Today, OpenAI researchers are releasing a new scientific paper on the company’s website and on arXiv.org (embedded below) revealing a new algorithm they’ve developed by which large language models (LLMs) such as OpenAI’s GPT-4 (which powers some versions of ChatGPT) can learn to better explain themselves to their users. The paper is titled “Prover-Verifier Games Improve Legibility of LLM Outputs.” This is critical for establishing trustworthiness in AI systems especially as they become more powerful and integrated into fields where incorrectness is dangerous or a matter of life-or-death, such as healthcare, law, energy, military and defense applications, and other critical infrastructure.
No comments:
Post a Comment