I didn't take the test, but one thing I worry about is that the existence of mediocre fakes makes people more confident in their ability to spot fakes when in reality they might just not be catching the ones that are well-done. There's aren't many feedback loops in most people's lives whereby if someone believed something that was fake they'd be corrected and learn to be more skeptical.
>one thing I worry about is that the existence of mediocre fakes makes people more confident in their ability to spot fakes when in reality they might just not be catching the ones that are well-done.
Yeah, most of the "fake" audio clips were blatantly obvious to literally everyone with a functioning brain (and hearing ability I guess) and if you thought that represented anywhere near the state of the art in faking voices you'd be far more confident than you deserve to be.
If I hadn't already heard very good audio deepfakes I would have left the survey with far less concern than I presently have around the issue. I came in expecting that the quality of the fakes would get far more convincing over time and it really didn't, but the capability definitely exists (especially around public figures with much audio available for training)
Plot twist: 100 % of the material here was faked, only 50 % well done and 50 % not so well done! Now that's a test of how well people actually recognise this stuff.
