3 ms·How was it different?by mach1ne 4y agoHow was it different?drexlspivey 4y agoIt uses reinforcement learning, they had humans rank the responses of GPT-3 and used that as a training set