note to users
this is not a usual roadmap/guide/collection. these are links and things i've got from one and half years of scrolling searching and after overcoming gatekeeping instincts ive collated them from my DMs to myself. feel free to contact me wrt adding stuff or new sections you want. do not try to complete everything here.
deep learning foundations
core resources for understanding modern deep learning theory, transformer architectures, and fundamental concepts in neural networks.
books & comprehensive guides
- https://fleuret.org/public/lbdl.pdf little book of deep learning
- https://sebastianraschka.com/llm-architecture-gallery/ goated page for handy reference of all popular llm arch
transformer architecture & attention
- https://transformer-circuits.pub/ whole thread is goated, read if you're into mechint
- https://www.youtube.com/playlist?list=PLoROMvodv4rOY23Y0BoGoBGgQ1zmU_MT_ language modeling from scratch stanford
- https://www.youtube.com/playlist?list=PLoROMvodv4rOCXd21gf0CF4xr35yINeOy course on transformers and LLMs CME295 stanford
- https://medium.com/@milana.shxanukova15/100-nlp-questions-answers-attention-part-bf86e07251de NLP interview q
mathematics for deep learning
vision & multimodal models
- https://www.youtube.com/playlist?list=PLoHa6Id98MYIBpRz1NDFeW3JaYG5r1RvX lots of stuff about vlms
- https://arxiv.org/pdf/2602.20330v1 VLM circuit tracing
mechanistic interpretability
resources dedicated to understanding how neural networks actually compute, including feature circuits, mechanistic explanations, and interpretability techniques.
comprehensive guides
- https://dynalist.io/d/n2ZWtnoYHrU1s4vnFSAQ519J comprehensive guide to mechanistic interpretability by Neel Nanda
- https://docs.google.com/spreadsheets/d/1oOdrQ80jDK-aGn-EVdDt3dg65GhmzrvBWzJ6MUZB8n4/edit?gid=0#gid=0 Neel Nanda's 200 Problems on Mechanistic Interpretability
lecture series
exercises & practice
- https://transformer-circuits.pub/2021/exercises/index.html goated series of exercises for mechint
- https://arena-chapter1-transformer-interp.streamlit.app/ ARENA
specialized topics
- https://arxiv.org/pdf/2406.11717 refusal is mediated by a single direction (arditi et al.)
- https://www.neuronpedia.org/ neuronpedia (insane stuff tbh)
ai systems engineering
practical engineering approaches to scaling, training, and deploying large language models and ai systems.
training & scaling
- https://blog.yellowday.day/posts/gpt_oss_from_scratch/ gpt from scratch
- https://arxiv.org/abs/2603.07685 MoE scalable training
- https://djdumpling.github.io/2026/01/31/frontier_training.html frontier training methodologies
gpu & hardware
- https://www.kannav.dev/blog/blog_cinco primer on gpu internals
industry oriented stuff
- https://workatafrontierlab.com/ pretty self explanatory
courses & lectures
structured learning paths and comprehensive courses on ai, reinforcement learning, and generative models.
reinforcement learning
- https://www.youtube.com/playlist?list=PLqYmG7hTraZDM-OYHWgPebj2MfCFzFObQ silver's rl course op
- https://www.youtube.com/playlist?list=PLoROMvodv4rN4wG6Nk6sNpTEbuOSosZdX stanford's rl course
- https://huggingface.co/learn/deep-rl-course/en/unit0/introduction HF course on deep RL
generative models & diffusion
- https://diffusion.csail.mit.edu/2026/index.html diffusion course from mit
- http://kuleshov-group.github.io/dgm-website/ course covering deep generative architectures
foundational references
- http://www.incompleteideas.net/ Richard Sutton's incomplete ideas
recent papers & developments
cutting-edge research papers and novel approaches that shape the future of ai.
recent research
- https://arxiv.org/pdf/2603.14567 my senior's paper on an alternative approach to top-k
- https://subliminal-learning.com/ a cool method of teacher-student learning mechanism
emerging paradigms
- https://www.youtube.com/watch?v=rIwgZWzUKm8 world models 7 hour interview
- https://openai.com/index/parameter-golf/ self-explanatory
repos & datasets
- https://github.com/duoan/TorchCode Leetcode for Pytorch
- https://www.kaggle.com/datasets/parthplc/facebook-hateful-meme-dataset multimodal dataset for understanding harmful content detection (memes included, they're funny)
research methodology & writing
guidance on conducting research, reading papers effectively, and communicating findings clearly.
how to approach research
- https://www.youtube.com/watch?v=L4NnV_YpS2o how to get into frontier research, really nice tips
- https://youtu.be/nL7lAo95D-o yacine's guide on how to read papers
- https://how-to-get-so-creative-it-feels-illegal.tx how to get so creative it feels illegal
paper writing & communication
- https://nicholas.carlini.com/writing/2026/how-to-win-a-best-paper-award.html paper writing guide
- https://neelguha.github.io/blog/2026/templates/ Neel's template on ML papers
- https://www.youtube.com/watch?v=MfMq4sVJSFc Neel's interview