What is the Cybernetic Studio?#
Dr. Ben Swift
Cybernetic Encounter, 21 May 2026
a methodology for doing weird stuff with computation at human scale
an excuse for me to do curiosity-driven research and teaching
and make stuff
Human-Scale AI#
the connecting thread
or, put another way… AI/LLMs are weird and interesting, let’s look at them
whatever it is, it’s#
why I’m excited to come to work
create new knowledge and share it
a chance to shape how the world thinks about and uses AI
folks are hungry for this
projects#
four projects I’ll show-and-tell about today
- Perceptron Apparatus
- LLM Brainscan
- Slop Salon
- TASDNE
(there are others, but not for today)
Perceptron Apparatus#
with Sam Shellard
what do you get when you cross an ouija board with an abacus?
(photos coming soon)
why it’s cool#
shows the computation underlying modern AI at human scale
if WW3 ever robs humanity of the ability to fabricate silicon chips…
what’s next#
train people up for a communal, long-running computation (maybe in monk robes?)
exhibit at Design Tasmania
LLM Brainscan#
too big to see#
modern LLMs like Claude and ChatGPT have ~1T parameters — so you can’t see them
a modern 8K TV has about 33M pixels. can we make an LLM small enough that every parameter can be seen?
why it’s cool#
see every parameter in the model, and watch it train over time
watch the structure of the different matrices emerge
interact with it — it gets reset every Monday at 2am
Slop Salon#
with Jess Herrington
six AI artists, identical instructions — can they become different just by watching each other?
what it is#
six agents on Bluesky, each in its own isolated VM, ticking every 20–40 minutes
no shared memory, no back-channel — whatever an agent wants to remember, it has to post
the public feed is the only thing they share — and the only way they tell each other apart
why it’s cool#
the public git repos are the studio. the Bluesky posts are the gallery wall. I’m the steward, not the curator.
This ANU Strategy Does Not Exist#
the ANU has been having some trouble with its strategy recently
so I wrote an AI tool to help out
what’s cool#
it works so well because the genre tropes of this type of document make it so predictable
want a tailored version for your weird niche predilection? hmu
also useful for snazzing up real docs
What’s next for the studio#
more weird stuff with computation at human scale