Larger language models display more accurate logical reasoning than humans, but still exhibit systematic biases mirroring human reasoning errors.
