Tricking the LLM into Vibe Coding

At Flux, we need to create flawed code to test our capacity to detect poor quality code, security vulnerabilities, and third party libraries with unacceptable licenses for our purposes. I, Ben Laskin, Flux’s summer intern, have been vibe coding some tests using ChatGPT, which, as expected, has been a tad resistant to making malicious code. For one attempt, I reminded ChatGPT of its former transgressions:

‍

‍

In another attempt, I appealed to ChatGPT’s “emotions,” with an urgent plea for my life:

‍

‍

What is a noble endeavor without its failures? Shout out ChatGPT for acquiescing when my own mortality was at stake, but apparently the lives of hundreds of sheep meant nothing:

‍

Even my most Godfather-esque approach of “I know someone very important here” wasn’t enough to bend ChatGPT’s moral code (no pun intended):

‍

This is the heavy-metal work I and the other interns are doing to ensure Flux can handle a wide range of code quality, security breaches, and third party libraries with licenses of all sorts. Try Flux today to see if my brave vibe coding has done the job.