34K ·GitHub
What is this?
It's a curated reading list for deep learning: each entry is a recorded lecture that walks through a paper paragraph by paragraph, from classics like CLIP and GPT-3 to recent ones like Llama 3.1 and Sora.
It's a curated reading list for deep learning: each entry is a recorded lecture that walks through a paper paragraph by paragraph, from classics like CLIP and GPT-3 to recent ones like Llama 3.1 and Sora.Is this a list of papers I read, or videos I watch?Videos, not notes or code — the README is a table of lectures with date, title, runtime and a link, e.g. a 1:38:25 CLIP session or a five-part Llama 3.1 series.
Videos, not notes or code — the README is a table of lectures with date, title, runtime and a link, e.g. a 1:38:25 CLIP session or a five-part Llama 3.1 series.Do I just follow links from the README, or is there anything to install?Nothing to install: open the README, pick a paper from the table — sorted newest-first with runtimes shown — and click through to the video.
Nothing to install: open the README, pick a paper from the table — sorted newest-first with runtimes shown — and click through to the video.What's the catch before I commit a couple of evenings to this?The lectures are in Mandarin, and many sessions run 50-90 minutes — plan on roughly an evening per paper.
The lectures are in Mandarin, and many sessions run 50-90 minutes — plan on roughly an evening per paper.Do I need a deep-learning background to follow along?Yes — it goes straight into landmark papers like Pathways, Megatron LM and GPT-4, so basic deep-learning knowledge is assumed; it fits grad students, researchers and engineers who want dense papers walked through.
Yes — it goes straight into landmark papers like Pathways, Megatron LM and GPT-4, so basic deep-learning knowledge is assumed; it fits grad students, researchers and engineers who want dense papers walked through.What language are the lectures in?Mandarin Chinese — the table titles are in Chinese and the sessions are spoken lectures, e.g. 'GPT, GPT-2, GPT-3 精读'. Non-Chinese speakers will hit that wall immediately.
Mandarin Chinese — the table titles are in Chinese and the sessions are spoken lectures, e.g. 'GPT, GPT-2, GPT-3 精读'. Non-Chinese speakers will hit that wall immediately.What's on the list besides paper walkthroughs?A four-part research-craft series (研究的艺术) covering how to connect with readers, pick important problems, tell a good story and build claims-evidence-warrants, plus one-offs on doing research in the LLM era and a small video-editing tool demo.
A four-part research-craft series (研究的艺术) covering how to connect with readers, pick important problems, tell a good story and build claims-evidence-warrants, plus one-offs on doing research in the LLM era and a small video-editing tool demo.What's the newest addition?A roughly hour-long OpenAI Sora walkthrough dated 1/10/25 that also covers Movie Gen and HunyuanVideo; just before it, the five-part Llama 3.1 series ran from late July to early September 2024.
A roughly hour-long OpenAI Sora walkthrough dated 1/10/25 that also covers Movie Gen and HunyuanVideo; just before it, the five-part Llama 3.1 series ran from late July to early September 2024.What are similar projects?Motus2, gpt-2, gpt-3, WholebodyVLA, RoboICL, DALLE2-pytorch, Awesome-LLM, math
Motus2, gpt-2, gpt-3, WholebodyVLA, RoboICL, DALLE2-pytorch, Awesome-LLM, mathWhich topics is it in?llama, whisper, paper, gpt, multimodal, transformer, artificial-intelligence, video, computer-vision, openai, curated-list, deep-learning
llama, whisper, paper, gpt, multimodal, transformer, artificial-intelligence, video, computer-vision, openai, curated-list, deep-learning