The most prestigious law school admissions discussion board in the world.
|
 |
 
|
The reason AI deepfakes are never more than 10 seconds long
Poast new message in this thread
Date: March 15th, 2026 1:39 PM Author: Jared Baumeister tancredi marchiolo is too stupid to guess (http://www.autoadmit.com/thread.php?thread_id=5845891&forum_id=2\u0026show=posted#49745467) |
Date: March 15th, 2026 1:39 PM Author: tancredi marchiolo ur so smart im sure u will tell us (http://www.autoadmit.com/thread.php?thread_id=5845891&forum_id=2\u0026show=posted#49745468) |
 |
Date: March 15th, 2026 1:41 PM Author: Jared Baumeister You're an expert at using Stable Diffusion, the answer should be OBVIOUS to you. But it's not (http://www.autoadmit.com/thread.php?thread_id=5845891&forum_id=2\u0026show=posted#49745474) |
|
Date: March 15th, 2026 1:41 PM Author: tancredi marchiolo im glad ur smart enough to tell us (http://www.autoadmit.com/thread.php?thread_id=5845891&forum_id=2\u0026show=posted#49745475) |
|
Date: March 15th, 2026 1:42 PM Author: Jared Baumeister Pop quiz: what increases along with video length? (http://www.autoadmit.com/thread.php?thread_id=5845891&forum_id=2\u0026show=posted#49745478) |
|
Date: March 15th, 2026 1:43 PM Author: Jared Baumeister Wow this is an RSF-free thread now (http://www.autoadmit.com/thread.php?thread_id=5845891&forum_id=2\u0026show=posted#49745481) |
|
Date: March 15th, 2026 1:44 PM Author: tancredi marchiolo im sure nsam will let us know any day now (http://www.autoadmit.com/thread.php?thread_id=5845891&forum_id=2\u0026show=posted#49745488) |
|
Date: March 15th, 2026 1:47 PM Author: Jared Baumeister I'm sure you'll find a 2 minute deepfake any day now and prove me wrong 🤣🤣🤣 (http://www.autoadmit.com/thread.php?thread_id=5845891&forum_id=2\u0026show=posted#49745495) |
|
Date: March 15th, 2026 1:54 PM Author: tancredi marchiolo ive already made way longer than 2 minutes are u dumb?? (http://www.autoadmit.com/thread.php?thread_id=5845891&forum_id=2\u0026show=posted#49745521) |
|
Date: March 15th, 2026 1:58 PM Author: Jared Baumeister 9 seconds:
https://x.com/i/status/2033199387908698178 |
|
Date: March 15th, 2026 2:02 PM Author: norwood ultra presumably each frame of a convincing video needs to be consistent with all the other frames (not just the immediately preceding frame) and that makes longer videos dramatically more computationally intensive (http://www.autoadmit.com/thread.php?thread_id=5845891&forum_id=2\u0026show=posted#49745547) |
|
Date: March 15th, 2026 2:10 PM Author: Jared Baumeister Yes but that's not the only reason. (http://www.autoadmit.com/thread.php?thread_id=5845891&forum_id=2\u0026show=posted#49745578) |
|
Date: March 15th, 2026 2:11 PM Author: norwood ultra what's the other reason (http://www.autoadmit.com/thread.php?thread_id=5845891&forum_id=2\u0026show=posted#49745583) |
|
Date: March 15th, 2026 2:17 PM Author: Jared Baumeister The likelihood of there being AI artifacts in just a ten second video is already high, and you have to police it closely to detect them. When you generate a two minute video, the likelihood of there being at least one bad artifact is effectively 100%, and your likelihood of catching it is lower because the video is so long. There will also be countless minor artifacts and you'll never catch all of them.
Furthermore, if/when you find artifacts in your 2 minute video, you have to run the whole process again and wait for the AI to make another two-minute video, with fingers crossed that it will have zero artifacts this time (PS it will still have dozens). Bottom line is you'll never have confidence that your two minute video is good enough to fool the masses |