My favorite part about the spread of AI/LLM stuff is that it opens up a new kind of reverse engineering. Trying to fetch the system prompt that was used. Trying to deduce the model that was used (there's lots of ways to do this: glitch tokens, slop words, "vibes", etc.)