Videos created by the Sora neural network turned out to be difficult to distinguish from real ones

Users had difficulty distinguishing videos created by the OpenAI Sora neural network from real ones. This became known from survey HarrisX company has over 1 thousand Americans aged 18 years and older.

Respondents were randomly shown four videos generated by the Sora neural network and four videos shot on a real camera. The task was to determine whether the video was generated by AI or filmed by a human. 60% of respondents made mistakes in five out of eight videos, from which the study authors concluded that content created with the help of AI is becoming more and more realistic and less and less distinguishable from the real thing.

OpenAI presented Sora neural network in February 2024, it is capable of generating high-quality videos up to a minute long using a text description. The neural network can create photorealistic videos and animate static images. Sora is not currently available to the general public.

Evgeniy Fedunenko