From the rationalists are a net negative for society dept: Scott Alexander’s latest (that I’m not linking) is all about how you should be using the slop machine to tell you who to vote for.

He’s even so kind as to share his prompt:

I’ll be voting in the June 2026 California primary. I’m a centrist liberal abundance YIMBY whose favorite political writers are Kelsey Piper, Matt Yglesias, and Ezra Klein. I’m wary of government overreach, but I’m not a doctrinaire libertarian and want to help people when we can figure ways to do it that work. I’m going to ask you about each race on my ballot, and I’d like for you to list the various candidates’ bios, policies, endorsements, your read on the most important differences between them, and your advice for me as I try to make my choice.

Pretending hallucinations and training data bias aren’t a thing must be making some people’s lives so much easier. While we’re at it, let’s also magical think away any possible dire consequences of giving the handful of ultrawealthy unwell weirdos behind LLMs as a service even more direct political influence.

Also the prompt sample itself is just showoffy¹ nonsense, isn’t it? Even if LLMs were as overcompetent as they’re being hyped there’s no way all that stuff can be deterministically parsed into a concrete set of values that you can check against whatever the LLM digs up from the internet, combined with all the close-enoughs hardcoded in its training data, there’s just enormous room for the chatbot to answer whatever the hell it wants.

1. That’s me trying not to overuse the term “virtue signalling”, but it seems clear siskind is using the prompt to set a sort of partly line for his (outer circle / not completely eugenics pilled) followers. That’s probably also the point of including so much chatbot attributed political slop in the article, ostensibly as data points.