I built NiceTryGPT, an open-source project for CTF authors who want to make cheap LLM shortcuts less effective without making challenges harder or more annoying for humans. The idea is simple: «Increase uncertainty, not complexity.» NiceTryGPT keeps the same vulnerability and learning objective, but removes obvious patterns that can turn a challenge into a one-shot LLM answer. Current examples include IDOR, path traversal and SQL injection. I'm not trying to make CTFs “AI-p...