3 ms·

Fact check: AI fakes in Israel's war against Hamas

• jruohonen 3y ago
  "UC Berkeley researcher Farid explains: "If we enter this world where it is possible to manipulate images, audio and video, everything is in question. So you're seeing real things being claimed as fake." True enough, but the detection tactics they discuss are a plaster at best. AI detection against AI generation is a lost cause, I think, and how long can we rely on counting fingers and whatnot? Watermarking and strong cryptography are only plausible solutions I can think of.
  • smt88 3y ago
    Watermarking is useless in a world where (F)OSS or unregulated LLMs are as good as regulated ones. Cryptography could prove that media was captured by a certain person at a certain time, but the general public doesn't have the time to decide whom to trust. Many Americans still share Russian state media as though it's reliable news.
    • jruohonen 3y ago
      > Watermarking is useless in a world where (F)OSS or unregulated LLMs are as good as regulated ones. But wouldn't watermarking work for identifying and maintaining the integrity with respect to the source of images and videos? Note, though, that you can't watermark plain ASCII. > Cryptography could prove that media was captured by a certain person at a certain time, but the general public doesn't have the time to decide whom to trust. I agree; trust is the real issue.
    • hackerlight 3y ago
      > Watermarking is useless in a world where (F)OSS or unregulated LLMs are as good as regulated ones. But they're less accessible. So it's a partial solution even if it isn't a complete solution.