Comment on Structured outputs on the Claude Developer PlatformparentComments−xpe9moThis makes me wonder if there are cases where one would want the LLM to generate a syntactically invalid response (which could be identified as such) rather than guarantee syntactic validity at the potential cost of semantic accuracy.
Comments
This makes me wonder if there are cases where one would want the LLM to generate a syntactically invalid response (which could be identified as such) rather than guarantee syntactic validity at the potential cost of semantic accuracy.