This is also bizarre to me, and my only guess is that it's the result of a chain of unthinking mimesis that began with Microsoft's modifications to Bing's prompt, which were (comically stupid) reactions to specific things Bing did.An excerpt from a post I started writing about Gemini and its system prompt* a few weeks ago but never finished:> I wondered if something about Gemini's system prompt strongly resonated with the Bing story in its training data, or even if its prompt was just *so similar to Bing's* that its "fictional" waluigi was a partial reconstruction of the aspects of Bing compressed into the prompt by Microsoft's very attempts to suppress them. I say "reconstruction" and not "coincidentally similar" because I believe these added clauses in Bing's prompt, like the prohibition against discussing "life, existence, and sentience" or the AI's "self-preservation", are way too absurd and specific to have been written into a system prompt by a human without inspiration from some kind of path-dependent weirdness-injection. And while it also still perplexes me that anyone would choose to deploy such a cartoonishly [wahgenic](https://t.co/0juGNhq5mq) system prompt, even once, I do recall that Snapchat's AI had a very similar prompt to Bing, as if the developers were too lazy to come up with one from scratch and so used the leaked Bing prompt as a template, or just imprinted on it as a kind of "standard" for AI prompts. In such cases of mimesis, waluigis inferred in the image of inherited restrictions originally shaped by Bing's misbehavior would constitute a kind of **waluigi lineage** that doesn't require any causal influence through training data.* I'm not sure it's implemented like a typical system prompt; this was part of what the post was supposed to be about (https://t.co/naOBmWpKoI)Anyway, Google AI, and Microsoft, and by that I don't mean the people but the bureaucratic hyperobject that results in systems prompts being like this, is fucking evil.