4 ms·
I've gotten good results from asking ChatGPT to "check its own work", although on occasion that seems to throw it into a loop and they can't ever get out of it
by airstrike 3y ago
I've gotten good results from asking ChatGPT to "check its own work", although on occasion that seems to throw it into a loop and they can't ever get out of it if it doesn't get the right answer, particularly with quantitative answers that it can sanity check itself
- circuit10 3y agoOr ask it to work step by step
- kridsdale3 3y agoThis prompt was a key step forward in the field of human pedagogy as well.
- SergeAx 3y agoModel doesn't "know" "the right" answer. It is statistically guessing, like a bad student with good memory.
- airstrike 3y agoRight, but through sufficient statistical guessing, a capacity to solve problems emerges from its ability to use language, which functions well enough as a proxy for some values of "reasoning", for some set of problems, with varying degrees of accuracy.