Great work! I'm referring your paper on my current project. But I'm still wondering how we can integrate this simple lie detecting classifier into current AI practice to improve our language models in practice. To my best knowledge, OpenAI takes its own approaches for AI alignment such as RLHF or reward modelling. If you have thought about this, I'd definitely love to hear your idea or future map.
Great work! I'm referring your paper on my current project. But I'm still wondering how we can integrate this simple lie detecting classifier into current AI practice to improve our language models in practice. To my best knowledge, OpenAI takes its own approaches for AI alignment such as RLHF or reward modelling. If you have thought about this, I'd definitely love to hear your idea or future map.