Reinforcement learning towards broadly and persistently beneficial models (opens in new tab)

This is an unofficial We find that reinforcement learning on realistic scenarios targeting beneficial traits can produce broad improvements across dozens of benchmarks measuring aligned and beneficial behavior. These alignment gains generalize beyond the domains used for training and persist under adversarial pressure. As AI systems become more capable and autonomous in high-stakes settings like health, science, education, and coding, they will need to remain helpful, honest, transparent, and...