Chaia Ella
Region: US
Wednesday 17 November 2021 01:28:20 GMT
44031
8687
334
196
Music
Download
Comments
The Lich of Westmire :
I keep ending up on spider Tik Tok somehow despite always clicking non interested and blocking every spider post I see 😭😭😭 help me 😭😭😭
2021-11-17 02:11:38
165
shy by sunshine :
the algorithm is showing me critical analysis of the algorithm. by the correlation of our poor self esteem it and I are now friends
2021-11-17 06:40:17
346
David.ai :
One thing to consider- ML is not JUST deep learning. Likely, tiktok uses a “ensemble algorithm”. Some of the feeders to this will be explainable.
2021-11-17 12:44:41
88
BeerandCircus :
my fav anecdote of accidental bias was a self-driving car that learned driving based on proximity to grass. almost crashed when first saw a bridge.
2021-11-17 03:09:11
66
Emma :
cpg grey did a great video about this, while a individual line can be understood, and clusters general purpose vaguely grasped, the whole is beyond
2021-11-17 02:03:26
47
Chris :
Wrote my thesis on deep recommendation systems, they generally don't work the way you're describing, so the dangers are subtly different
2021-11-17 04:59:00
35
ace_of_bass :
great video! I do want to say that it is possible to analyze these networks, it's just that engineers care much more about results than understanding
2021-11-17 02:58:46
30
Ian :
Machine learning is absolutely fascinating. Also, more than a little spooky
2021-11-17 04:00:43
21
anti carson :
i always appreciate your communication skills with this sort of stuff. very engaging and effective!!
2021-11-17 02:12:28
20
FiercePufferFish :
FB's algorithm is an extremely successful predictor of manic episodes.
2021-11-17 19:46:56
19
cat fangs :
just and a console.log 🤪
2021-11-18 04:23:27
11
Michael Berlinger :
True but any brain (artificial or not) can be described the same way. Try and explain what a color is to a cat for example...
2021-11-26 05:34:55
10
Quinn :
On this topic I can only recommend the art, talks and books by James Bridle, they're a genius when it comes to making these abstract concepts graspabl
2021-11-17 01:37:27
8
EcoEngineer :
of course you can answer these questions. you do an ablation study. you remove one input/component at a time and monitor the results.
2021-11-17 07:31:30
7
Beatrice :
I also have this theory that ADHD and other neurodivergent neurotypes get shown videos based on the way they interact with videos.
2021-11-17 02:22:50
6
Hakunyaa :
I think the best way I've seen of describing this is ai (and its associated imitations) don't have morals, they have objectives
2021-11-17 05:45:00
6
Kelflcwlalwo :
Tell me you're not a ML engineer without telling me you're not a ML engineer
2021-11-18 07:15:00
4
left n depressed :
ml systems are just a mystery box with "number go up" haphazardly painted on the side
2021-11-17 03:33:54
3
Serena Aguirre :
me watching this wondering how long it will take before I’m on machine learning TikTok … is this the metaverse 🧐
2021-11-17 07:13:20
3
Ash :
my dad loved ai and he told me a story of some people trying to use ai to spot tanks/military stuff hidden in photos. the ai turned out to be super +
2021-11-17 08:04:59
3
JessieW :
I need to send this video to my therapist because this is exactly my fear and it feels like forbidden knowledge
2021-11-17 02:51:44
3
Someone Is Wrong :
Yup, spoke to a YouTube engineer once & their platform is like this too. Got the impression he tried not to think about it too much cos it scared him
2021-11-17 14:43:06
2
Phill :
The "black box" that is machine learning is honestly so fascinating... But like... Eerie 👀
2021-11-17 02:45:33
2
AmadiTalks אמדי :
Plus we know that they have put a thumb on the scale in the past, they admitted to hiding disabled, Black, fat &queer users’ videos to “avoid abuse.“
2021-11-17 03:55:19
2
xopherx :
This is a very good explanation.
2021-11-17 01:40:08
2
DemiGoose :
They absolutely do watch time, a lot actually. Make a new account and don’t like anything, but pick one type of video to watch longer
2021-11-17 08:50:39
1
Roy C :
putting it another way, we can always ask how often person X will get video Y. but the algorithm can't tell you why they chose that
2021-11-25 07:32:41
1
John Mammal Enjoyer :
Get the incense! we must convince the machine spirit to not be bigoted!
2021-11-17 06:23:17
1
I Am Dead Too :
I think we'd have much more interpretable ML if the biggest tech companies had incentive (or requirement) to do so
2021-11-17 05:32:57
1
Elliot Rayne :
Terrifying!
2021-11-17 01:38:02
1
Emilio Bedelio :
Aaaaaaaaaaaaaahhhh
2021-11-17 01:46:23
1
FastestDrawInTheEast :
I am very confident that it looks at not just how fast you swipe, but where on the screen and with which hand.
2021-11-17 09:11:15
1
Julie Karasik :
The algorithm doesn't care about anything other than keeping eyeballs on the site, and it will do anything it can to accomplish that goal.
2021-11-20 00:04:14
1
Ducky Bimbo :
Some work is being done on this. This who are interested can look up "explainable AI" and "Trustworthy Autonomous Systems". More to do though!!
2021-11-17 09:26:03
1
smeezer :
I stay away from my fyp. The algorithm thinks I should be anti vax, religious and into diesel trucks. Which would describe significant swatches of...
2021-11-17 02:32:44
1
Pascal :
This is true, but we could use multiple regression analysis to get some idea of what matters
2021-11-18 04:07:13
1
FaerieMachinist :
This is one of those terrifying things in my brain's "here's why you can't sleep" Playlist.
2021-11-19 00:58:07
0
lolcopeharder :
is the analogy of ML models being a "black box" correct?
2021-11-18 13:38:43
0
michelle_your_belle :
i think about this constantlyyyyyy.
2021-11-18 05:12:02
0
lolcopeharder :
can you ever export weights and biases and learn from them?
2021-11-18 13:39:28
0
Paul Lawniczak :
I welcome the comming of our new robot overlords
2021-11-18 19:38:04
0
gulfcoastyachtpros :
I think an advanced ai could contemplate morality
2021-11-18 23:41:24
0
Diego :
that's why explainable ai is a hot topic right now
2021-11-19 04:24:19
0
Mike :
Maybe Frank Herbert was right and we need to jihad the machines before it's too late
2021-11-19 17:53:45
0
lithuinsbane :
Not all ML is neural nets. Some of the algorithms are easily shown mathematically.
2021-11-19 16:57:48
0
Feigen :
Weapons of Math Destruction would be a great book for this audience to read
2021-11-19 14:45:43
0
anisotropic.reaction :
I've been working on an unsupervised ML project for a bit, and each time I train from scratch at some point it's output sometimes surprises me 1/2
2021-11-19 02:03:00
0
MarcKS :
Amoral machines
2021-11-19 05:25:00
0
To see more videos from user @butterflyastrophysics, please go to the Tikwm
homepage.
