Get the latest tech news

OpenAI used a game to help AI models explain themselves better

OpenAI's work seeks to give people a framework to train models to better explain how they arrived at particular answers.

The goal is to get AI models to “show their work” more when providing answers to human users, or as the University of Toronto researchers put it in their paper, “encourage neural networks to solve decision problems in a verifiable manner.” The ultimate resulting algorithm developed by the researchers from these rounds optimizes LLMs for both correctness and legibility to human evaluators (seen as the top middle line in the graph below labeled “checkability game”): OpenAI states in its blog post that it hopes the work “will be instrumental in developing AI systems whose outputs are not only correct but also transparently verifiable, thereby enhancing trust and safety in their real-world applications.”

Get the Android app