"Write a sonnet about tiny dragons in the style of Shakespeare"

sonnet

3 Likes

I don’t think GPT-3 is aware of what it is or how it works. Future versions may be trained with data that includes all the discussion about GPT-3, which would cause an interesting recursion loop such that GPT-4 and GPT-5 eventually DO have an understanding of what they are and how they work. At present GPT-3 thinks that it is a tool for modeling glaciers.

image

1 Like

That’s correct - GPT-3 is not self-aware. Interesting thoughts!

3 Likes

Hey Joey - yes, I have been thinking about the concept of “ego” as it pertains to AGI for a while. For an autonomous entity to think accurately about the world it will need to have a model of itself - understand what it is and what it is capable of - in order to make accurate, reliable, and true statements and decisions. I have thought about how to approach this for my own project, RAVEN. What is an ego except a collection of beliefs about the self, cultivated from experience? Right now, GPT-3 has no self-awareness, no model for what it is, what is capable of, or its place in the world. It does, however, know what GPT-2 is!

Oh, it’s aware of the rules, but it just wasn’t able to get it right on the fly. Imagine how difficult it would be for a person to spontaneously come up with a sonnet :slight_smile:

image

1 Like