GPT Statement generation

[markwhiting]

We should try electing statements that satisfy particular design points, and elicitations styles, e.g., write a statement from a politician, vs write a statement of something a normal person might say is common sense

• In the wild: Try extracting common sense from content, e.g., given a news article that mentions common sense, can we get GPT to make an accurate sentence about what was implied as being common sense by that use of the term.
• Direct elicitation: Get it to just say a bunch of things it thinks are common sense given a knowledge category.

[markwhiting]

See notes on Watts-Lab/commonsense-platform#87 as a refinement of what we might do here.

[markwhiting]

For in the wild, here's some of the prior data: https://github.com/Watts-Lab/commonsense/tree/master/Statements/Rhetorical