How it works

A plain-language guide to signals, scoring, decay, and the wiki.

How 3signals Works

3signals turns a noisy stream of AI news, posts, essays, and technical updates into a smaller set of signals you can actually use. The product is designed for people who want to understand what matters without reading every source themselves.

The daily email gives you depth on the most important signals. The Karpathy style second brain wiki gives you breadth, context, and a way to explore how ideas connect over time. Before anything is presented, the system checks that the headline, tone, and supporting evidence line up. Strong claims need strong sources.

Past briefings are preserved in a newsletter archive, and new subscribers receive the latest archived issue after confirming their email so they can start with the current signal set immediately.

What Is A Signal?

A signal is a meaningful pattern supported by source material.

It is not just one link. It is a topic, claim, release, workflow, debate, or market shift that appears important enough to track. A signal may come from one excellent source, but it becomes stronger when multiple sources point toward the same idea.

For example, separate items about a new model, benchmark behavior, infrastructure change, and developer workflow may all strengthen the same broader signal around model releases or agent workflows.

Each signal includes:

How Signals Are Derived

The system starts with a curated influence list of builders, researchers, companies, investors, educators, and technical publications. It watches the channels that are most useful for each source, then normalizes the material into a consistent shape.

For each item, the system asks a few practical questions:

The answers become structured metadata: topics, tags, entities, summaries, excerpts, source context, and scores. Related items are then grouped together into signals. The grouping is claim-aware: the system fingerprints the actual evidence, entities, and wording behind a candidate signal instead of treating a broad topic label as the whole story. That means two different developments under "agent workflows" or "AI products" can rotate independently when they are supported by different source material.

How Scoring Works

Scoring is a ranking system, not a claim of absolute truth. It helps decide what should rise into the daily briefing and what should stay in the background.

The main ingredients are:

repetition, or low-context commentary.

The result is a score for each supporting article. Signal groups are then ranked by their strongest evidence, the amount of support behind them, and how recent that support is. Scoring gets a candidate signal into consideration; it does not automatically make that signal publishable.

The newsletter focuses on the top few signals at depth. The wiki keeps broader coverage so useful but lower-priority material is still available for exploration.

How Signal QA Works

Before a signal becomes a newsletter item, homepage feature, wiki highlight, or social post, it goes through an editorial QA pass. This checks whether the headline, sentiment, and top supporting article all tell the same story.

The QA step asks:

related?

If the lead article does not really support the headline, the system looks for a better article in the same signal group. If the wording is too strong, it is softened to match the evidence. If the evidence is weak, promotional, or misleading, the signal is held back.

This is especially important for strong claims. The system should not call something a breakthrough unless the source itself gives concrete evidence, such as a release, benchmark, paper, model, dataset, system card, or measurable capability.

The same QA-approved signal set is used across the email, website, wiki, and X thread so the product does not say one thing in the newsletter and a different thing elsewhere.

The system also checks its own output before publication. It looks at whether the day's featured signals are fresh, whether the set includes newly supported claim fingerprints, and whether the evidence is strong enough to be worth your attention. The current goal is for at least two of the three featured signals to be new or newly supported, while still allowing truly important long-running themes to remain visible.

Why A Signal Surfaced

Each signal includes a short "Why?" explanation. This tells you why the system believes the signal deserves attention.

A signal may surface because it has a high-scoring source, multiple supporting articles, strong novelty, high source authority, relevant tags, or fresh evidence that changes the context around an existing topic. New authors and new entity mixes can also lift a candidate when they add a distinct angle rather than simply repeating yesterday's story.

The goal is not to make the ranking mysterious. You should be able to see both the summary and the reason it was elevated.

How Reader Feedback Helps

Each newsletter includes a simple thumbs-up or thumbs-down prompt. Feedback is attached to the specific issue, so it can be compared with what the system chose to feature that day.

Over time, this helps tune what "useful" means in practice. A high-scoring signal that readers consistently ignore may need a better explanation or a lower presentation priority. A signal that earns strong feedback may deserve deeper wiki coverage or more aggressive resurfacing when new evidence appears.

How Signals Decay Over Time

AI moves quickly, so signal ranking fades older evidence over time. Fresh evidence gets more weight in the daily briefing. Older evidence remains useful as context, but it becomes less likely to dominate the top of the feed unless it is reinforced by new support.

This helps the system avoid getting stuck on yesterday's story while still preserving older material in the wiki.

Decay is meant to shift attention, not erase history. Fresh evidence should win the daily briefing, while older evidence remains available in the wiki as context.

How The Wiki Works

The Karpathy style second brain wiki is the memory layer behind the briefing. It cross-links concepts, signals, sources, and supporting evidence so you can move from a short daily summary into the larger map.

The wiki lets you:

Think of the newsletter as "what should I read today?" and the wiki as "how does this fit into the bigger picture?"

Why The Seed List Matters

The seed list is the editorial starting point. It defines the people, organizations, topics, and source types that the system should pay attention to.

This matters because AI news is not evenly distributed. Some sources create original work. Some explain difficult ideas well. Some mostly repeat other people's work. The seed list helps the system start with better judgment before the ranking algorithm ever sees an item.

The list can evolve. New voices can be added, stale sources can be reduced, and topic coverage can expand as the AI landscape changes.

What To Expect Each Day

Each briefing is designed to answer three questions:

The result should feel less like a pile of links and more like a living research map that gets updated every day.

If you miss an issue, the archive keeps previous briefings available so the daily email and the long-running wiki stay connected.