Is GPT4 A Surprise?
More than a decade ago, IBM’s Watson beat Ken Jennings. Now, if we take a crude Moore’s law doubling of capability every two years, that would be over six doublings. Is GPT4 64 time better than Watson? Maybe even more. It definitely seems that way.
The issue we have is that it’s hard to understand exponential growth looking forward. In twelve years, 2035, whatever incarnation of language model might be 64 times more capability than GPT4. What does that even look like?
Are we potentially looking at thought completion instead of sentence completion? Will webcams or cell cams running at 240 fps be able to see our micro-expressions as they occur and the technology make emotion assessments based on this? Could AI systems not only write out responses to email but simulate several exchanges to predict outcomes?
It makes me wonder about the endgame of such technologies. They might up creating human determinism. To us, it will still look like fun and cool things.