TikTok machine learning
- Thread starter TimCook
- Start date
Android Tipster
Elite Member
- Sep 7, 2015
- 1,732
- 749
An interesting thing that happened yesterday when I uploaded to my tatts account a front-facing video of a full tattoo. TikTok flagged the video almost immediately. It took me a few seconds to even realize that the body part was a boob!
TimCook
Newbie
- Dec 26, 2020
- 21
- 7
There is no doubt that uniqueness of videos may not help 99% of the time and TikTok can find identical videos. Regarding body parts, colours are (probably) detected with high accuracy. At least flip the video 360 degrees, mirror it... Machine learning is highly likely to be able to detect the same videos on its platform.
- Jul 28, 2016
- 848
- 295
Also, kinda related. With the recent changes in tiktok algorithm, it's harder to just paste videos from yt/ig, one of the methods is to apply some filters and flip the video horizontally, what about stickers:
1) does applying some stickers on the video would help against shadow bans?
2) would it make harder for algorithm to detect content from other platforms?
1) does applying some stickers on the video would help against shadow bans?
2) would it make harder for algorithm to detect content from other platforms?
nguyen quang vinh
Newbie
- Oct 18, 2021
- 4
- 1
Maybe
Juherianto
Regular Member
- Oct 25, 2020
- 492
- 351
Wow Tik Tok having face recognition also!
- Jul 28, 2016
- 848
- 295
You'd be surprised how much data they collect from their users. No wonder they have problems in the EU haha.
limitless1337
Junior Member
- Mar 2, 2016
- 175
- 67
They collect a lot of data, even your actions, what you do, what you watch, who you stalk, what you search , with such anything is possible. for example they already know what your schedule is, what you like to eat, what tv show is your fav. you name it.
miamitech
Newbie
- May 10, 2022
- 7
- 6
Yea. TikTok has the most advanced ML by far. It's incredible how accurate their "For You" page is. After a few hours they already know what you're into and the app has you hooked. It's similar to the Uber model where after two rides you're hooked for life. On TikTok, after two hours, you're hooked for life. They know what makes you laugh, what you're interested in, which videos you share with friends. who your friends are, people near you, your contacts, they have everything about you collected and fed into their AI algorithm. They're destroying every other platform.
321oiram
Newbie
- Apr 19, 2022
- 5
- 3
They are competing, and other platforms are trying to improve. These companies are not going to give up so easily.
All this competition will eventually give birth to Skynet. And the end of the whole species.
The few surviving humans will have the consolation of hiding in a cave and connecting to some old servers with Neuralink, trapped in an infinite loop, nothing new.
timothywcrane
Supreme Member
- Apr 25, 2009
- 1,339
- 742
I am interested in how they get the throughput out of the models they use. Finding video uniqueness can be done using crypto hashes to define uniqueness and perceptual hashing for similarity, but with the amount of videos they process, my question is how they get the throughput they do. Has to be more than horizontal scaling... right?