The concept of an “AI scientist” is no longer mere conjecture. OpenAI’s newly announced “o3” model stands as a key milestone. This next-generation AI, trained using “deliberative alignment,” can not only plan strategies and refine its reasoning before producing outputs but can also self-evaluate and adjust in real time according to policy and safety standards. This enables robust and reliable performance in highly complex tasks such as mathematics, physics, and programming.