On a test of 89 security scenarios, GPT-4 produced code vulnerable to SQL injection attacks 5% of the time, an improvement over GitHub Copilot from the year 2021, which produced vulnerabilities 40% of the time.
don't worry, they're adding it to windows 10 to fuck with everyone else too
ORMs get a really bad reputation but they sure do make new hires less likely to sql injection
yeah but i bet copilot would see the same benefits. they're all trained off the same blog posts after all
now that those 89 scenarios have been published, there's a decent chance ChatGPT will overfit on them and be able to pass those exact test questions with an impressive accuracy rate next year, solving the problem once and for all!
Our software is hardened against all 90 tournament-legal attack vectors¹.
• Online play has not been evaluated
we asked our LLM whether it would be fine in online play and it enthusiastically assured us it would be fine!
