Marks another tally under, "Things I've Seen This Week That Make Me Increasingly Wary of an LLM-Based Society".
Man, it would suck if we built important infrastructure on non-auditable systems that were themselves built on data that doesn't reflect the real world, because didn't set aside the time and effort to verify the integrity and veracity of that data.
This might not be the most interesting topic to many here, but ask yourself: how many other datasets that influence the likes of GPT-4 are essentially wrong about the state of the world?
