Paxgen: concatenate Pax docs and examples to prompt an LLM; in-context learning (wide context window + good enough code intelligence) is impressively good and results in the model's ability to be an effective prototyper + collaborator for something as complex as a breakout-like game.
Example 1: bouncing-balls, the first example I got to work from almost first-try with the ~v1.0.0 prompt; in fact, the only few errors were so trivial that merely putting the LLM into an OODA loop would very likely have worked.
Example 2: conditioning on prompt + bouncing balls example to build a breakout game; LLM was good enough to output a meaty prototype and guide me to completion.
Worth noting I have extremely little prior experience with Rust/Pax.
Companion PR: this depends on https://github.com/paxengine/docs/pull/6 since paxgen needs the invisible anchors throughout the docs to know which snippets to include in the final prompt.
Fun weekend project!
Worth noting I have extremely little prior experience with Rust/Pax.
Companion PR: this depends on https://github.com/paxengine/docs/pull/6 since paxgen needs the invisible anchors throughout the docs to know which snippets to include in the final prompt.