectively means that people are required to be psychotic centaurs in order to keep up with review demand. There’s an LWN secret link from the documentation lieutenant which summarizes the linked LKML thread. My former lieutenant, in the Direct Rendering Manager (DRM) subsystem, is apparently having a bad time:

Airlie asked his sub-maintainers; most of them identified him as the biggest pain point. More seriously, it was that the DRM subsystem is still seeing too many regressions. There is a push within the subsystem to move development to GitLab in order to use its CI features; he is not standing in the way because he wants to see how it turns out.

This is a more systemic problem than one might expect, because DRM (and DRI, Direct Rendering Infrastructure) are partially shared between Linux and the BSDs. This subsystem has creaked along under the weight of corporate expectations for decades, mostly fueled by a kind of quiet sponsorship of young university folks from GPU vendors along with outsourced drivers produced behind corporate privacy and thrown over the wall. Every vendor, even our eternal nemesis nVidia, has made repeated concessions in order to try to figure out the maintainer burden, but it seems that we are still dependent on Airlie’s stoutness and willingness to get shouted at by Torvalds.

Previously, on Awful, Gutsick Gibbon spent an hour ripping apart T. Beale’s (VD’s) anti-evolution book. She was still angry though, so she got her prior collaborator and four friends to join her for a complete dissection of VD’s bullshit claims. Content warning for literally everything awful we discuss here, including sex crimes, neo-Nazis, Silicon Valley bullshit, and remarkably bad arithmetic. My choice sneer is her chosen video title, “Six Scientists Roast One Racist Loser (Heartwarming) (Four Hours) (Hopecore)”. GG’s requested that we not give VD any attention other than reminding him that he’s such a pathetic loser that even white nationalists disavow him as too boring and insipid, so I’m not posting this to a top-level topic, but I’m getting a lot of household chores done while listening to it in the background and maybe you will too.

While doing research on our friends, I found (a disowned Reddit comment which found) another banger from the guy who pointed out that pinball is too chaotic for superintelligences to predict (discussed previously, on Awful): whole-bird emulation requires quantum mechanics, or “WBE requires QM” for short. Apparently WBE was a minor meme inside the community at some point, but I’m not sure who first noticed the situation. In short, birds with certain chemicals in their eyes may have the ability to sense magnetic fields due to QM effects. (I’d normally recommend WP on this, but the relevant article is bizarrely skeptical about the situation, although it does include a decent diagram.) As such, our friends consider the possibility that their vaunted superintelligence tasks (ancestor simulations, Newcomb predictors, etc.) are harder than WBE, which itself is much harder than e.g. cracking classical cryptography with Shor’s algorithm or some other not-yet-reached quantum-computing task. The author left a great sneer in the comments:

Adding a compass is unlikely to also make the bird disoriented when exposed to a weak magnetic field which oscillates at the right frequency. Which means that the emulated bird will not behave like the real bird in this scenario. You could add this phenomenon in by hand. Attach some detector to your compass and have it turn off the compass when these fields are measured. More generally, adding in these features ad hoc will likely work for the things that you know about ahead of time, but is very unlikely to work like the bird outside of its training distribution. If you have a model of the bird that includes the relevant physics for this phenomenon, it is much more likely to work outside of its training distribution.

Ro-ro-rotate your model of the bird, rotate your model of the bird for science!

Famous mathematician Terry Tao has published slides for “Math 2.0”, a talk where he discusses how mathematics and chatbots can integrate. I suspect that Dr. Tao has fucking lost it. I don’t know how far the chatbots have misled him vs. how much of this is his unawareness of non-number-theoretic maths, but his “ablation studies” and “neighborhood of a theorem” are already studied in reverse maths, and chatbots have not lowered the complexity of proof search, so there’s not really a sense in which chatbots replace any of the proof-oriented parts of practicing mathematics, regardless of the quality of what they say.

Related: a doctor who denied COVID-19 vaccine efficacy (and thus should not be trusted) has chimed in to say that Dr. Tao doesn’t understand cancer research; in particular, he doesn’t understand that (1) we already throw a bunch of chemicals at the wall to see what’s effective, and (2) terminal cancer patients rarely hesitate to sign up for human-challenge trials and experimental pretrials, on account of having decreasingly little to lose. The orange-site commentariat have several similar complaints about Dr. Tao’s position.

Today I learned that Gwern has a folder of Nano Banana comics. Moreover, on HN, he’s proud of them. Masterful gambit, sir.

Ever wondered what kind of workplace environment Elon cultivates? Reading coverage of Tesla’s trial, where complaints were brought by Black folks who felt harassed and discriminated against, gives us an inkling.

CRD attorney Juan Gamboa asked [Judge] Borkon to bar Tesla’s counsel from contacting anyone on the agency’s witness list. He told the court the CRD had tried to reach at least four other listed witnesses over the past several days without getting a response, and that they’ve fallen “off our radar.” The agency believes Tesla’s lawyers got to them too, either with payments not to testify or by talking them out of it. Gamboa wants to stop Tesla from offering to represent witnesses “through persuasion and payment.”

You’re going to hear a lot of coverage of this trial, I expect, due to the following antics from Elon’s “hardcore” attorneys. For context, the N-word is an anti-Black slur which is, in the USA, completely unacceptable to use in court or the workplace.

Tesla’s outside counsel told the judge the N-word is a “term of respect and endearment” in the Black community. … “This trial is not about the N-word — the N-word is not on trial here,” Jones said.

Huh, so what’s actually on trial?

Edwards said a non-Black coworker called him the N-word, cursed at him, and threatened to beat him up. He reported it right away. A month later, he found out his managers had written him up over the incident, and he was later passed over for promotions after applying to more than 100 jobs.

Oh, that’s pretty awful. That’s depressingly common in blue-collar factory employment, though. I wonder if there’s anything unusual about Tesla’s internal culture when it comes to race relations?

The CRD’s Brett Watson told the court the evidence will show managers calling Tesla “the plantation,” with some referring to themselves and Elon Musk as “slave masters,” and graffiti including swastikas, “KKK,” and nooses left on factory walls for days. The second witness, former regional security manager Ozell Murray, testified that his supervisor, Ray Sethna, now Tesla’s senior director of global security, told him to warn a Black Marine veteran he was hiring: “just be sure [the veteran] knows he’ll be called those words.”

I was at a loss for words for a few minutes after this. Dr. Seuss and Jim Henson together could not create such a caricature of a goose-stepping bigoted incompetent business-muppet as Elon.

Started a new file in my notes: what are some ten-words-or-less domain-specific questions that completely, totally, hilariously stump the chatbots? Everything here was tested with whatever DDG’s currently wrapping, both in knowledge panels and full chats, and the responses were pathetically wrong or uninformed. My thesis is that, with such short prompts, the user is doomed to receive a milquetoast average response; the bot correctly identifies the specific domain but elaborates a global non-specific approach that isn’t sufficiently nuanced.

Gwern’s been updating those comments! This was in 2023, and in 2025 he was still so mad about it that he wrote a list of ways to cheat at pinball and edited the comment to add a link.

I’ve done some of the numbers here, but don’t stand by them enough to share. I do estimate that products like Cursor or Claude are being sold at roughly an 80-90% discount compared to what’s sustainable, which is roughly in line with what Zitron has been saying, but it’s not precise enough for serious predictions.

Your last paragraph makes me think. We often idealize blockchains with VMs, e.g. Ethereum, as a global distributed computer, if the computer were an old Raspberry Pi. But it is Byzantine distributed; the (IMO excessive) cost goes towards establishing a useful property. If I pick another old computer with a useful property, like a radiation-hardened chipset comparable to a Gamecube or G3 Mac, then we have a spectrum of computers to think about. One end of the spectrum is fast, one end is cheap, one end is Byzantine, one end is rad-hardened, etc. Even GPUs are part of this; they’re not that fast, but can act in parallel over very wide data. In remarkably stark contrast, the cost of Transformers on GPUs doesn’t actually go towards any useful property! Anything Transformers can do, a cheaper more specialized algorithm could have also done.