Discussion about this post

User's avatar
David Yates's avatar

>And there is no need to fear we will program a creative AI because we have no idea how creativity works. Heck, we still have no idea how consciousness works, never mind advanced functions like creativity.

I agree very much with the overall thesis of this piece, but I'm not sure this is a very strong argument, considering that LLMs are not so much the result of deliberate programming as a surprising accident of scale. The transformer architecture, which modern LLMs are based on, was originally developed for machine translation. LLMs arose from feeding way more data than ever before into this transformer architecture, and the results surprised everyone. We don't really understand how or why these things work as well as they do. So understanding how something works does not seem to be an essential prerequisite for creating it, in this realm at least.

Expand full comment

Failed to load posts

Ready for more?

Cookie Policy

We use cookies to improve your experience, for analytics, and for marketing. You can accept, reject, or manage your preferences. See our privacy policy.