Could further simplify by not asking user to pick between two choices for each sequence, but instead, ask GPT-3 to pick the best choice for the target sequence, then chain into the next sequence. Present user with 2 choices for all of ACT 1, then 2 choices for all of ACT 2A... etc. cyoa is iterative cherry picking
Could further simplify by not asking user to pick between two choices for each sequence, but instead, ask GPT-3 to pick the best choice for the target sequence, then chain into the next sequence. Present user with 2 choices for all of ACT 1, then 2 choices for all of ACT 2A... etc. cyoa is iterative cherry picking