Oddbean new post about | logout
▲ ▼
10c87bb7 | 1 years ago (raw) | root | parent | reply | flag 
 @1273d431 X (Twitter) as well https://stackdiary.com/x-terms-of-service-for-ai-training-go-into-effect/
▲ ▼
gyptazy | 1 years ago (raw) | root | parent | reply | flag 
 @b53c11cd as I said, I don’t see any issues there. Everyone should be aware that public postings can be misused…

- Written things (blog, social media,…) for for text
- Photos (Facebook, instagram,…) for photos
- Podcasts and YT for voice cloning
- …

Recently, I copied a voice from YT tech talks of one of my mate (and my own one). I also gave it a try to create a stable diffusion model from a random person of instagram. 
See also: https://mastodon.bsd.cafe/@gyptazy/110936994541115009
▲ ▼
10c87bb7 | 1 years ago (raw) | root | parent | reply | flag 
 @1273d431 the future is going to be creepy.
▲ ▼
gyptazy | 1 years ago (raw) | root | parent | reply | flag 
 @b53c11cd indeed! And I’m not quite sure if it’s going into the right direction. From a technical aspect - yes it’s great. But it’s so easy to recreate photos and voices of people. We can adapt their pronunciation and art of speaking to make it more authentic.

More afraid I’m regarding deep fake videos including an adapted art of speaking and their voice. While this is still much work and the masterlevel, we will soon reach models for videos. At this point, I’m absolutely afraid of upcoming political fake stuff and created porn content. Last one is already a problem since it just needs some pictures of someone’s crush on Instagram…