Sorted by New

Wiki Contributions


I think most people on LW try to keep their speculations to a minimum mainly to avoid embarrassment for when they don't come true. I have no such worries when it comes to this technology specifically, since the outcome is so obvious that it would be more unlikely for it to not happen.

While I'm not certain about any particular other content that might scratch the same itch, I definitely plan to write more posts on other topics similar to this one.

There's nothing stopping people from making fan films at the current moment, generally with the limitation that it isn't put up for sale. I would find them being able to shut down progress on this tech dubious at best, but certainly not outside the realm of possibility.

It's important to consider the idea that these models could also be used in tandem with actual footage as a cheap alternative to the modern CGI pipeline. Instead of paying hundreds of artists to painstakingly make your CGI alien world, you could ask a future model to inpaint the green screens with an alien world that it generates.

It wouldn't surprise me at all if we soon see an update to DALL·E similar to the instruct series for GPT-3. Not a total reworking, but a significant enough change that it produces improved results.

I could sit here and speculate on potential workarounds, but you're probably correct. It makes sense that if you want to place yourself in movies, you would need to first build a comprehensive model of yourself for the AI to work with. Fortunately, this is the kind of thing you'd only need to do once.

Eliezer should have taken Cotra up on that bet about "will someone train a 10T param model before end days" considering one already exists.