A researcher’s February 1 social-media post, which has since been removed, hinted use of ‘reinforcement learning’ to advance its AI systems.