Comment by kqr
10 days ago
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.
10 days ago
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.