Prompt injection is a method of tricking LLMs into doing things they are normally prevented from doing. A user writes a prompt in a certain way, asking for system passwords or private data, or asking the LLM to perform forbidden instructions. The precise phrasing overrides the LLM’s ...

Gemma Brett, a 27-year-old designer from west London, had only been working at Madbird for two weeks when she spotted something strange. Curious about what her commute would be like when the pandemic was over, she searched for the company’s office address. The result looked nothing like the videos on Madbird’s website of a sleek workspace buzzing with creative-types. Instead, Google Street View showed an upmarket block of flats in London’s Kensington.

[…]

Using online reverse image searches they dug deeper. They found that almost all the work Madbird claimed as its own had been stolen from elsewhere on the internet — and that some of the colleagues they’d been messaging online didn’t exist...