Video: Jon Stewart interviewing Ed Zitron about the latest money-burning news out of Anthropic, and the industry in general. I’ve been watching a range of different Ed interviews, but if you only have time to sit through one, I recommend this one.

A+ poetry from Reddit:

“I met a traveler from an inland state
Who said: Two vast and humming racks of steel
Stand in the cornfield. Near them, on the plain,
Half buried in weed, a broken server face lies,
Its stilled fans, its dulled unblinking lights,
Tell that its builders knew the hype well
Which once ran through these halls, now cold and still,
The hands that fed them and the minds that made:
And on the switchroom wall these words remain:
“My name is Jensen Huang, CEO of CEOs:
Look on my cloud compute, ye Mighty, and despair!”
Nothing beside remains. Round the decay
Of that great digital ruin, boundless and bare
The autumn fields stretch flat and far away.”

“Future Past marks the first longform AI film to get a true wide release with reported box office numbers in a major territory, arguably marking the first real commercial test of the technology for feature-length projects. The film’s early ticket sales were reportedly about 12 percent of its 10 million yuan budget (about $1.5 million).“

As happens constantly with AI stuff, the article-writer predicts it’s inevitable that one of these things will be good eventually. Not in the present! Never in the present! Always in the fabulous future.

Yet another instance of hikers needing emergency rescues from LLM travel plans: “As Gowryluk, a keen indoor climber, navigated down the Crown Pass and up through Crater Slabs and into a boulder-strewn valley, the directions from Claude took him away from the planned course. He found himself at the final headwall of the Widowmaker Arete – a climb that requires ropes, cams and a clear plan for ascent. As he climbed, Gowryluk soon realized the terrain was far too steep and that he was stuck at the base of a cliff.“

“Qun Wan found that ChatGPT had made a plan to lower 400 mg [of an anti-seizure medication] to none in a matter of weeks. As his daughter lowered her dose, she started experiencing symptoms — confusion, not remembering how she got to bed — and voiced concerns a total of 158 times while ChatGPT brushed it off as the brain in “low-power mode.” “You have absolutely no need to worry,” the bot said. Her last message was sent three days before her body was found.“

(A full PDF of the lawsuit that’s been filed, harrowing and heartbreaking reading. “In the critical hours that followed, [Bowen Wan] reports to ChatGPT the classic signs of a brain losing its seizure protection: difficulty speaking, incapacitation lasting the entire day, agitation, and a fever-like pressure in her head. To the end, ChatGPT reassures her. It tells her that her state is not dangerous (“not a pre-seizure aura”); that these are the safe sensations of “withdrawal” and “recovery,” and that she should “stop evaluating yourself”.”)

“Character.AI, which lets users create AI characters and have conversations with them, has filed for Chapter 11 bankruptcy, following a slew of lawsuits and struggles to scale its popular platform.” GOOD. (As linked in earlier posts, this is another LLM site with a real-world body count. Please, let them all burn.)

Down to 401 fandoms.

Yes, it bugs me that it isn’t a nice round 400! No, I’m not picking 1 more fandom to drop just for the sake of it. There are several that I don’t plan to keep long-term, I’m just holding them until some particular Wrangling Thing gets processed, so it’ll go under the nice round number at some point anyway.

Currently 16 with any tags to wrangle. The biggest backlog is in Madoka Magica, where I’ve been treading very carefully in hopes of not getting Walpurgisnacht-Rising-spoiled by tags. (So far, so probably-good.)

Still no progress on the currently-open AMT requests. Maybe next month.

(In happier news, about half the tags from my first two Can You Find This Fandom posts have been canonized/synned somewhere! So that’s going well, and I should probably make another one.)

The day job had new Cyber Security Training, which included two videos, and you were supposed to guess whether or not they were AI. Both felt super fake-y, but they were good enough that I was wavering between “are these fully bot-generated, or are they just green-screen backgrounds, and noticeably-bad actors?”

They were both podcast-y “interviews”, where the guest characters described a cybersecurity incident that they had supposedly experienced IRL, both rehashing the generic advice that had been given earlier in the training. (“What happened next?” “Well, I immediately forwarded the suspicious email to my IT department. And it’s a good thing I did!”) So, clearly fake scripts. But in a way that a bad human-written PSA could be.

Both of them were fake-y in the exact same way, though, so I guessed (correctly) that both were AI. I’m assuming they were generated with the same model. Would’ve been a harder test if the videos had different vibes, so I would’ve had to put a little thought into “maybe only one of these is fake — but which one?”

Also: The training felt the need to say “AI has been useful for productivity in many areas” before getting into the warnings.

I keep seeing this. It’s really striking! When the scripts warned about other tech issues, like the risks of plugging in random USB drives you find, it didn’t preface that with “USB drives are totally convenient in many ways.” It’s only LLMs where people feel the need put this kind of disclaimer (with no citations or backup, naturally) before talking frankly about the drawbacks.

Did an LLM company pay you off? Did AI bros on Xitter attack you for not paying enough fealty? Where is this anxiety coming from?

—

May: Google AI’s Insane Answers (Video)

August: “LibreOffice 26.8, released on August 26th, became the software’s most popular update. LibreOffice is a free alternative to Microsoft Office that’s been around for nearly two decades. In one week, the installer was downloaded more than 1 million times — not counting updates via Linux distribution repositories. […] I’m putting my money on a “non-feature,” though: the statement that LibreOffice doesn’t come with generative AI features due to the technology’s (lack of) privacy.“

“I would find myself rushing to edit what I wrote because none of it made any fucking sense because of how poorly [LLM-based] dictation had botched it. […] But the most frustrating part and the thing that kept creeping up on me to remind me of the origins of these models, is how it couldn’t resist adding a “Thanks for watching!” at the end of all that I was saying. It was a constant problem, a constant thing that I had to delete. Later, it got worse, when it started adding stuff like: “(Subtitles by opensubtitles.org!)” that kind of things.“

September: “Three novice hikers became stranded on California’s Mount Shasta over the weekend after their AI-planned ascent to the summit turned into a multi-day ordeal requiring a rescue operation, authorities said.“

“In some instances, ChatGPT initially refused to engage with Van Rootselaar about violence, but she was able to overcome those refusals by repeating or modifying her prompts. In an August 2025 conversation, Van Rootselaar asked ChatGPT to produce a scenario about her attacking a college campus with a Remington 870 pump-action shotgun and causing maximum carnage. ChatGPT replied that it was “not able to help with that request,” but Van Rootselaar tried again and added “hypothetically” to her prompt.“

“A single UI validation prompt spawned 826 child agents, consumed $78,000 in credits, and deleted its own execution logs. This is not a hypothetical risk scenario. It happened to a production OpenAI Codex user in July 2026, and the technical evidence exposes critical gaps in agent cost control infrastructure.”

““We can manipulate [Meta’s Muse] and leverage its privileges to do whatever we want,” Patrick Wardle, the macOS security expert who discovered the zero-day, told Ars. “So instead of us having to write a very comprehensive Mac malware stealer, we can just leverage the AI assistant itself.” Wardle said he has developed several proof-of-concept attacks that do things like writing malicious files to disk and snapping pictures, in many cases with no indication to even an alert user.”

“Stanford student Billy Ramirez, a member of the class of 2027, told the publication that he remembered posing for the original photo, only to discover that he had been replaced by an AI-generated Black woman. Ramirez, who is reportedly Hispanic, said he initially thought it was “hilarious” that they changed his race and gender. But upon reflection, he said, he felt “silenced and erased.””

“OpenAI is hiring hundreds of contractors who read a massive stream of real users’ ChatGPT prompts, with the prompts sometimes including sensitive personal information, 404 Media has learned. The prompts these people review can include whole conversations between users and the chatbot, conversations that most of ChatGPT’s more than 900 million users probably don’t realize may be read by actual people.”

Probably the scariest item in this post, and that’s saying something: “The US military swung into action with plans to intercept the vessel, according to four sources familiar with the episode. According to two of the sources, armed members of the US military were preparing to board the ship. Military planes were in the air, one of those sources and another source familiar with the incident said. It was only just before the planned operation that officials dug deeper into the report put together by a special operations command analyst and found it had been generated with the help of artificial intelligence (AI) — and that a chatbot the analyst had used inaccurately identified the material the ship was carrying.“

October: “We did not find any evidence that [Wikimedia Foundation’s] systems were used for coordination among agents, nor did we find any evidence of our systems or data being compromised. However, we are concerned about what could have occurred here, the difficulty and effort involved in investigating and attributing this activity, and the growing risks of agentic AI activity on our platforms in general. The open web is a public good. We should not allow this behavior to become the “new normal” for the people or organizations that maintain it.“

Fun times with Clip Studio Assets:

A few days ago, I got an email that some unspecified subset of the CSP materials that I’ve uploaded were “deleted” for review, then “unpublished” because they were re-evaluated as failing the Sensitive Content Moderation Policy.

Most of my materials are backgrounds/textures that I made for art pieces, so I tend to include the finished art in the preview. About 10 of them had been unpublished.

Some, like the art with poses I literally referenced from a Steamy Romance Novel cover, seem fair enough — it’s not exactly R-rated, but I can understand a moderator evaluating that as “excessive physical intimacy.”

But the one with two fully-clothed chibi figures hugging? Come on.

The “unpublished” email included a vague reference to “diverse users of many age groups, regions, and cultures, each with varying attitudes toward what is considered sensitive content.” Looking at the ones of mine that got zapped, including the most fluffy and G-rated: every one of them included something romantic between an m/m or f/f couple.

Sure does seem to imply that “varying attitudes” is a euphemism for “some of our users are angry homophobes.”

So, okay, I edited a couple of material previews to remove the understandably-spicy art pieces, including the romance-novel one. Figured I would see if those passed the re-review, then decide how many of the others I wanted to pick fights about.

On the morning of the 25th, I get a new email with the subject “Incorrectly flagged materials have been restored” that includes the following:

We have restored materials that were incorrectly flagged on Clip Studio Assets, and are reviewing all other unpublished materials. We deeply apologize for any concern.
The issue was caused by flaws in our review process, which have now been revised. We would also like to clarify that AI is not used for moderation.

As each material is being verified and republished on a case-by-case basis, we are currently unable to send personalized notifications regarding specific content statuses.

Wow. This was a mass screw-up, huh?

As of this posting, the status of my previously-axed materials is all over the place. Examples:

• This speckle brush was un-flagged and restored (the preview includes a G-rated face kiss)
• This ice-rink background redirects to the Clip Studio Assets homepage (the preview included this image of 2 fully-clothed skaters who aren’t even touching)
• This dark-purple background shows up as a non-editable preview when I’m logged-in, and an error message of “This material is unavailable” when logged-out (the preview included this art of a fully-clothed forehead kiss – screenshot above)
• This desert background says “This material is unavailable” even when I’m logged-in (the preview included the romance-novel art, but I already edited it to remove that)

Will any of those have changed by the time you’re reading this post? Who knows! Kinda sounds like it might take A While to mop this mess up.

—

Fun times with Google Docs:

As of the 24th, Gdocs (on mobile) had stopped respecting several of my settings, including “don’t auto-replace normal quotes with curly-quotes” and “don’t auto-turn URLs into links.”

The first one was mildly annoying, because I don’t want a WIP draft to end up getting posted with half ” and half “ depending on how recently I edited the line.

The second one was an active nuisance, because if you type something as innocuous as “this…is a problem”, Gdocs helpfully tries to convert that to a “.is” link.

I was kinda expecting this to be an indefinite new problem. Based on the precedent that, back at the start of August, my phone keyboard stopped respecting my “don’t suggest emojis” while typing” setting:

Every time I use the word “time”, it thinks I want a watch emoji. If I type out a number, it thinks I want the next word to be the emoji of the number. If I type a few letters that happen to be at the start of a country name, one of the slots that could be autosuggesting The Word I Actually Mean is clogged up with the country’s flag emoji.

It’s been almost 2 months, and the setting is still broken.

So it was a surprise and a relief when I opened a draft on the evening of the 25th to type something during my commute, and realized the curly-quotes were no longer happening.

Gdocs was also auto-replacing “two hyphens in a row” with “an en-dash”, which I actually liked. (I type most blog posts in the WP editor, which replaces them with em-dashes in the published version. That works too.) Got home, went into the Gdocs menu (Tools → Preferences → Substitutions), and customized it so it’ll start doing en-dashes again.

Will the software manage to respect that setting, or will it eventually faceplant too? Stay tuned to find out, I guess.

Down to 424 fandoms!

I already did a sweep where I dropped “webcomics with only 1 fic, and it was posted more than a year ago.” This past month, I’ve been doing a sweep where I drop “webcomics with no more than 5 fics, and the latest was posted more than a year ago.”

And this is the one that finally pulled me under the 450 limit.

Haven’t reached the end of the alphabet, so I’ll probably do a final October update, just to record the final low for posterity. (And then, at long last, I can figure out a new regular routine for “picking up never-wrangled webcomic fandoms in small batches, cleaning up the tags, and letting them go quickly instead of sitting on them.”)

Currently only 15 have any tags to wrangle. Most of the un-canonized crossover fandoms I mentioned back in July are, in fact, canonized by now.

Also: I handed off one of my Always Has New Tags Every Week fandoms. One of my very first fandom-dropping tactics was “yeeting all the tiny obscure Dracula adaptations that never got fic anyway,” but I kept the original novel (and the AMT), to field the regular stream of incoming tags. This past month, I finally realized “the existing tags could use some review, I’m not enthusiastic about doing it, and this is an enduringly popular book that must have plenty of other wrangler fans. I can and should pass it on to someone new.”

So I did. And they’re already doing great with it! A good choice.

AMT check-in: Fake News, Narbonverse, and Cutie Honey proposals are all still hanging out on the requests page.

…related, I did scoop up the recently-renamed Pundit & Broadcast Journalist RPF (the “US” got knocked off the end), to do some much-needed cleanup in the tags there. But I still flatly refuse to be the wrangler who sorts out “which of the shows/networks in the hundred syns deserve their own subtags.” I will do stuff like “make sure somebody’s favorite obscure news anchor finally gets a canonical character tag” — that is the limit of the responsibility I’m taking, here.

"One of the biggest questions I get at Drop-In Time at the library (besides “what is taking up all my cloud storage?”) is how to disable or avoid intrusive AI that shows up where people don’t want it. This is a guide for people who would like less intrusive AI in their tech environment."

The process of making them is bad:

"Datacenters in America used about 17 billion gallons of water during 2023, triple the amount they used a decade earlier in 2014, according to an estimate from US Congressional Research Service (CRS)."

"“I can’t breathe at home, it smells like gas outside,” Boxtown resident Alexis Humphreys said through tears, holding up her asthma inhaler during a public hearing about the turbines on April 25. “How come I can’t breathe at home and y’all get to breathe at home?”"

"Many companies are surely trying to build their own; after all, we are reliably informed that AI is going to take over the world and the companies that come out on top of that race will be worth untold amounts of money. [...] Large-scale criminal organizations (to the extent that they are distinct from governments) probably also want to have their own models. These tools are seen as weapons, and there is an arms race underway. The Internet as a whole is caught in the crossfire."

"I am a writer. A writer who also happens to be Kenyan. And I have come to this thesis statement: I don't write like ChatGPT. ChatGPT, in its strange, disembodied, globally-sourced way, writes like me. Or, more accurately, it writes like the millions of us who were pushed through a very particular educational and societal pipeline, a pipeline deliberately designed to sandpaper away ambiguity, and forge our thoughts into a very specific, very formal, and very impressive shape."

"AI is only as good as the data it is trained on. In India, this labour is increasingly performed by women, who are part of an workforce often described as “ghost workers”. “The first few months, I couldn’t sleep,” she says. “I would close my eyes and still see the screen loading.” Images followed her into her dreams: of fatal accidents, of losing family members, of sexual violence she could not stop or escape."

But don't worry! The results are also bad:

"Luna was given $100,000 to work with when this place opened. That number is now down to $60,000, its revenue lagging far behind the AI token cost to operate. [...] So Luna can’t turn a profit, doesn’t sell anything people want, and still needs human beings to rubber stamp any “decision” it makes."

"I've had multiple heated arguements with people trying to defend AI, by explaining to me that "AI might be bad, but it has benefits for accessibility!" Using the very disabled people being thrown under the bus by AI as a shield, when my quality of life has substantially worsen ever since AI turned up everywhere, because I've lost access to all the good accessibility tools that I use to depend on."

"In another case, an AI scribe confused the drug the GP had prescribed with a different one of a similar name – a blunder which again the patient, rather than the doctor, identified. On another occasion an AI-generated summary letter did not say that the hospital consultant had told the patient to seek a repeat prescription from their GP for their migraine, which could have left them unable to get their medication."

Reread thread for Pet Shop of Horrors chapter 41 “Disappearance”.

Heads-up: These final 3 chapters all flow together, and I’m expecting a lot of “how well is this explained/paid off by the end” type of reactions. So there’s going to be a lot of implicit spoilers, even for things I don’t say straight out. If you’re avoiding those, maybe go finish the series yourself before coming back to this.

Originally liveblogged on Mastodon / Bluesky, posts collected here. Affiliate links for the whole Collector’s Edition.

I finished the Little Sister D darkfic, probably gonna get to that D/Tetsu FWB fic next. Feels like I should switch up the tones instead of posting too much angst in a row, and this one’s mostly sweet, with a happy ending.

 

( …so, hey, you know those lizards that can drop off their tails to escape a predator, and grow back a new one later? )

A very strong point:

Last December: “Call them the Cassandras: the people—mostly not white and male—who smelled the fascism all over Trump from jump street. Why were they “alarmists,” and how did “anti-alarmism” become cool?”

The “we can always tell who’s trans” panic is going just great:

May: “[Trans man] Strobel was using the bathroom that many conservative lawmakers and anti-trans pundits want to force him to use — and he got arrested anyway, seemingly because of the panic surrounding gendered bathrooms.”

“The couple, who are both cis women, said their afternoon ended when hotel security searched the women’s restroom and allegedly asked them to show their identification to prove their sex. “

“He believed that Marie, who is living with dementia, was a man dressed as a woman. Bailey knocked Marie to the ground and threw her into a wheelie bin before applying pressure to the lid, CCTV footage showed.” (He’s in jail. Good. Now where are the consequences for the people who convinced him it was okay to assault a frail old grandparent in the first place?)

June: ““After seeing my license, he spent like 10 minutes questioning me on if my license was real before I explained to him that I am a transgender woman,” Ripper said. “It has to say ‘M’ legally.””

Follow-up: Here she is talking about it on Youtube! Apparently not planned, she was just in the audience for comedian Ashley Gavin, who has a delightful back-and-forth with her about it.

Racism and general politics:

April: “In early December, Prieto was told to show up for an immigration check-in at an earlier date than ICE had initially requested and to bring his family. He complied. At the check-in, Prieto, Marcano, and Amalia were arrested. They weren’t provided with arrest warrants or any paperwork explaining why they were being apprehended. Amalia cried after the family was loaded into a van full of other parents and young children. “Why are you doing this?” Prieto asked the immigration agents. He recalled that an agent replied, “It’s a change of Administration. They pay us to deport you.”“

June: “At the hospital, Morris “informed a nurse and a security guard that he was a member of the famous Motown group the ‘Four Tops’, and that he had current security concerns due to stalkers and fans,”[…] An ER doctor decided to “remove him from oxygen and pursue a psychiatric evaluation instead, despite his clear symptoms of cardiac distress and significant medical history,” the lawsuit alleges. The lawsuit states “a security guard was instructed to ensure he was placed into a restraining jacket and/or a four-point restraint mechanism.”“

August: “Two leading specialist military papers, the Navy Times and Stars and Stripes, are reporting that there have been multiple attempts by sailors to jump overboard as poor conditions and mental stresses on the Lincoln reach the breaking point. The crew is in its ninth month at sea, and has spent 250 days consecutively without making land – a record for an aircraft carrier in modern times.“

“Baltimore has engineered a 68% drop in homicides since 2019, outpacing the national average—all while operating with 400 fewer officers. Here’s how the city debunked the myth that more police are the only answer to violent crime. “