The researchers at the University of California, Santa Cruz, and Johns Hopkins showed that, in simulated trials, AI systems and the large vision language models (LVLMs) underpinning them would reliably follow instructions if displayed on signs held up in their camera’s view.
Yes, the thing that can be prompt injected can be prompt injected.
The headline is false. These are simulated vehicles, not vehicles.
I could see this IRL, though. The more the industry tries to stuff driverless cars in to every city (and let’s not lose site of this being fully the wrong solution to the public mobility), more and more people are going to try messing with them, to varying degrees of success. We will find out whether something like this works in fairly short order, I would say.
Test it on actual self driving cars on a controlled course, instead of pretending ChatGPT is a “self driving car”.