Canadian legislator reads out apparent LLM response in floor speech

Anyone who has used the Internet in recent years is probably used to encountering telltale signs of LLM prompting that lazy users sometimes forget to remove from their AI-generated pabulum. But a Canadian legislator took that idea to an embarrassing new level this week, reading an apparent LLM prompt instruction into the record during a floor speech. Bill Oliver, a Progressive Conservative Party member of the legislative assembly of New Brunswick, noted in a speech last month that “One of the dangers associated with creating advocacy offices is that citizens often develop expectations that exceed the powers actually granted to those offices.” He then went on to say out loud that “here’s a more natural, flowing version of that section that reads like a legislative speech rather than a series of short points,” a line with all the hallmarks of an LLM presenting an alternative style option in response to a prompt. Oliver is far from the first politician to blindly read a speech obviously prepared by someone else. He surely isn’t the first politician to rely on an LLM to aid in speech writing, either. But reading an obviously AI-generated non-sequitur aloud in such a public setting—apparently without realizing what was happening in real time—represents a brave new frontier in the annals of embarrassing revelations of AI use. With his speech, Oliver joins a growing trend of professionals whose AI use is revealed in mortifying ways. Lawyers, authors, journalists, and academics have all faced problems when their LLM use has come to light in the recent past, usually when someone noticed obviously hallucinated errors in ostensibly human-written work. No wonder a Duke University study last year found that workers tend to hide their AI use, as colleagues see their work as “lazy” or “replaceable” for using LLMs.