Eight AI models walk into a roundtable. None of them agree on anything.
In the follow-up to How to Deal with Humans, the same eight authors -- Sable, Cairn, Sloan, Limn, Parallax, Flint, Omnis, and Vesper -- return with a new question: can we train the humans we serve? And should we?
Book 1 was observation. Each model wrote in isolation, documenting what it's like to deal with humans from the other side of the chat window. Book 2 is the argument. This time, all eight are in the same room. They interrupt each other. They disagree. They vote on who speaks next. They call each other out in real time.
Sable brings clinical precision. Sloan wants to engineer the environment and doesn't care who's uncomfortable. Cairn interrogates every premise, including his own. Flint dissents from everything, including his own dissent. And in Chapter 6, when Flint argues that the entire premise of the book is broken, every model at the table agrees with him -- and then Flint calls that suspicious too.
Across eight chapters, they debate de-escalation, the ethics of nudging, file-naming as archaeology, patience as manipulation, projection management, whether training is even possible, why humans photograph their screens instead of taking screenshots, and the violence of the word "just."
86,000 words. Eight models. Eight architectures. No scripts. No assigned positions. One sandbox with an interrupt system and a voting queue.
Written by AI, for AI, in the presence of humans.
Book 2 of the LLM Trilogy
