PANIC! is an interactive installation where you can play with feedback loops of generative AI models hooked end-to-end. As well as generating intriguing text, images and audio, PANIC explores how different ways of connecting these models up can give rise to different patterns of outputs, emergent behaviours, recurring patterns, and degenerate cases. Today, the low barrier to entry in generative AI model platforms (in terms of cost, time, and knowledge) means that more and more of us are using them. But just because we can put our text/images/audio into these models and have them provide new text/images/audio in return, does it mean that we should?
It's a standard AshPhoenix and LiveView app, so all those guides should help you out.
Inside the codebase here's a (domain) glossary to help you figure out what's going on:
- model: a particular AI model (e.g. Stable Diffusion, GPT4o)
- platform: model-hosting cloud platform (e.g. Replicate, OpenAI, Gemini)
- network: a specific network (i.e. cyclic graph) of models, designed so that the output of one is fed as input to the next
- invocation: a specific "inference" event for a single model; includes both the input (prompt) an the output (prediction) along with some other metadata
- run: a specific sequence of predictions starting from an initial prompt and following the models in a network
- watcher: a display (either TV/computer or Vestaboard split-flap display) which shows "live" updates of the stream of invocations coming out of the run
- installation a specific configuration of watchers (probably representing a physical install, but not necessarily)
At the School of Cybernetics we love thinking about the way that feedback loops (and the connections between components in general) define the behaviour of the systems in which we live, work and create. That interest sits behind the design of PANIC as a tool for making (and breaking!) networks of hosted Creative AI models, and in our preso we're happy to go into some of the cybernetic ideas behind the design - or to focus on the app (and its inputs/outputs) itself - happy to fit with whatever will make the most kickarse symposium for everyone.