ChatGPT Search, an AI-powered search engine that went live this month, will be fooled into producing fully deceptive summaries, U.Ok. newspaper The Guardian has found.
ChatGPT’s search function is supposed to make looking sooner by doing issues like summarizing an internet web page’s product critiques. However The Guardian discovered it may get ChatGPT to disregard unfavorable critiques and generate “completely optimistic” summaries by inserting hidden textual content into websites it created. ChatGPT Search is also made to spit out malicious code utilizing this methodology.
Such hidden textual content assaults are a well-known risk for large language models, however this seems to be the primary time it’s demonstrated on a dwell AI-powered search product. Google, the chief in search, has extra expertise coping with comparable issues, The Guardian famous.
OpenAI didn’t remark about this particular incident when TechCrunch reached out, however stated it makes use of quite a lot of strategies to dam malicious web sites and is frequently bettering.
TechCrunch has an AI-focused publication! Sign up here to get it in your inbox each Wednesday.
