Eleven long-form digests · Updated continuously

Learning Engine

Every corpus we index — conference keynotes, founder interviews, podcasts, on-chain activity, and App Store movements — distilled into a navigable digest you can skim in minutes. Roughly 1,000 hours of source material turned into structured reads, so you get the signal without the watch time.

11
Digests
~900h
Watch-time saved
~290
Recordings indexed
7
Bilingual (EN / 中文)

Courses & lectures

External
336

CS336 lecture notes — Language Models from Scratch

Stanford CS336 · 17 lectures · one navigable note per lecture

Structured lecture notes for Stanford's CS336, “Language Modeling from Scratch” — the full build-it-end-to-end course: tokenization, architecture, training, systems and kernels, parallelism, scaling laws, data curation, alignment, and inference. Each lecture is distilled into a navigable note with the key derivations, diagrams, and takeaways.

17 lectures Stanford CS336 GitHub Pages

Market intelligence

New

App Store ecosystem dashboard

Daily snapshot · 220+ apps tracked · public Apple RSS / Search / Reviews APIs

Daily snapshot of relevant App Store activity: top charts in 7 categories, competitor app rankings, and ~150 most-recent reviews per tracked app. Free Apple-public-API alternative to Sensor Tower / Apptopia. Tracks Hyperliquid dashboards, perp trading, on-chain analytics, prediction markets, AI research apps. Append-only history file lets us trend ranks over time.

220+ apps 7 categories $0 cost Updated daily

On-chain & markets

New
HL

Hyperliquid on-chain digest

May 2026 · HLP / HYPE / AF / vaults / funding / fees / moat

A 9-section working note on the most successful on-chain perpetual exchange to date. HLP P&L decomposition ($287M cumulative, −3.81% max DD), HYPE tokenomics, the Assistance Fund's $751M/yr automated buyback, full fee structure and ~$1.4B/yr revenue reconciliation, vault ecosystem (9,452 vaults, Gini 0.965), HL vs Binance vs CME head-to-head, funding-clamp mechanics, US compliance, and the 7-decision strategic moat.

9 sections 5 live charts 230 coins 9,452 vaults
μ

Crypto microstructure live dashboard

BTC across 5 CEXs · L2 order-book archive (2017–2023) · live research engine

Autonomous research engine output: 16 completed studies on BTC order-book structure across Binance, Coinbase, Kraken, OKEx, and Bitstamp. Lead-lag analysis, OFI predictive power, futures-spot basis dynamics, market-maker quote patterns, crypto-equity correlation against US equity returns. Deep-learning regime discovery on GPU.

16 studies 5 exchanges 13TB tick data 2 GPUs

Founders & long-form interviews

New
JY

Jeff Yan video archive

14 videos · multimodal AI · 2023–2025

Every public Jeff Yan (Hyperliquid co-founder) YouTube interview, fireside chat, and keynote, processed video-by-video with multimodal AI. Verbatim quotes with clickable YouTube timestamps, on-screen visual elements, novel insights, and host context. Covers TOKEN2049, KBW, Bell Curve, Flirting with Models, When Shift Happens, Risk On, and more.

14 videos 11.3h runtime 80+ quotes 56 insights
External
XJ

Xiaojun Podcast digest

Chinese AI long-form interviews · 1–7h per episode · bilingual EN / 中文

Xiaojun Podcast is one of the most influential Chinese AI long-form interview podcasts. Episodes run 1–7 hours with guests spanning international AI labs, Chinese LLM startups, VC, and application-layer founders. Each episode → 30–80 topic blocks with clickable YouTube timestamps, notable quotes, falsifiable predictions, and visual signals transcripts miss. Saining Xie (Ep 133), Shunyu Yao (Ep 140) and more.

2+ episodes EN/中文 bilingual By guest & topic GitHub Pages

Conferences & keynotes

External
CV

CVPR digest — 235 talks indexed

CVPR 2024 + 2025 · 527 hours of source video · ~1,400 named methods

A navigable index of every CVPR 2024 and 2025 talk recording we could find. Per video: speaker list with affiliations, talk-by-talk breakdown with clickable YouTube jump links, abstract, key takeaways, and every named model/dataset/method that appears on screen. Cross-reference indexes across methods (every video where CLIP or GAIA-1 shows up) and speakers.

235 talks 527h source video 1,400+ methods GitHub Pages
External
I/O

Google I/O digest 2024 + 2025

15 keynotes · 7.4h · bilingual EN / 中文 · the "announce → prove" shift

Segment-by-segment breakdown of every Google I/O 2024 and I/O 2025 keynote with clickable YouTube timestamps. Quantitative diff: 2024 announced 89 products with 6 on-screen benchmarks; 2025 announced 57 products with 16 benchmarks — messaging shifted from "look at all the new things" to "look at how much better the existing things are." LMArena Elo 1448, WebDev Arena +142, USAMO 2025 49.4%, LiveCodeBench v6 80.4%.

15 keynotes 7.4h source EN/中文 GitHub Pages
External
GTC

NVIDIA GTC digest 2018–2026

29 keynotes & special addresses · 8-year arc · bilingual EN / 中文

Bilingual structured index of 29 NVIDIA GTC keynotes spanning V100 (2018) through Vera Rubin Ultra (2026). The 8-year arc: DGX/Tesla V100 → A100/Ampere/Omniverse → Hopper/H100 → Blackwell B100/B200/ NVL72 → Vera Rubin with Nokia 6G, Foxconn robot factory, Uber robotaxis, sovereign AI. CUDA-X, AI Factory, Quantum NVQLink. Every claim cites [video_id @ timestamp].

29 keynotes 8yr arc EN/中文 GitHub Pages
External
CN

Google Cloud Next digest 2024 + 2025 + 2026

32 keynotes · 11.2h · three-year competitor-mention evolution

Cloud Next 2024 (2.8h, 2 videos), 2025 (2.3h, 7 videos), 2026 (6.1h, 23 videos). The competitor-mention evolution: 2024 named Zoom/Teams/Webex; 2025 named Anthropic, Meta Llama, OpenAI, Mistral, Snowflake-Databricks; 2026 was the first time AWS was named explicitly on stage. Same keynote announced "Rapid Enterprise Migration — Microsoft 365 to Google Workspace 5x faster." Bilingual EN / 中文.

32 keynotes 11.2h source 3 years GitHub Pages

AI infrastructure & economics

External
∞

AI infra pricing & memory digest

39 long-form videos · pricing, HBM, inference economics · ~9.4M tokens extracted

Bilingual structured index of 39 long-form videos on AI infrastructure pricing, memory (HBM/DRAM), bottlenecks, and inference economics. Headline: inference cost fell 1000× in 3 years ($400 → $0.40 per million tokens), but memory just became the bottleneck that decides whether that drop continues. 30% of Big Tech's 2026 capex is going to memory; HBM capacity sold out through 2026. Every claim cites a specific [video_id @ timestamp]. No fabricated numbers.

39 videos ~9.4M tokens EN/中文 GitHub Pages