The only useful way to define an AGI is based on its capabilities, not its implementation details. Based on capabilities alone, current LLMs demonstrate many o

The only useful way to define an AGI is based on its capabilities, not its implementation details. Based on capabilities alone, current LLMs demonstrate many of the capabilities practitioners ten years ago would have tossed into the AGI bucket. What are some top capabilities (meaning inputs and outputs) you think are missing on the path between what we have now and AGI?