alt Hacker NewsWorse! It's trained to output coherent reasoning, so by putting the assumption last there's a risk it massages the assumption slightly to fit the conclusions it has already drawn.
Worse! It's trained to output coherent reasoning, so by putting the assumption last there's a risk it massages the assumption slightly to fit the conclusions it has already drawn.
That's a real danger, yes.
If it's the reasoning kind, then it'll run through one iteration in the background before it composes its emissions for the meatbag.