Embedding Trajectories experiments

This repo is a place for us to experiment (research woo!) with different ideas about semantic trajectories, genAI, embeddings, TDA and more.

Motivation and research questions

Anyone who's played with Panic (or watched others play with Panic) has probably had one of these questions cross their mind at some point.

The purpose of this project is to give us answers to these questions which are both quantifiable and satisfying (i.e. we feel like they represent deeper truths about the process).

The hope is that these methods are useful outside Panic as well, to understand and predict semantic trajectories/narrative structure/the shape of information flows in animal and machine communication).

how did it get here from that initial prompt?

• was it predictable that it would end up here?
• how sensitive is it to the input, i.e. would it still have ended up here with a slightly different prompt?
• how sensitive is it to the random seed(s) of the models?

is it stuck?

• the text/images it's generating now seem to be "semantically stable"; will it ever move on to a different topic?
• is it predictable which initial prompts lead to a "stuck" trajectory?

has it done this before?

• how similar is this run's trajectory to other runs?
• what determines whether they'll be similar? initial prompt, or something else?

which models have the biggest impact on what happens?

• do certain models dominate the trajectory? or is it an emergent property of the interactions between all models in the network?

Repo structure

• fake-panic/ contains a couple of python scripts for running a "fake" version of Panic (Panic itself is in a separate repo, ask Ben if you need access).

• lit-trajectories/ contains python scripts for chunking + embedding classic texts from The Canon^TM and looking at their trajectories in embedding space.

There will be more subfolders as we come up with more experiments---although we can also share code between experiments too. For now, just get stuff done and we can refactor as we go.

See the READMEs in each subfolder for more info about each project.

TODO

• create a unified data model for all trajectories (timeseries of semantic data points), perhaps with Pydantic?
• move the TDA code across from Sungyeon's other TDA repo
• switch to a vector database for the embeddings (in fact, for all data)
• run this on cybersonic

Authors

Ben Swift, Sungyeon Hong, hopefully others :) Check the commit history to see who did what.

Licence

MIT