About What the AI

Curated AI intelligence for every level.

The Idea

AI is moving fast. Too fast for any one person to keep up with everything — and most news sources either drown you in hype or assume you already have a PhD in machine learning.

What the AI is built on a simple idea: everyone deserves AI news at their level.Whether you're just curious about what ChatGPT actually does, or you're deep in the weeds of transformer architectures, you should be able to find signal without the noise.

We pull from dozens of sources — Hacker News, arXiv, RSS feeds, X/Twitter, Reddit, GitHub Trending — and use AI itself to classify, summarize, and organize everything into five levels of depth. The result is a feed you can actually use, no matter where you are on your AI journey.

The Levels

Curious — Non-technical, just AI-curious. Major launches and plain-English explainers.

Practitioner — Using AI tools daily. Prompt tips, workflow automation, feature updates.

Builder — Power users and low-code builders. Custom GPTs, MCP servers, API integrations.

Developer — Software engineers building with AI. SDKs, agent frameworks, architecture patterns.

Researcher — ML engineers and AI researchers. Papers, training techniques, model architectures.

How Ranking Works

Every article in the feed gets a "Most Relevant" scorethat determines its position. This isn't a simple recency sort — it's a composite signal designed to surface what actually matters.

What goes into the score

AI quality classification (0-100 pts) — Every article is classified by an AI model that evaluates depth, originality, and significance. Scores are calibrated across a wide range: groundbreaking launches score 90+, solid coverage lands 55-74, and thin content falls below 35.
Engagement signals (up to 20 pts) — Hacker News score, Reddit upvotes, GitHub stars, and Bluesky engagement are combined. An article trending across multiple platforms gets a bigger boost.
Cross-source virality (up to 10 pts) — Articles discovered by 2+ independent sources get a bonus, since widespread coverage is a strong signal of importance.
Reader votes (up to ±40 pts) — Your upvotes and downvotes are the heaviest signal. Community feedback matters more than any algorithm. Each net vote is worth 8 points.

What pushes articles down

Time decay — Articles lose relevance over time on a logarithmic curve: steep in the first day (~14 pts), then gradually flattening (~37 pts after two weeks). This keeps the feed fresh without instantly burying good content.
Model release supersession — When a newer model in the same family is announced (e.g., GPT-5 after GPT-4), the older announcement is automatically demoted so you see the latest.

The result: a feed that balances recency, quality, community signal, and real-world impact — not just what was published most recently.

We Want Your Feedback

This is an early-stage project and we're actively building based on what people actually want. If something feels off — the classification is wrong, a source is missing, a feature would make your life easier — we want to hear it.

The best way to reach us is through GitHub Issues. File a bug, suggest a feature, or just tell us what you think.

You can also contribute directly — suggest glossary terms, vote on articles to help improve quality rankings, and bookmark what matters to you. Every interaction makes the feed better for everyone.

Part of Some Shovels

What the AI is part of Some Shovels — digital artifacts for builders. We make tools that help people do real work with technology, without the fluff.

someshovels.com