@TheGlobalMinima - Study Path Through Top
@TheGlobalMinima on X The author says a site called paperswithcode.co, which tracks the most cited papers in modern generative AI, is the best way to learn the field. The pitch is to read the top 10 most cited papers at a pace of one or two per week, then reread, reimplement, and write up what you learned.
- Covers language modelling, inference optimisation, RL, system scaling, agentic AI, RAG, memory, and benchmarking in one place, which the author says is rare in nearly five years of generative AI.
- Argues these subfields overlap in practice: agents need system scaling, memory needs inference optimisation, and RL depends on understanding environments and benchmarks.
- Recommends a slow, repeated reading routine: one or two papers per week, reread, break down, code it, and write it back.
The post is a recommendation for a book (or resource) that the author says covers modern generative AI unusually well. Their claim is that it spans language modelling, inference optimisation, reinforcement learning and its methods, system scaling, applied topics like agentic AI, retrieval-augmented generation (RAG), and memory, plus environments and benchmarking. The author argues these areas blur into one another in practice: agents need system scaling, memory needs inference optimisation, and RL depends on understanding environments and benchmarks. Their point is that a single resource tying these together is rare in roughly five years of modern generative AI.
The second half of the post is a study plan the author attributes to @TheGlobalMinima. Go to paperswithcode.co, open the "most cited" list, and read the top ten papers at one or two per week. Read each one more than once, break it down, implement it in code, and write up what you learned. The author's case is that the most influential work of the last decade is concentrated there, so a disciplined pass through it builds a solid foundation.
Worth knowing before you save it: the post itself gives no title, author, publication date, or table of contents for the book, and the "paperswithcode.co" link is the author's pointer rather than a verified source. Paperswithcode.com was the original site, and it has since been folded into Hugging Face, so check the domain before you rely on it. The praise is also enthusiastic and unsupported by specifics, so treat the coverage claim as the author's opinion until you've looked at the actual table of contents yourself.