Viewpoints for AI

The Viewpoints

If you ever hit a wall in a creative exercise where someone told you to "imagine you're a mountain" or "pretend a monster is chasing you," you might like the viewpoints.

Me from when I was a performer. I'm smiling because I don't know what an Abstract Syntax Tree is yet

The viewpoints (and I'm talking about the system I learned, the Anne Bogart version) are a vocabulary to talk about performance:

• Architecture
• Spatial Relationship
• Topography
• Shape
• Gesture
• Tempo
• Duration
• Kinesthetic Response
• Repetition

The idea is that these labels make a framework which allows you to talk more specifically about rehearsal, performance, anything happening in time and space.

"I really liked the part where you did that wavey thing with your hand"

becomes:

"I liked how you extended the duration of the wave gesture."

Not scientific. Still somewhat flowery language. But a shared vocabulary that points towards the kinds of things a performer has the ability to adjust.

Possible Viewpoints of AI:

Tempo

I have no idea what an LLM is doing when it generates tokens. I do know whatever's happening happens fast.

Human thought is fast. I can go from weird sensation on my arm while driving into a full spiral over an imagined tick in about two seconds.

LLM token generation is fast and getting faster. Tempo changes the user experience. Same model, but it'll feel a whole heck of a lot smarter/more magical as it gets faster.

sync / async

Are you talking back and forth with the LLM or spec-ing work and then reviewing later?

freaked-out-ness

How are you feeling?

I appreciate when people own this. You can be right about AI or wrong about AI and still be pretty freaked out by just the uncanniness of the computer talking back to you.

For me, this comes and goes in waves. I'm not mourning the craft of software engineering. Mad respect, but boy did I spend a lot of time hunting for semi-colons and matching brackets in 2022.

I do think it's absolutely nuts that this technology exists, though, and I feel myself anxiously bracing for all the ways it's going to reshape the world.

hopes

my inbox: 6 pleas for my attention, a sneaky one got me!!!! (and one of which is me doing it to myself!)

What do you want from AI?

I want a three-day work week and the agency to control the software that controls what words and pictures pass in front of my eyeballs.

I hope for a world where "smart" doesn't matter as much, and isn't as important to our identities.

I hope if algorithmic feeds are like a teenager smoking a cigarette, AI can be the parent coming home and making the kid smoke the whole pack. I hope we all throw up.

Deterministic/non-deterministic spectrum

The "black box" weights introduce quite a lot of non-determinism. You can fight back against this with deterministic scripts. It's a spectrum.

"They're non-deterministic systems!"

Yes. So add a few pinches of determinism back into the dough.

fears

To name a couple of mine:

• losing the need to ask for help, we're gonna lose connection/relationships/signal
• what if the systems just keep getting better?
• AI twitter is on twitter in spite of the right wing echo chamber, right? Right? Right?!?!?

more:

• slop persistence (are you hoarding or letting it flow?)
• oversight (are you reading the code? Really? How often do you read the code and how bad do you feel about not reading it all?)
• blast radius
• prompt length (are you saying "yeah sounds good" or "Here's teh 50-point plan...")
• verifiability of your domain
• your knowledge of the domain