Note that the post is a little misleading; the paper is about jobs AI can be applied to, not necessarily jobs that are at risk.
They also were not studying people using LLMs in these specific roles. They used LLMs (with all the caveats that implies) to a) analyze a database of conversations with Bing Copilot and sort the requests into various categories, like "gather information" (by far the most common) or "work with the public", and b) match up these tasks with occupation descriptions in a job database and with Bureau of Labour statistics.
Like I'm sure the paper's still crap but that isn't exactly what they're saying.
@V0ldek @BlueMonday1984 Thank you for linking the source, because in replies someone linked the actual paper (it's table S3).
Note that the post is a little misleading; the paper is about jobs AI can be applied to, not necessarily jobs that are at risk.
They also were not studying people using LLMs in these specific roles. They used LLMs (with all the caveats that implies) to a) analyze a database of conversations with Bing Copilot and sort the requests into various categories, like "gather information" (by far the most common) or "work with the public", and b) match up these tasks with occupation descriptions in a job database and with Bureau of Labour statistics.
Like I'm sure the paper's still crap but that isn't exactly what they're saying.