این روزا همه جا صحبت از ایجنت‌های هوش مصنوعی دانشمنده، اما واقعیتش اینه که بیشترشون فقط با آزمون و خطای فله‌ای و بدون هیچ درک عمیقی دنبال جواب می‌گردن

AI scientist agents are great at optimizing and finding solutions through endless trial and error. But is that really what science is about? As Terrance Tao eloquently put, the role of math and science should be more than that – they are "lighthouses" that guide and inspire exploration through understanding and insights. Can AI agents discover novel insights through experimentation? To study these gaps, we worked with domain experts to introduce EurekaBench, a benchmark spanning 6 science domains that evaluates an agent’s ability to discover genuinely new scientific insights. (1/n)