As recently as Dec 2024, the typical AI researcher thought AI wouldn't solve a Millennium math problem until 2054, prove publishable math theorems until almost 2040, or find and patch a security flaw for a system with 100k+ users until the early 2030s

As of December 2024, AI researchers generally predicted that artificial intelligence would not solve a Millennium math problem until 2054. They also estimated that AI would not be able to prove publishable math theorems until nearly 2040. Furthermore, the consensus among researchers was that AI would not be capable of finding and patching a security flaw for a system with over 100,000 users until the early 2030s, according to a report from aiimpacts.org.