Shelf
Articles, videos, repos, and papers I recommend: a running list, not a ranked catalog.
Nothing in this category yet.
Reading now
- VIDEOKarpathy: Let’s build GPTReading
Build a tiny GPT from scratch in an afternoon. Pairs well with your Vanilla Transformer repo.
- BOOKDesigning Data-Intensive ApplicationsReading
Slow read, but the storage/replication chapters stick forever.
- PAPERQLoRA: Efficient Finetuning of Quantized LLMsReading
Quantized LoRA adapters: fine-tune big models on a single GPU without full-weight copies.
- PAPERRLHF: InstructGPT paperReading
The InstructGPT paper: SFT plus human preference ranking, the blueprint behind chat-model alignment.
- PAPERAttention Is All You NeedFinished
The transformer paper. Still the clearest mental model for why modern LLM stacks look the way they do.
- VIDEO3Blue1Brown: But what is a neural network?Finished
Best intuition-first intro before you touch backprop math.
- ARTICLEThe Missing Semester of Your CS EducationFinished
Shell, git, vim, debugging: dense and practical.
- PAPERNeural Machine Translation by Jointly Learning to Align and TranslateFinished
Attention before transformers: align-and-translate with a soft search over source states.
- PAPERSequence to Sequence Learning with Neural NetworksFinished
Classic seq2seq with LSTMs: the baseline story before attention took over.
- PAPERUniversal Language Model Fine-tuning for Text ClassificationFinished
ULMFiT: pretrain a language model, then fine-tune with discriminative rates and slanted triangular schedules.
- ARTICLEPaul Graham: How to Do Great WorkQueued
- ARTICLEAnthropic: Core Views on AI SafetyQueued
Useful framing when you are explaining RLHF work on a resume.
- ARTICLEStreamlit docsFinished
Quick reference when shipping ML demos with Streamlit.
- VIDEONeural Networks: Zero to Hero (playlist)Queued
Karpathy playlist: watch in order if you want the full arc.
- PAPERChinchilla: training compute optimal modelsQueued
Why “bigger is always better” stopped being the whole story.
- BOOKThe Pragmatic ProgrammerFinished
Classic craft book: DRY, tracer bullets, rubber ducking.
- ARTICLEVercel: Astro deployment guideFinished
- VIDEOAndrej Karpathy: State of GPTFinished
Broad survey talk: good before diving into a new architecture paper.