Let's think step by step in ML Reasoning

[2205.11916] Large Language Models are Zero-Shot Reasoners,
  Takeshi Kojima et al.
https://arxiv.org/abs/2205.11916

If you feed a machine-learning language model "reasoning" questions,
it gets some right and some wrong.  Depending on the model and how
it was "trained." If you give the same question to the model but add
"Let's think step by step", it gets them right.

Apparently the magic phrase depends on the kind of model, and the kinds of
training.  What phrase could we use on humans, to …

ml reasoning