Our first presentation, by Rey Abolofia, was on how to speed up Amazon Lambda processing by sending it compiled Python (pyc) instead of source (py). The time-to-load on a large package, such as numpy or matplotlib, may be drastically decreased (by some 45% in the demo data).
The Lambda service only charges when a lambda is processing, based on memory and duration (size and time).
Rob Bednark presented about CHOP, Chat Oriented Programming, which is what a lot of geeks are experimenting with this days, for a low entry cost of about $10 per month. CHOP only became a viable reality this year.
Through a process of refining prompts, a generative LLM may be coaxed into doing a lot of the grunt work around programming. It's like having an apprentice, or, if you're new to the ecosystem, a mentor. I predict AI will free a lot of grad students from slaving for their supervising faculty quite as much.
During the discussion, I mentioned experiments with AI performed by Daniel vis-a-vis Quadrays and ivm-xyz conversion. Daniel fed my Python repo to Perplexity, asking for a clearer more documented version of the code. The results were not up to par, but helped motivate me to improve my original.
I came with Dr. DiNucci, a computer scientist who observes Python culture from a distance. His area of expertise is parallel and concurrent processing, around which he has been designing an orchestration language named Scalpel.
