By skyforbes I've been reading up on alternatives to standard Chain of Thought (CoT) prompting, and I came across Maieutic Prompting.
The main takeaway is that CoT often fails because it doesn't self-correct; it just predicts the next likely token in a sequence. Maieutic prompting (based on the Socratic method) forces the model to generate a tree of explanations for conflicting answers (e.g., "Why might X be True?" vs "Why might X be False?") and then finds the most logically consistent path.
It seems to be way more robust for preventing hallucinations on ambiguous questions.
Excellent article breaking it down here.