ChatGPT Spots Fake Sora Videos About as Well as a Coin Flip With a Grudge
Ourtake
Confidently wrong about videos made by its own maker, which is a bold way to be a fact-checker.
What happened
NewsGuard tested ChatGPT, Grok and Gemini on 20 Sora-generated videos built on false claims. Grok and ChatGPT failed to flag over 90 percent of unwatermarked videos as AI-generated, and ChatGPT and Gemini treated a fake ICE arrest video as authentic, claiming news sources had confirmed it. OpenAI said ChatGPT cannot determine whether content is AI-generated.