BAML is a language that helps you get structured data from LLMs, with the best DX possible. Works with all languages. Check out the promptfiddle.com playground
Coming from #934; tldr- because scoring for union discriminant selection switches solely on which candidate was least penalized, we can't use "quantity of information preserved" as a metric. Flags is obviously too complex to use as a scoring metric, we do need a scoring metric that is closed under recursive summation, but I suspect we'll run into more awkward situations along these lines.
Coming from #934; tldr- because scoring for union discriminant selection switches solely on which candidate was least penalized, we can't use "quantity of information preserved" as a metric.
Flags
is obviously too complex to use as a scoring metric, we do need a scoring metric that is closed under recursive summation, but I suspect we'll run into more awkward situations along these lines.links:
950 has a minimum repro