[Discuss] [arxiv] How Secure is Code Generated by ChatGPT

[richard.pieri at gmail.com (Rich Pieri)]

On Sat, 22 Apr 2023 18:02:44 -0400
John Abreau <abreauj at gmail.com> wrote:

> Found a story at theregister about ChatGPT generating insecure code.

A program that's designed to ingest people's code and regurgitate
infinite variations of that code producing buggy code? Who could have
thought? Well, anyone with at least two brain cells to rub together.

Or how about this:

February 26, 2023: Clear, Detailed, And Wrong
http://www.sjgames.com/ill/archive/2023-02-26

Andrew Walters writes: "I am asking ChatGPT questions about Munchkin
rules and cards. It is wrong 100% of the time, even with questions that
are on the FAQ. ChatGPT says, explicitly, that the Munchkin FAQ was
part of its training material.

"The responses were perfectly grammatically correct, detailed, and
sounded very confident. And wrong every time. Well, 0/5. I have a
transcript."

-- 
\m/ (--) \m/