Only does depth first search #206
Description
The prompting strategy does depth first search which forces the models in to a space where they hit a local minima but can't generalize to find the global minima of the problem.
Doing depth to some Nth step, then fully rewinding using the knowledge of the prior minima would allow the model to keep probing at new local minimas until it can generalize to find a better possible global minima.
It gives the model a more informed understanding of the problem / solution space via data collected going down various depths.