GPT-3 Paper Explained: How "Language Models Are Few-Shot Learners" Changed Everything
GPT-3 paper explained: OpenAI's 175B-parameter model reached 71.2% on TriviaQA few-shot with no fine-tuning. What held up, what did not.
AI research decoded: the papers, the datasets and the people behind them.
Researchers, PhD students and engineers working on frontier projects.
The 8 Google Researchers Who Invented the Transformer: Where Are They Now?
Where the eight Attention Is All You Need authors are as of October 2026: Cohere, Sakana AI, Essential AI, NEAR, Inceptive, and Shazeer's move to OpenAI.
InstructGPT and RLHF Explained: The Paper Behind Human-Feedback Fine-Tuning of GPT-3
InstructGPT explained: how RLHF let a 1.3B model beat 175B GPT-3 in labeler preference, what the paper measured, and where the evidence is thin.
Chinchilla vs Kaplan Scaling Laws: Were the First LLMs Trained Wrong?
Chinchilla vs Kaplan scaling laws: compute-optimal model size grows as C^0.50 in one paper and C^0.73 in the other. Here is why they differ.
Attention Is All You Need Explained: The 2017 Google Paper Behind the Transformer
Attention Is All You Need explained: how the 2017 Transformer works, its 28.4 BLEU result, what the ablations show and where the evidence is thin.
DeepSeek-R1 Paper Explained: How Reinforcement Learning Alone Taught AI to Reason
DeepSeek-R1 paper explained: pure RL lifted R1-Zero from 15.6% to 71.0% on AIME 2024. What the results show, and where the evidence is thin.
Scaling Laws Explained: What the OpenAI Paper Showed, and What Later Work Corrected
Scaling laws explained: Kaplan et al. (2020) fit loss to power laws across seven orders of magnitude, then Chinchilla changed the data-to-model split.