If LLMs can't write, I doubt it can lead us to AGI
SMRTR summary
Despite being impressive at coding, AI consistently fails at writing in a human voice, even when given strong examples to follow. That's because code has clear, checkable answers and rewards repeatable patterns, while good writing does the opposite. Until AI can handle open-ended, judgment-based tasks like writing, it likely isn't close to reaching true general intelligence.
SMRTR provides this summary for quick context. The original article belongs to Hacker News.
Read the original article