Post #4167343
2026-07-28 15:13 UTC
Excellent piece by @pluralistic@mamot.fr:
https://pluralistic.net/2026/07/28/hitl-ers/#ai-ai-oh
If a human domain expert asks an AI agent a question about that domain, he or she can get a fast response, and then decide whether or not it makes sense. Good deal! Every really interesting use case for LLMs that I've seen matches this pattern.
And every instance where a non-expert chooses to trust an LLM's output is a mistake.
Replies (1)
-
@gooba42@mastodon.social 2026-07-28 15:31
@mikeolson@mastodon.social @pluralistic@mamot.fr Even incorrect responses have value to an expert who is tinkering with an idea and can use the randomized inputs to inspire new perspectives. "Wrong, but interesting" has value to an expert that it doesn't for a non-expert.