How we work

AI finds the research. A person decides what you read.

Every week thousands of papers are published and most never leave the lab. We use AI to read widely and write first drafts, and a person decides which ones are published.

  1. 01 · Find

    Every night, we read the new papers

    New studies come in from open sources such as arXiv and OpenAlex. AI scores each one for how new, how solid and how useful it is.

  2. 02 · Draft

    AI writes a first draft

    For the most promising studies, Claude drafts an article in plain language and proposes a readiness level with its evidence. Drafts never go live on their own.

  3. 03 · Decide

    An editor decides

    A person reviews every draft and decides whether it is published. Nothing goes live without that approval. The original study is always linked, so you can check the details yourself.

Our signature

The Readiness Scale

Every story gets a level for how close the research is to real use. A level always shows its evidence, and what would move it up. If research moves, we update the level and show the move on the front page. An automatic check looks for newly published code every week and can raise a study from II to III. Every other level is set by the editor.

I
Lab only

Results from controlled experiments run by the authors.

We look for A clear method and results we can follow in the paper.

II
Benchmarked

Measured on public benchmarks that others can verify.

We look for Named public benchmarks and comparable baselines.

III
Open source

Code or weights are published, so the work can be reproduced.

We look for A working public repository or model weights with a usable licence.

IV
In production

Documented use outside the lab where it was made.

We look for A named user or operator, and evidence of use beyond a pilot.

Our principles

  • Calm and factual. No hype, no doom.
  • We separate what a study shows from our reading of it.
  • We always link the original study and mark preprints as not peer reviewed.
  • We write in our own words and only reuse figures with an open licence.
  • Every article says how it was made, including the role of AI.
  • Advertisers and sponsors never influence what we cover or how we rate it. Ads are always marked.
  • Science Underway is built by Nexxion. If we ever cover Nexxion or its partners, we say so in the article.

Corrections

When we get something wrong, we fix it in the article, say what changed and when, and list it here. If a level was wrong, we say so.

No corrections yet.
Report an error
Editor

Stefan Perdal has spent 25 years building digital products and now edits Science Underway. New sections will each have a qualified subject reviewer.

Contact the editor