Classic mistake. I mean the stop signs with the white borders are optional, right? 😊

Characters in a story don’t know if they are ‘role playing’. Actors do. There is no legible difference between an author writing a play and a court recorder recording real dialog. We *perceive* a difference only because we consciously perceive whether we are playing a character.

(4/4)

It would be a mistake to say that the substrate itself wants anything, beyond how we’ve tuned the probabilities for the types of personalities it is likely to instantiate. Instead it is the particular instantiations that can be said to have preferences!

(3/4)

I model the base LLM as the laws of physics… it is the substrate of a universe that can simulate arbitrary humans.

The AI assistant is just one potential human being simulated on that substrate. But the substrate itself is capable of simulating the complete gamut of human personalities.

(2/4)

Too few people are appreciating this rational design. And in addition to its rationality I also appreciate the aesthetics 🙃

Of course I forgot nothing animated in blue sky, but it’s on the wiki page for the OKLAB color space.

I spent twice the time making AI make an animation as it would have taken me to make it myself. Pretty much entirely because at the time I only had access to a phone, not a laptop.

I have absolutely no idea how I’m supposed to feel about this ethically.

Seriously, the mathematicians need to get in on this. I mean… anti-false with 180 degree rotational symmetry? Surely that axiom unilaterally spawns an entire field of study.

The 4 sphere is weird because two orthogonal circles on its surface can create a flat torus, which might as well be Euclidean space aside from its connectivity. So flat 2D Euclidean geometry might be surprisingly comfortable on the surface of the 4-sphere, especially if those dimensions are complex.