Comment by TobTobXX
10 days ago
Wouldn't it make more sense to state the assumptions first? Because then the model has this critical check in its context and can react appropriately. Otherwise, it will just write this step, but what's written before is already written.
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.
Fair point, though I almost never use fast so I'm not sure how much it matters. Can try playing around with the instructions. The main objective was to make me aware of any assumptions made, not necessarily make it behave differently.
It would make the most sense to clarify the assumptions with the user first.