H. Claire Taylor: 1
[To hack an AI] all you really have to do is have two LLMs “talk” to each other for 5 minutes and they both erupt into gibberish. I recently read a book (More Everything Forever by Adam Becker) that pointed out that LLMs don’t hallucinate sometimes. Every response is actually a hallucination. Thought that was very astute! And also how certain people in charge of things seem to be functioning.
* Subscribe to her Substack and she can afford to write more!
Check out her home page
- 1. Friends don’t let friends order from Amazon ↩︎
Comments