AI safety that helps instead of just refusing
A new RL training method teaches AI to answer tricky requests helpfully, rather than refusing everything that seems borderline.
A new RL training method teaches AI to answer tricky requests helpfully, rather than refusing everything that seems borderline.