Embed Notice

@kaia I appreciate that, thank you. Your third paragraph there is worrisome to me -- 'dialectic approach' and 'time [...] is minimized' are severe red flags.

I would strongly suggest in cases where the dialectic/time pieces feel good when you get an LLM answer to immediately contradict the LLM 1-5 times, record all responses, and then ask a person with domain knowledge to validate the output (as opposed to accepting the first). That's not perfect, but 'better'