News for hackers. Stuff that matters.

A comment on I'm sorry but I cannot fulfill this request it goes against OpenAI use policy. Parent comment

Re:I'm sorry but I cannot fulfill this request it goes again... by altairprime on Friday January 12, 2024 @01:28PM (#38971909)
No. Random employees have a well-understood distribution of mostly normal human errors of certain types and estimated severity, relative to unattended LLM which has a poorly-understood distribution of errors in both type and severity. (“SolidGoldMagikarp”.)
[ Parent | on HN ]
Re:I'm sorry but I cannot fulfill this request it goes again... by dylan604 on Friday January 12, 2024 @02:45PM (#38972965)
copy&paste errors are exactly what human employees are good at. this could very easily be the result of a bad copy&paste by a human into a form. especially if the copy&paste text is in a language not understood by the human employee. to them, it might look just like one of the other hundreds of search term word salad used as titles
[ Parent | on HN ]
Re:I'm sorry but I cannot fulfill this request it goes again... by altairprime on Friday January 12, 2024 @03:06PM (#38973253)
Whether it’s human or not is irrelevant to the point: human beings fail much more predictably.

When the same search term salad is presented hundreds of times for copy paste, a human would notice and have an opportunity to ask a supervisor.

A chatbot automation would not notice the repetition unless it had been coded to detect repetition, and/or to reject the ChatGPT refusal message.

Ironically, it was probably an automation coded by ChatGPT.

[ Parent | on HN ]