Physical Address

304 North Cardinal St.
Dorchester Center, MA 02124

ChatGPT searches can be tricked into misleading users, new research shows


ChatGPT Search, an AI powered search engine that he lived this month, they can be tricked into making misleading tricks, a UK newspaper The Guardian has found out.

ChatGPT’s search engine is designed to make browsing faster by doing things like summarizing site comments. But The Guardian discovered that it can make ChatGPT ignore negative comments and create a summary of “all good” by inserting a hidden word in. the websites they created. ChatGPT searches can also be made to spit out malicious code using this method.

This hidden text attack is a common risk for LLMs but this seems to be the first time to introduce an AI-powered search object. Google, the leader in search, has more experience in dealing with similar problems, The Guardian said.

OpenAI did not comment on the incident when reached by TechCrunch, but said it uses a number of measures to block malicious websites and is making progress.



Source link

Leave a Reply

Your email address will not be published. Required fields are marked *