The white papers that actually moved the field, each linked to the original and paired with my plain-language summary, a glossary of the terms, and what it changes in practice.
01
Vaswani, Shazeer, Parmar, Uszkoreit, Jones, Gomez, Kaiser & Polosukhin — Google Brain / Google Research
The 2017 paper that replaced word-by-word reading with “attention” and, in doing so, quietly invented the architecture behind every large language model since.
02
Wei et al.
Asking a model to show its work raised PaLM from 17.9% to 58.1% on GSM8K - and only above roughly 100B parameters.
03
Lewis et al.
Trained the retriever and the generator together. Narrower, and more interesting, than what "RAG" came to mean.
04
Hu et al.
Fine-tune two small matrices instead of the whole model, and swap adapters at serving time.
05
Hoffmann et al.
The field had been building models too large and training them on too little data.
06
Yao et al.
The thought/action/observation loop that every agent framework now implements.
07
Rafailov et al.
Collapsed the two-stage RLHF pipeline into a single loss, with no reward model to train.
08
Dao et al.
Attention was memory-bound, not compute-bound. Tiling it into fast memory is why long context got cheap.
09
Ouyang et al.
Turned a next-token predictor into an assistant, and found a far smaller aligned model was preferred over a much larger unaligned one.
Summary
All original papers remain the work of their authors and link to the official source (arXiv or publisher). Summaries and glossaries on this site are my own commentary.