Learning from Failure to Tackle Extremely Hard Problems

6 minute read

This is a blog post about our work on BaNEL: Exploration Posteriors for Generative Modeling Using Only Negative Rewards (https://arxiv.org/abs/2510.09596).