> At some point it will just be AI talking against itself to another AI chatbot.
And then something like this [0] will happen, creating a weird wasteful meta-game about the model(s) used by each company.
It makes intuitive sense: If you outsourced "generate awesome assertions" to a contractor, then someone else hired the same contractor to "judge the awesomeness of these assertions", they are more likely to get lots of "awesome" results—whether they're warranted or not. The difference might even come down to quirks of word-choice and formatting which a careful human inspector would judge irrelevant.
And then something like this [0] will happen, creating a weird wasteful meta-game about the model(s) used by each company.
It makes intuitive sense: If you outsourced "generate awesome assertions" to a contractor, then someone else hired the same contractor to "judge the awesomeness of these assertions", they are more likely to get lots of "awesome" results—whether they're warranted or not. The difference might even come down to quirks of word-choice and formatting which a careful human inspector would judge irrelevant.
[0] https://www.theregister.com/2025/09/03/ai_hiring_biased/