To uncover a deepfake video call, ask the caller to turn sideways (metaphysic.ai)

Article summary

The article discusses a method to uncover deepfake video calls by asking the caller to turn sideways, as deepfakes often lack profile views in their training data. The article highlights the abundance of training data for notable actors, but the lack of such data for ordinary people. The method's effectiveness is based on the distinct lack of source material to train models on profile views. This approach can be used to verify the authenticity of a video call.

Main themes

• deepfakes
• video call verification
• scroll hijacking
• chatbot limitations
• training data
• verification methods
• absurd requests

What commenters say

• Some users experienced scroll hijacking while trying to read the article, which made it difficult for them to continue reading.
• The downvoting system on the platform can be broken, leading to low-quality content being downvoted unfairly.
• Training models to generate sideviews can help prevent detection of deepfakes.
• There is a nearly limitless source of training data on social media that can be used to circumvent known circumventions.
• Absurd requests for verification, such as showing specific body parts, can increase in absurdity if not met with significant pushback.
• Simple questions can be used to bust chatbots and reveal their limitations.
• Chatbots' responses can be awkward and incoherent when faced with unexpected questions.
• The effectiveness of a verification method depends on the ability to gauge the response rather than just checking if the answer is correct.