Most of my math academics do that.
They write an issue on the board and ask us to resolve it. Then they ask, “What number of of you bought X? Elevate your fingers.”
“What number of did you get, Y? Elevate your fingers.”
That’s how we realized democracy. Besides, the bulk doesn’t imply the proper reply in math.
Jokes apart, this can be a legitimate technique to determine which reply could possibly be right after we can’t predict with sufficient accuracy which one is right. Not solely in school rooms however it’s practiced nearly all over the place.
AI engineers additionally follow it…
AI generates solutions with chance. Nobody is aware of in the event that they’re right, however the mannequin thinks they could possibly be. If we discover it’s incorrect, we are saying the mannequin hallucinates.
However can we democratically scale back the possibilities of hallucination? — That’s known as self-consistency.
A gaggle of researchers discovered the next of their 2022 research paper.
Sampling a various set of reasoning paths, and choosing essentially the most constant reply (self-consistency) boosts the efficiency of chain-of-thought prompting.
