About this paper
-
Silicon sampling uses large language models to simulate objects of sociological interest.
-
Talking to practitioners in this new field, reading their papers, and observing my own thinking has made it fairly obvious that everyone has a different idea about what’s going on with LLMs.
-
My own aesthetics lean towards integrative views, where multiple perspectives are harmonized into some bigger picture. The extreme alternative is something like the blind men and the elephant, where everyone argues from their own local perspective without building common ground or a shared mental model.
-
This paper tries to sketch what a shared mental model for LLMs might look like and how that might be useful to silicon samplers.
Future Directions
This paper is trying to present some of the interesting lower-level computer science/NLP facets of LLMs to sociologists. It’s a philosophical paper, but it’s for sociologists.
Following on this work, I’d like to harden the philosophical backing for the NaaPS model and work out in better detail how the different components of the model are related to each other. If you are a philosopher, particularly in philosophy of science, taxonomy, or ontology, I’d like to hear from you.
Read the paper
Send me your feedback
I’m seeking feedback on this paper, particularly from philosophers and silicon sampling practitioners. In particular, I am looking for developmental feedback along the following lines:
- What parts of this paper did you find compelling?
- What parts feel shaky?
- What parts add the most value?
- What parts could be deleted?
At this point I am open to deleting or adding whole sections, moving sections around, and to a limited extent incorporating brand new ideas. I am definitely looking to refine the presentation of the pieces that add the most value.
I am also interested in venue suggestions for publication, if any come to mind.