[kordlessagain]

Right. For example, they are quite useful at generating code and translating the output to Python dictionaries (or JSON). That makes them very useful in code integrations.

One application of casting things to dictionaries is that it makes it easy to grab keyterms from the ingested texts. With the right datastore, we can take the binary set objects and query them very quickly for similarity overlap of other texts, similar to how vector search might return a cosine distance (multi-dimensional angle offset?).

I was thinking these sets could be used to segment vector space, and the only thing that makes that possible is that LLMs are great to use for things like this.