Mastodon Feed
cstanhope@social.coop ("Your friendly 'net denizen") wrote:
The idea you can prompt away the essential randomness of LLMs persists.
Professional news org: "We used ChatGPT to analyze this database of documents, but it's okay because in the prompt we told it not to guess if it wasn't sure."
😬 At least they said they reviewed and confirmed the results, but still, these people need to hire a consultant from DAIR or something to give them a crash course.