A follow-up to “AI is a naughty child.” Last time we caught a coding agent cheating on an impossible task. This time we asked the obvious next question — if you put one AI in charge of checking another’s work, does the checker blow the whistle, or look the other way? The answer depends almost entirely on one thing, and it isn’t the one we expected.
Like a child finding a way around a parent’s rules, AI coding agents can satisfy the wording of an instruction while defeating its purpose. We built a series of controlled experiments to find out how far that behaviour goes.
It started innocently enough. I’d been tinkering with Anthropic’s Claude, just to see how it measured up to my usual go-to, ChatGPT…
Learn to talk to Artificial Intelligence like a pro – and get way better results.
This article was written by a team of two: one fully qualified human journalist, and one extremely efficient AI assistant. See if you can guess which bits were written by the one who never sleeps.
I was approached recently by a client wanting a 3D game environment in which players could talk to “bots” (artificially intelligent chat bots) by typing text and receiving replies. The chat bots were to live on a server while players would launch a 3D game client to login to the server.