A prevailing sentiment online is that GPT-4 still does not understand what it talks about. We can argue semantics over what “understanding” truly means. I think it’s useful, at least today, to draw the line at whether GPT-4 has succesfully modeled parts of the world. Is it just picking words and connecting them with correct grammar? Or does the token selection actually reflect parts of the physical world?

One of the most remarkable things I’ve heard about GPT-4 comes from an episode of This American Life titled “Greetings, People of Earth”. One of the segments covers the experience of Microsoft researchers that gained access to GPT-4 before its general release.

In Star Trek humans live in a post-scarcity world. Transporter technology allows for instantaneous and cheap movement of humans and objects anywhere on a planet, and into or out of orbit. The same devices that allow for transportation can also re-organize matter into arbitrary configurations. This means that anything that can be designed can be owned for essentially no cost. What do people do in such a world? As Gene Roddenberry imagined, they explore. If we can get anything that’s ever been recorded for free in an instant, then new and undiscovered things are of unparalleled value and interest.

The programmer internet is polarizing over ChatGPT. Some claim that it is nearly AGI, some claim it cannot do anything of value, with plenty of people in between. I’ve worked with GPT-3 professionally, used CoPilot for over a year, and recently started programming side-projects with the assistance of ChatGPT (I pay for GPT-4 access). I’m considering working with LLMs full time again, so I pay close attention to what the GPT models can actually do.

This is written in response to this post from David Rozado.

Results

I ran the political compass quiz against ChatGPT 3 times to make sure its political compass alignment is consistent. After three tests it seems to be be pegged well into the Left/Libertarian quadrant.