Wrote a letter/novella of three years of AI thoughts wtaysom.github.io/imitation-engi… presents a cross between science fiction and victorian "science" demonstration with the real, surreal, magically real, and magical all braided together with little comic skill and full earnestness.
Finally finished my explainer on why LLMs can in fact learn world models, and why larger neural nets generalize better than smaller ones (against all theoretical predictions) youtu.be/UKcWu1l_UNw?si…
@pedroth9 Great article! I wish I know about it when I was scripting my video, it would have made it a lot easier haha. I knew this derivation was too simple to not have been discovered before, but I couldn't find it anywhere. Hopefully it becomes more popular.
I released my #SoMEpi entry: the simplest way to derive Taylor's polynomial. I haven't seen this derivation presented anywhere else before, and I think it's really neat, so I made it a video on it. After this I am back to machine learning content (Solomonoff induction next!)
@GregBlue123 I am releasing a video for SoMEpi very shortly, but it's not machine learning focused. Next machine learning video won't be for some time.
My latest explainer video on Mamba is out! Mamba is an exciting new neural net architecture that has the potential to replace transformers. Come learn how Mamba works from the basics. And don't worry, no state space theory required!
@praveenkumar_92 For example, CNNs are usually justified by saying that they make the model translation invariant, but locally connected nets are NOT translation invariant, they only reduce the input dimension, and they perform just as well (e.g. ieeexplore.ieee.org/document/91749…).
Just released a new video on how generative AI models work! Deep dive on auto-regression and diffusion and their advantages/disadvantages youtube.com/watch?v=zc5NTe…
@praveenkumar_92 I was teaching the ML course at my university. I thought the lecture slides from previous years did a really poor job of justifying why these methods work, so I tried to make better ones. The actual process was a combination of reading lots of papers and experimenting myself.
Just released a new video on how generative AI models work! Deep dive on auto-regression and diffusion and their advantages/disadvantages youtube.com/watch?v=zc5NTe…
Really appreciate @AlgorithmicSimp's ability to explain the big picture of how NN model's work. Other explanation's miss the forest for the trees.
youtube.com/watch?v=zc5NTe…
1 Followers 40 Following- programmer, but doesn't know any real languages
- too dumb to train foundation models, too stubborn to create a chatgpt wrapper
12K Followers 1K FollowingBuilding Humain Brain at @Humain |Ex- @Microsoft | EX-@awscloud | This account is a duet with my cool branding squad | https://t.co/3bI4YEcieR
36 Followers 51 FollowingEcology, evolution, and statistics | Professor, Institute of Freshwater Biology, Nagano University | PhD | OpenBSD user | ORCID 0000-0001-7464-0754
9K Followers 258 FollowingGeorge Lowther, Author of Almost Sure blog, on maths, probability and stochastic calculus.
Also on YouTube https://t.co/VyOijwbe9l
1K Followers 5K FollowingWhen poverty is rampant, and everything is regulated, everything is a crime, and everyone is a criminal. This means YOU.
Very, very anti-semetic.
11K Followers 7K FollowingVisiting fellow: Dept of Engineering Science - #UniversityofOxford and Course Director: Artificial Intelligence: views my own