How should we test AI for human-level intelligence? OpenAI’s o3 electrifies quest

Nature – Experimental model’s record-breaking performance on science and maths tests wows researchers.