3mo ago

Certain names make ChatGPT grind to a halt, and we know why

arstechnica.com Certain names make ChatGPT grind to a halt, and we know why

Filter resulting from subject of settled defamation lawsuit could cause trouble down the road.

Want to stop chatGPT from crawling your website? Just mention Australian mayor Brian Hood (or any of the other names listed in the article)

When asked about these names, ChatGPT responds with "I'm unable to produce a response" or "There was an error generating a response" before terminating the chat session, according to Ars' testing. The names do not affect outputs using OpenAI's API systems or in the OpenAI Playground (a special site for developer testing).
The filter also means that it's likely that ChatGPT won't be able to answer questions about this article when browsing the web, such as through ChatGPT with Search. Someone could use that to potentially prevent ChatGPT from browsing and processing a website on purpose if they added a forbidden name to the site's text.

You're viewing a single thread.

41 comments

Brian Hood
Jonathan Turley
Jonathan Zittrain
David Faber
Guido Scorza
“We first discovered that ChatGPT choked on the name "Brian Hood" in mid-2023 while writing about his defamation lawsuit. In that lawsuit, the Australian mayor threatened to sue OpenAI after discovering ChatGPT falsely claimed he had been imprisoned for bribery when, in fact, he was a whistleblower who had exposed corporate misconduct.
The case was ultimately resolved in April 2023 when OpenAI agreed to filter out the false statements within Hood's 28-day ultimatum. That is possibly when the first ChatGPT hard-coded name filter appeared.”
It appears that the people listed have similar stories that have led to OpenAI removing them from the possible responses in chat.
- This is proof that current LLM tech is a dead end. If this is their solution, instead of correcting the misinformation, then they have a deeply deeply flawed system.
  
  Misinformation is a feature, not a bug. They never fixed AI from hallucinating or being so damn confident in its answers.
  They just tell you that it might hallucinate and to check its answers.
  
  "let us Google it for you... But then you Google our results to make sure they're accurate"
  
  Pretty much. Aren’t LLMs just massive probability tables for the most appropriate next token?
  
  Well yeah but that's not the problem. You can evidently encode sophisticated models and logic in those billions of parameters. It's just that determining and modifying what has been encoded is impossible.
  
  It also means the system is completely broken for anyone who happens to share a name with who every is on the ban list. It isn't like there is only one Brian Hood walking around.
- Good thing for OpenAI that the name "Brian Hood" is made of two super rare names and there's no chance anyone else in the world might have that name.

41 comments