Colin Popell

I wrote about my experiments with embedding on json documents - not embedding a field IN a document, but the document as a whole! And, more than that, how I ended up doing what I call 'agentic propositional flattening', though I'm not 'done' yet. Link below.

more procedural PBR for cloth

custom cloth solver. building towards something (for fun)

???

to simulate a cotton garment from scratch, you must first create a universe

simulating sugar maples (we are having some struggles) with some thanks to @zetalyrae

It's funny when people act like agents have lifetimes when the persistence is basically just an appended text file (or database, whatever) and the life of the transformer ends as soon as it stops sending tokens.

you can just do things (browsable recursive summarizers. not done)

one of my more expensive API days

Okay, let's be real: how many of you guys are just building startups that exist to market or idea generate for startups?

Free advice: If you are doing ANY kind of structured LLM work, think about the schema of any kind of operations and build the following around it: - validation - automatic repair where possible - LLM recreation - (risky) LLM repair