Reading list

Obsidian CEO on file-over-app, owning your data, durable note systems (vault-as-product + how-to-think)

the NotebookLM source→audio→extraction workflow; **action: figure out how to fold this into the vault ingest flow** (see actionable below)

clean first-principles treatment (Musk/Parrish framing)

George Mack: inversion as the single most powerful problem-solving move

reusable prompt that turns an agent into an incremental Socratic teacher that quizzes you to mastery (tooling for this lane)

7-chapter from-scratch guide: NNs → math → classical ML → hardware → transformers → vision → agentic

~10k-word intuition-level survey of modern deep learning (VLMs, encoders, architecture path-dependency)

rigorous from-scratch math + code for Rotary Position Embedding (HF Llama impl)

Gundersen (quant researcher): math fundamental, foundational read

AMS roundup (3B1B, Strogatz eigenvector intuition); math fundamental 2 (lighter/resource-list)

Fei-Fei Li / World Labs: world models as renderer/simulator/planner; spatial-intelligence thesis ("prolly a later read")

CMU SEI primer on training hardware (GPU/memory/NVLink/multi-GPU) — slightly dated

Illustrated Transformer, LLMs-from-scratch, Chinchilla, ReAct, MemGPT, DreamerV3, Cosmos, OpenAI Evals

35 RL Qs (PPO/GRPO/DAPO/GSPO, credit assignment, vLLM/SGLang) — an RL-interview benchmark

the 6 areas modern AI/ML interviews span; don't skip classical ML

12-stage roadmap — **skip stages 1-3**, use as a checklist for the rest

EXPLORE: group theory + positional encodings, reverse-engineering an NN, visualizing piecewise-linear NNs

Vivek: labs hire for building useful abstractions / judgment under incomplete info ("a race for better abstractions, not intelligence")

forkable Claude Code job-application pipeline (/setup, /scrape, /apply drafter-reviewer, /upskill) — **evaluate adopting for the campaign**

what gets noticed vs ignored on the recruiter side

best opener (per the learning map)

Anthropic eng (primary)

Anthropic eng (primary)

PostHog: the one practical-lessons keeper

Primary: 20 improvements, 11% speedup

Primary: 3-file architecture

best analytical breakdown

Anthropic eng

SOUL.md/MEMORY.md/DREAMS.md (vault-as-product)

first-principles walk through agent memory

Shpigford: USER.md + MEMORY.md (for David)

markdown knowledge graph + MCP server (directly vault-relevant)

server design, OAuth, vault patterns

Anthropic eng

Lilian Weng (canon-grade)

non-dev built full CoS in 36 hours (vault-as-product)

Unsloth practical guide

explainer for the MIT RLM paper

Wiecki, intuitive Metropolis-Hastings

Wiecki, dependence beyond correlation

Monte Carlo, Kelly, copulas with code

IR = IC * sqrt(N)

daily pre-market for Holding

Google's TimesFM

30 lessons (Holding)

Hashimoto on building blocks vs apps

Dorsey + Botha: Block's AI-native org

Aschenbrenner: AGI timeline (must-read)

5 moats that survive AI

Paul Graham

SF social capital, practical

SF landing (settling-in)

whole company on one laptop (vault-as-product)

Dan Koe on generalism (the compounder shape)

"you can just build things"

Sentra founder on enterprise memory → company world model. ("Ashwin knows his shit — question is whether it's relevant to me rn.")

Obsidian + CC second-brain build (4 skills). ("Hype or legit? could be garbage or valuable" — read with skepticism, it overlaps your own setup.)