Yes, small models checking conversations and pruning GPT answers if needed is definitely what should be done, it's much more reliable than one-shot prompts. Bu

Yes, small models checking conversations and pruning GPT answers if needed is definitely what should be done, it's much more reliable than one-shot prompts. But OpenAI is already hemorrhaging money, so they definitely can't afford to run 2 inferences for every answer.