Man watching your videos have helped me broaden my perspective of the concepts at play and have allowed me to create new and useful things versus regurgitating the same product repeatedly. thank u g
just gotta harvest a bunch of good commits with good comments. after the llm is trained you could do like a dpo finetune sorta thing where you ask the model to tell the difference between each commit edit for githu thing
Regarding the multi-resolution approach for text. I was pondering something similar to myself the other day, considering the granularities of char, phoneme, word, Latin style fusions, phrases, sentences, paragraphs. I'm curious how it would compare to a multi resolution transformer that started with a sequence of small (overlapping?) context windows that merged into a full sized context window over a series of layers.
![일상 속 기분 좋음 [Proto disco meme]](http://i.ytimg.com/vi/DwmEIaZ7XGQ/mqdefault.jpg)
Man watching your videos have helped me broaden my perspective of the concepts at play and have allowed me to create new and useful things versus regurgitating the same product repeatedly. thank u g
Very nice, will be helpful for my bachelor thesis. Thanks!
This channel needs way more views
Wow these papers are hella brand new!
right?!?!
ASR automatic speech recognition
Continual training and model merging sounds like a natural way to have continually improving models. That adapt to user requirements.
Superbbb!!!!!!!!!!!!
just gotta harvest a bunch of good commits with good comments. after the llm is trained you could do like a dpo finetune sorta thing where you ask the model to tell the difference between each commit
edit for githu thing
Regarding the multi-resolution approach for text. I was pondering something similar to myself the other day, considering the granularities of char, phoneme, word, Latin style fusions, phrases, sentences, paragraphs. I'm curious how it would compare to a multi resolution transformer that started with a sequence of small (overlapping?) context windows that merged into a full sized context window over a series of layers.
Wow that domain adaption paper is literally my phigments 3b paper 😅
How far out are we from having ai systems that can automatically implement/verify reported on breakthroughs in papers?
"Guys, we should do another blockchain!"
Crypto bros gonna be so mad 😂
There is a redundant chapter (3:36) for a paper that (I think) was not included, making all the later chapters one out of sync
ah it’s an automated script that writes the timestamps and that happens sometimes
@Tunadorable You should fix that, the timestamps are crucial to watch the interesting ones.
Oi, what's up with ai today?
nuttin much hbu?
oi