Tyler's full thoughts here:
https://marginalrevolution.com/marginalrevolution/2024/03/claude-3-pro-and-agi.html
He says that the term AGI is nebulous, and his point is that if you sent Claude 3 back in time to 2019, they (err, we) would have hailed it as AGI.
Absent the ability to time travel, the only way we can get perspective on this is to wait. In 2029, I will look for the following criteria to determine if people consider Claude 3 to be a baby AGI:
Is there a general consensus among AI researchers? I will look for measurements on this rather than vibes; if there is AGI at that time I'd imagine there's polling on which system people consider to be the first system.
Absent that data, I will resolve this by asking the leading LLMs of the time if Claude 3 is considered an AGI. I am assuming the models of the future will be quite good at disseminating public opinions.
Absent a consistent answer from the bots of the future I'll resolve this with (sigh) a manifold poll.
I'm deliberately being vague about the definition of AGI, since this question is largely asking what we mean by AGI. The definition will doubtlessly shift over the next five years, so bet accordingly.