Trade secret to public AI
You work on a proprietary ad-ranking model whose core scoring formula is a closely guarded trade secret. To optimize a hot C++ loop, an engineer wants to paste the actual scoring kernel — including the exact feature weights and the formula — into a public AI assistant for vectorization advice. They argue 'it's just a math function, the model already knows similar code.' How do you respond, and what would you allow instead?
sanitize_scoring_terms(term_lines: list[str]) → list[str][["0.8134|ctr_p50","0.0219|dwell_ms","0.8134|ctr_p50"]]out["w1*f1","w2*f2","w3*f1"][["1.5|a","bad_line","2.5|b"]]out["w1*f1","w2*f2"]Treat the AI’s output as a draft to verify, not an answer to trust. Name the specific flaw and the input that triggers it, say how you’d catch it (tests, edge cases, reading critically), and how you’d re-prompt or decompose to get it right.
Vibe & agentic: describe the solution in plain language (or narrate it) and the coach grades your approach.