Comments by "Gilad Barlev" (@GSBarlev) on "" video.

One thing I cannot stand about myself: whenever a person in a video is wearing a shirt where part of the text or logo is obscured, I spend the rest of the video trying to figure out what it says. So that was my main stressor this evening. 😄
25
Good news on the art front: first copyright court case went the creators' way! Also doesn't look good for Zuck admitting he personally approved using a database of pirated content to train LLAMA.
11
 @captainkesh If you have friends who are still Muskovites, even after all of this... are they still your friends?
9
"Another video on AI that makes you look kinda clueless." <Wall of completely off-topic ranting> "I haven't watched the full vid yet though" ::chef kiss::
8
@lionelmessisburner7393 You're off here ranting about Alpha Fold and deep learning when the video was specifically about Transformers. Comparing the transformer architecture to RNNs is like comparing a squirrel to a bat—yes, both are small mammals that live in trees, but only one of them is capable of flight. Not only that, the video was focused hyperspecifically on one particular article that could have easily been written by Google's PR team! But—bottom line—you conclude she doesn't know what she's talking about when you clearly didn't listen to a word she said!!
4
@chatzehus Omg thank you. That totally lines up.
4
@potassiumyttriumsulfur5013 You are wildly misinformed—none of what you wrote is true. Read "Attention is All You Need."
1
@potassiumyttriumsulfur5013 Alpha Fold is a DeepMind model. Deep learning is an unsupervised approach utilizing recurrent network layers. Transformers omit the recurrent layers in favor of more "attention." They also must be trained using prelabeled datasets. These are foundational distinctions.
1
@potassiumyttriumsulfur5013 AlphaFold uses transformers, but they are not part of the core architecture. See the article: "AlphaFold is not an LLM."
1
This video isn't about AlphaFold, honey. DeepMind is an extremely complex recurrent neural network model framework that is capable of unsupervised learning and—yes—novel insights. Transformer models came out of a 2017 paper called "Attention is All You Need" that showed how you could build a chatbot cheaply (without those extremely complex recurrent layers) that could be pre-trained on a corpus of documents. Angela isn't saying "AI" isn't useful for science—anyone who's ever done anomaly detection or optical pattern recognition knows how useful it is—she's saying that these models cannot science. Because they can't. Because that's not how they work.
1
How close was she?
1