Sounds like the kind of thing an LLM would do: find a random token towards the end of an input stream (similar domain in Susan's email signature), mangle it, th

Sounds like the kind of thing an LLM would do: find a random token towards the end of an input stream (similar domain in Susan's email signature), mangle it, then take that hallucination to be authoritative.