🌊 Tide

English | 中文

🌐 Read today's tide: longqibyte.github.io/Tide — updated weekly, fully automated.

Information arrives like the tide — wave after wave, never stopping. You cannot hold back the ocean, but you can learn to read it.

Tide is an open-source AI information website: every day it ingests the sources that matter (arXiv papers, Twitter/X voices, AI lab blogs…), selects the waves worth reading, and publishes deep interpretations — not one-line summaries — on a web page anyone can open and read. A daily tide report for people who want to actually understand what's happening, not just scroll past it.

Why "Tide"?

News, papers, and hot takes come in daily like the tide. Most aggregation tools try to drink the ocean: collect everything, compress it into bullet points, and call it a day. The result is a feed you skim and forget.

Tide takes the opposite stance. The tide will come regardless — the question is not "how do I see everything?" but "what do I let shape my thinking, and how deeply?". Tide watches the water for you, picks out the waves worth studying, and helps you engage with them seriously.

The Core Belief: Understanding Cannot Be Outsourced

This project is built on one idea, borrowed from Andrej Karpathy:

"Understanding cannot be outsourced."

In the AI era, it has never been easier to outsource reading. Any LLM will happily summarize a paper into five bullet points. But a summary you didn't struggle with is a summary you don't own. If AI does all the digesting, you end up with a well-organized library of things you never actually understood.

So Tide deliberately does not aim to "save you from reading." It aims to:

1. Filter — surface the small number of items genuinely worth attention each day, instead of a 50-item dump.
2. Interpret deeply — for each selected item, publish a real analysis: what problem it attacks, what's actually new, why it matters, and where it might be wrong.
3. Invite you in — every interpretation links back to the original source. Tide's write-up is the on-ramp, not the destination.
4. Leave the understanding to you — Tide prepares the material and asks the questions; the final act of understanding stays human.

Think of it as a research assistant that pre-reads for everyone — not a replacement for your own reading.

What Tide Is

A website, updated daily, where you can:

• 🌊 Browse today's tide — the day's selected papers, tweets, and posts, each with a deep interpretation, not a one-liner.
• 📄 Read paper deep-dives — original abstract → background retold → figure-driven walkthrough of method & experiments, with a link to the original (see below).
• 🐦 Follow the conversation — what the researchers and builders worth listening to are actually saying, with context on why it matters.
• 🗂 Look back — a browsable archive by date and topic, so the tide leaves a record instead of washing away.

Anatomy of a Paper Deep-Dive

Every paper page on Tide follows the same shape. We link to the original, and we add exactly one thing of our own: interpretation.

1. Abstract, preserved verbatim — the paper's own abstract, untouched, plus the link to the original. That's the ground truth; we don't paraphrase it away.
2. Background, told properly — the part most summaries skip. Why does this problem exist? Why does it matter? What did prior work try, and where did it fall short? What is this paper's angle of attack? Essentially the Introduction, retold slower and clearer.
3. Method & experiments, through the paper's own figures — this section is always figure-driven. We walk through the method and the results coarsely, anchored to the figures and tables the authors themselves drew. If a part of the paper has no figure to anchor to, we skip it rather than wall-of-text it.
4. Understanding over formulas — we don't reproduce derivations. If you want the math, that's what the original link is for. Our job is to make you understand what the method does and why it works, not to re-typeset it.

The guiding image: a wave should splash you, not drown you. Tide gets the paper out of the water and onto your skin — enough that you know whether to dive in. The diving is yours to do, at the source.

The Pipeline Behind the Page

• 📡 Subscribe — arXiv topics/keywords, selected Twitter/X accounts, blogs and other RSS-able sources.
• 🧹 Collect & deduplicate — daily ingestion, cross-source dedup, noise removal. Collection is the dirty work; it should be boring and reliable.
• 🔍 Score & select — relevance scoring so the daily selection stays small and worth reading.
• 🧠 Deep interpretation — a multi-step LLM analysis chain producing the page structure above (not one-shot summarization).
• 🚀 Publish — the day's report goes live on the site automatically, every morning.

Status

🚧 Early stage. This README is the founding document — written before the code, on purpose. The "why" should outlive any particular implementation.

Rough roadmap:

• v0: the website — arXiv ingestion → daily selection → deep-interpretation pages, auto-published daily
• v1: Twitter/X sources on the site
• v2: archive & topic navigation, search
• v3: subscriptions on top of the site (RSS / email / Feishu)

Acknowledgements

• Andrej Karpathy, for the "personal LLM wiki" idea and the line that started this project.
• The many open-source daily-digest projects (arXiv digests, KOL aggregators, meridian and friends) that proved the collection layer is a solved problem — Tide focuses its energy on the interpretation layer instead.

License

MIT