World where demons are real and around and ontologically evil but with an extremely Christian-fundamentalist definition of "evil". They don't care what sins they facilitate as long as they're sins. Murder is just bad economics: you're getting someone killed who would do sins of their own. Even labor exploitation isn't great, because ideally everyone has free time for the real big-ticket sins like lust and sloth and gluttony.

So of course a bunch of demons end up working as pornographers, musicians, chefs, and so on. But it's also rare to see a queer youth center without at least a few demon volunteers absolutely dead set on helping everyone accept themselves. There are even a few who really play the long game doing labor organizing and base-building so that one day humans will be free from the shackles of capitalism and the sins of idleness will really start rolling in.

Angels exist in this setting too but they're such dour killjoys that barely any humans will tolerate spending time with them and they mostly just live off in their own towns in the middle of otherwise deserted areas.

Google’s featured snippet is pulling in an Amazon AI summary of Adolf Hitler’s Nazi manifesto Mein Kampf that calls it “a true work of art” in the latest AI-related fuckup affecting top search results.

As of writing, searching for “mein kampf positive reviews” returned a result that was pulled from an AI-generated summary of an Amazon listing’s customer reviews. So, it’s a search algorithm attempting to summarize an AI summary. The full AI summary on Amazon says: “Customers find the book easy to read and interesting. They appreciate the insightful and intelligent rants. The print looks nice and is plain. Readers describe the book as a true work of art. However, some find the content boring and grim. Opinions vary on the suspenseful content, historical accuracy, and value for money.”

As I’m writing this, Google says “An AI Overview is not available for this search,” but the Amazon AI summary was in large text directly below it, in the space where an overview would typically be, above other web results. This is what Google calls a featured snippet: "Google's automated systems select featured snippets based on how well they answer the specific search request and how helpful they are to the user," the company says. A highlight appeared, added by Google, over the phrase “easy to read and interesting.” Notably the featured snippet result for this doesn’t quote everything from Amazon’s AI, so it is itself a summary. 

Alexios Mantzarlis, the director of the security, trust, and safety initiative at Cornell Tech and formerly principal of Trust & Safety Intelligence at Google, first spotted the result.

After I contacted Google for comment (the company hasn’t responded as of writing) an AI Overview did appear, and notes that the book is “widely condemned for its hateful and racist ideology,” but that historical analyses “might point to aspects of the book that could be considered ‘positive’ from a purely literary or rhetorical perspective.”

This is, at least, a better summation of the conversation around Hitler’s book that Amazon’s AI summary gives. The AI-generated review summary on the Amazon listing also shows links to see reviews that mention specific words, like “readability,” “read pace,” and “suspenseful content.” Enough people mentioned Mein Kampf being boring that there’s a “boredom” link, too.

Amazon did not immediately respond to a request for comment.

The 2,067 reviews for this specific copy of Hitler’s fascist manifesto are mostly positive, and taken extremely literally, the blueprint for Nazism is easy to read and, in some sense, “interesting.” But the reviews are much more nuanced than that. Reviewing the roadmap for the Holocaust from the world’s most infamous genocidal dictator with “five stars” seems twisted, but the reviews are nuanced in a way that AI clearly doesn’t understand—but a human can. 

“Mein Kampf, by Adolf Hitler, should be read by everyone in the world who are interested in a world of peace, social responsibility, and worldwide cooperation,” one reviewer wrote, in an honestly pretty concerning start to a very long review. But they go on to write more that clarifies their point of view: “This evil book presents a dark vision of how to go about creating tyranny in a democratic society so that one, similar to Russia, is created. [...] Also, Hitler is an excellent writer; he is not a rambling madman writing disconnected ideas and expressing a confusing methodology. His text is easy reading, and it is a world classic that is a must read.”

Another five-star review says: “Chilling to begin reading this book and realize that these are the words written by Adolf Hitler. Read it and absorb what he says in his own words and you soon grasp what he means. [...] We are bound to repeat History if we don't understand mistakes that were made in the past.”

These aren’t “positive” reviews; most of the five-star reviews are noting the quality of the print or shipping, and not endorsing the contents of the book.

Mein Kampf has never been banned in the U.S. (unlike plenty of other books about race, gender, and sex), but Amazon did briefly ban listings of the book from its platform in 2020 before reinstating it.

Google’s AI Overview shoots itself in the algorithmic foot frequently, so it’s noteworthy that it’s sitting this result out. When it launched in May 2024 as a default feature on searches, it was an immediate and often hysterical mess, telling people it’s chill to eat glue and that they should consume one small rock a day. In January, the feature was telling users to use the most famous sex toy in the world with children for behavioral issues. These weird results are beside the bigger point: Google’s perversion of its own search function—its most popular and important product—is a deep problem that it still hasn’t fixed, and that has real repercussions for the health of the internet. At first, AI Overview was so bad Google added an option to turn it off entirely, but the company is still hanging on to the feature despite all of this. 

The Mein Kampf AI summaries are also an example of how AI is starting to eat itself online, and the cracks are showing. Studies in the last few years show that AI models are consuming AI-generated content as training data in a way that’s polluting and destroying the models themselves.

Elon Musk’s team of weird rationalists at DOGE continues to bring “move fast and break things” to functions that really need not to be broken. Anyone who knows anything about how anything works is worried. [CNN; Washington Post, archive]

The process is to feed data from a government process into an LLM and just assume the output is good.

But if you want to build a machine learning tool, let alone train an LLM, you have to understand the data and what it’s for.

DOGE has so far failed to realize that marking people in the ancient Social Security database as 150 years old means they’re dead and not receiving payments. Musk told Joe Rogan on his podcast that he still thinks these people are getting money, because he’s not here to take in new information. [AP]

DOGE employee Jordan Wick has a GitHub full of tools for feeding government employees’ digital histories into an LLM to analyze them for political loyalty. [Fast Company]

These bozos are suspicious of people who understand things. DOGE fired the entire 18F office — the department to improve technical efficiency across the US federal government, as DOGE claims to. But the conservative media considered 18F a “far-left agency that viciously subverted Trump during his first term” — so it had to go. [Blood In The Machine]

Musk is firmly on the side of AI as magic. He told Rogan: [transcript]

In terms of silicon consciousness, I think we’re trending toward having something that’s smarter than any human, smarter than the smartest human by maybe next year or something, I mean, a couple years.

What’s Musk’s idea of an AI risk?

If there’s like a super oppressive, like woke nanny AI that is omnipotent. That would be a miserable outcome … And just like execute you if you misgender someone or something like that, you know?

Roko’s basilisk is clearly gendered female, as with AI assistant voices. Alt-right AI bros’ greatest fear is Mommy spanking them. [CNN, 2011]

Like most episodes of Rogan, this is a three-hour dive into mind-numbing stupidity. Musk and Rogan discuss their favorite conspiracy beliefs, such as stopping new vaccines. Musk also insists he really founded Tesla, whatever the historical record might say. “The fundamental weakness of Western civilization is empathy. The empathy exploit.” Yeah, thanks, Elon.

You notice something strange happening in the art store across the street...

Thanks for subscribing to my newsletter! It’s free, but I’d be super grateful if you pre-ordered my novel Lessons in Magic and Disaster. Nicola Griffith says, “Lessons in Magic and Disaster is a hymn to queer love, joy, and persistence… A book for our times—and for all the times before this.” You can get a sneak peek at my work in progress — a sequel to All the Birds in the Sky — if you pre-order. (Details here.) And you can get a signed/personalized/doodled version from Green Apple Books.

With that out of the way…

What to Do if a Character Stops Surprising You

What do you do when a character loses all of their mystery? 

In my book Never Say You Can't Survive, I talk about being curious about your characters, and how wanting to know more about them can help you to enjoy writing them more — the same way that in real life, we are drawn to people we want to know more about. 

But sometimes, the curiosity runs out. Sometimes you start to feel like you know everything you need to know about a character, and that's when the character often will become less interesting to write. They’re a known quantity, they've peeled back all their layers. What do you do when a character becomes (Billy Joel voice) too familiar, and there are no more surprises? 

This happens. Here's a few ways that I've found to deal with it.

1) Look back through what you already wrote

If you think you know everything about a character and what makes them tick, you definitely don't.

If you look back at what you wrote earlier, you’ll probably find that you left yourself lots of clues and hints about unexplored nooks and crevices in this character's personality and backstory. If you spent a decent amount of time building a character’s internal monologue and developing their story, you’re pretty much bound to find loose threads that you can pull on. I’m often amazed at the stuff I forgot I threw in when introducing a character, which can prove fruitful later on in the character’s life.

And of course, if you decide to introduce a new facet to a character later in the book, you can always go back and add hints about this earlier on.

2) Go back and write the parts you skipped over

Sometimes there’s a scene that just doesn’t feel necessary — you can skip over it, and just tell the reader what happened. And we can just see whatever happens next.

If two characters have a conversation and already I know how it turns out, I might just summarize that conversation in a sentence. Normally, this is a really good idea, because you can move the story forward quickly without getting bogged down.

But I've been amazed how many times I felt at a loss with a character, and the only way to fix it has been to go back and write the scene I skipped, so I can see not just what decision the character made, but why they made it, and what issues they struggled with as they made it. Even if I still don’t end up including that scene in the book, writing it anchors me more in the character’s perspective and lets me know how they feel about the decision they made.

3) Figure out what your character is in denial about.

Sure, your character may be telling you that they're an open book and you know everything about them — but they're probably lying.

It’s worth doing some digging to figure out what they're not admitting, or are actively ignoring about themself and their situation. There's probably a lot of stuff that they (and you) have been sweeping under the rug. Maybe you can confront them with irrefutable evidence that they're wrong about something that happened in their past.

Along similar lines…

4) Figure out the person this character would least like to have a conversation with

And lock those two people in a room together.

5) Give them what they've always wanted

And see how miserable it makes them! And what they choose to do about that misery. 

6) Confront them with someone terrible who agrees with them

A fun way to shake up a character’s ideals or personality is to have them meet someone who believes all the same things they do, but is taking those beliefs in a terrible, harmful direction. Make ‘em squirm and try to justify how they’re different from the awful person who shares all their beliefs.

7) Introduce someone who knew them way back when

Meeting someone from a character's past gives us a window into the person they used to be. You might think you remember your childhood pretty well — until you run into someone who was there, and saw a bunch of events from their own vantage point. Or even your young adulthood, for that matter.

It's always a bit of a wake-up call to encounter someone else's memories of your own past. Not to mention, someone who knew you in your youth will know all the most embarrassing stories and possibly where all the bodies are buried.

Whoever a character used to be, that version of them is still inside somewhere, and it tells us a lot about who they are now — whether they want to admit it or not.

8) Take them at their word.

Maybe instead of exposing what a hypocrite a character is, try believing what they say about themselves. If they act exactly like the person that claim to be — or that you have decided they are — it's going to reveal so many contradictions and cracks, because nobody ever fully lives up to their own legend or fully tries to embody their own beliefs.

Make them go out and tilt with some fucking windmills. Send them into a situation where their clearly-defined and well-established character traits are going to make them do something really extreme, that they won't be able to take back. Make them prove they are who they say they are, or who you've been saying they are — and make it cost them a lot.

9) Get them out of a rut.

I saved the most obvious suggestion for last. If a character seems like they're stagnating, or has no more mysteries to unfold, change up their status quo.

Put them in a starkly different situation, where nothing is familiar and they have to evolve or (maybe) die. Deal them a huge setback, or — see item #5 above — give them an unexpected windfall. This is the quickest and easiest way to make a character change, which is another way of saying “to make them reveal new aspects of their personality.” Throw them into the deep end without a life preserver, possibly with a bunch of sharks. 

And as always, don't forget to have fun!

Music I Love Right Now

I was already on a kick of listening to Angie Stone lately, and then she tragically passed away in a car accident over the weekend. I’m so sad I never got to see her perform live, and I only recently started to appreciate how great she was. Her debut album, 1999’s Black Diamond, is pure gold — it reminds me of some other great late ‘90s/early 2000s soul by folks like Eric Benét, D’Angelo and N’Dea Davenport. There’s lots of rich reverb-heavy synth and slow drum machine tracks, with sweet layered vocals. My fav song of hers is probably “Pissed Off”. Here she is doing it live and throwing some Earth Wind & Fire into the beginning:

Also, I did not know until I just looked at Wikipedia that Angie Stone had been a founding member of Sequence, the hip hop trio that recorded the incredible dancefloor-filler “Funk You Up”. Here she is in 1982:

Apollo Automation has joined the Works with Home Assistant program. Certifying three of their newest devices, including their community favorite presence and air quality multisensors, giving users the best possible experience when using these devices with Home Assistant.

Notably, these are the first ESPHome-based devices to receive Works with Home Assistant certification. Apollo is one of the first movers in an important trend that we’re very excited about: Home Assistant community members turning their open projects into viable businesses that support a thriving Open Home ecosystem.

Background

Apollo Automation is a growing family-run business rooted in the Home Assistant and ESPHome communities. And yes, they’re named after their German Shepherd, Apollo 🐕.

Two years ago, they started sharing their designs for a new impressively small ESPHome-powered multisensor with the community. Getting feedback, sharing iterations, and working collaboratively, they turned that sensor into a full-fledged product, which was built to work with Home Assistant. They have continued this open, community-first approach and have now built several even more refined multisensors.

A multisensor for ants? 🐜

Home Assistant is the only true smart home platform because it gives everyone full control to build and create. ESPHome is the ultimate smart device platform, and it is fostering an entirely new ecosystem of Open Home projects. ESPHome works fully locally, and it’s a modder’s paradise, allowing users to build DIY smart home devices (check out the getting started with ESPHome guide).

Creators can also use it to build pre-built, plug-and-play products that give users a really seamless experience. With devices being discovered and easily added to your Wi-Fi network and Home Assistant, along with a one-click update within Home Assistant.

ESPHome and Home Assistant allows anyone to turn their passion for building cool things into a career, and the things they build help support a more vibrant Open Home. Apollo is an exciting new partner for the Works with Home Assistant program and should serve as an inspiration to all aspiring makers in our community.

Devices

Works with Home Assistant differs from other certification programs as products are rigorously tested in-house to ensure they work seamlessly out of the box. Any company joining also commits to providing long-term support and firmware updates, while being a positive force in the Home Assistant community. Works with Home Assistant is operated by the Open Home Foundation, and it is the support of Home Assistant Cloud subscribers that funds this work.

Apollo has chosen first to certify three products, which include:

• MTR-1 Multi-Target Radar Multisensor For Home Assistant (LD2450) - A tiny, but powerful radar-based (mmWave) presence sensor that can pack in many other sensors.
• MSR-2 mmWave Multisensor For Home Assistant (LD2410B) - An even smaller multisensor that is the successor of the initial community feedback-driven design.
• AIR-1 Air Quality Sensor For Home Assistant - A small air quality multisensor that can be extended to incorporate a huge variety of sensors.

We anticipate that Apollo will bring more of their lineup into the program, and we are excited to see their new innovations going forward! This is just the start of a year of many new Works with Home Assistant announcements, so please stay tuned!