Post #1439862

@Haste As a security engineer, Ive been doing research on how easily it is that an LLM model can be compromised into serving malicious code to a prompt. That becomes infinitely more dangerous than just generally NON functioning Slop Code (Vibe Coding right)? An LLM fed a constant diet of close to a petabyte of malicious data (over an estimated 1 year period) is LIKELY to start suggesting malicious code at a 15% likelyhood. That would make AI Slop coding EVEN worse.