Can language models overcome brittleness and reach AGI?

The Verge describes a Decoder conversation in which Kevin Roose says he thinks large language models could reach artificial general intelligence (AGI), though he isn't certain. Host Nilay Patel remains skeptical, pointing to AI's brittleness on tasks without clear right-or-wrong feedback. They also discuss Roose's book, The AGI Chronicles, and his plans to make his new Machine Gods podcast with Casey Newton video-first.