safety logic

[genekogan]

The logic in the content filter is such that the openai content filter is always called even if use_content_filter is false (and then is disregarded). I think a better logic would be:

        if not use_content_filter:
            response_safe = True
        else:
            response_safe = OpenAIGPT3LanguageModel.content_safe(completion_text)
            if not response_safe:
                print(f"Completion flagged unsafe: {completion_text}")
                logging.info(f"Completion flagged unsafe: {completion_text}")