we could spend another decade continuing to explore the implications and applications of GPT-3.5-era models alone. the fact that we have GPT-6-level models capable of solving millenium prize problems now… we need much more exploration much faster.