Categories
2026
LLaVA: How a Small Bridge Lets a Language Model See
LoRA: Teach a Model Without Rewriting the Whole Thing
Speculative Decoding: Guess Ahead Without Changing the Answer
PagedAttention: When a Slower Kernel Makes the System Faster
DPO: Where Did the Reward Model Go?
FlashAttention: Move Less, Not Compute Less
Mamba: Why Remembering Less Can Understand More
DeepSeek-V2/V3: The Two Bills Behind MLA and MoE
LLaMA: No Architecture Revolution, Yet It Became the Open-Model Foundation
Chinchilla: Why Bigger Models Are Not Always Better
InstructGPT: Teaching LLMs to Be Helpful with RLHF
T5 / BART: Unifying Every NLP Task as Text-to-Text
GPT-3: In-Context Learning and the Few-Shot Explosion
DeepSeek-R1: Eliciting Reasoning in LLMs via Reinforcement Learning
Unix Domain Socket Source Code Analysis (OLK 6.6)
GPT-1/2: Generative Pretraining and the Seeds of Few-Shot
containerd Communication Architecture
BERT: What Exactly Is 'Bidirectional'?
Transformer: The Starting Point of Everything — What Problem Did It Actually Solve?
