Independent AI research community — Paris, France

Small models, big questions.

Open models and tools have made it possible for individuals and small teams to do serious AI research independently. nimblelab.ai provides the rest: community, collaboration, review, and a shared affiliation to publish under.

01

Community and collaboration

Meet other people doing AI research on their own or in small teams. Test ideas on each other, find collaborators and run experiments together.

02

Serious review

Get critical feedback before you publish, from at least two reviewers who weren't involved in the work. If it doesn't meet the bar, it doesn't carry the name.

03

A shared affiliation

Publish under nimblelab.ai. Every paper must be written by its human authors. The name tells readers the work has been checked, and it becomes part of a shared, public research record.

How it works

Serious research, without doing it alone.

AI research is no longer confined to universities and large labs. What independent researchers still lack is what institutions have always provided: people to discuss the work with, people to criticise it, and a name that tells readers the work has been checked.

Researchers join with their own questions and projects. Some work together on experiments or papers. Others contribute by discussing ideas, reviewing results or sharing expertise.

There's no requirement to have a particular academic background or publication record. What matters is the work, and your willingness to engage with criticism.

One rule we don't bend: papers are written by their human authors. Writing is where you find out whether you really understand your results, and our reviewers' time is best spent on work its authors can stand behind line by line. Generated text makes papers easier to produce and harder to trust, and keeping it out is how we keep the quality of submissions high.

Getting started

Joining takes one email. Here's what you get from there:

  1. 01
    A warm start. Tell us what you're working on or curious about, and we'll introduce you to the people and projects closest to it.
  2. 02
    Early feedback. Share ideas and experiments while they're still taking shape, and get input from people who know the field, long before you submit anything.
  3. 03
    Collaborators. Team up on experiments, check each other's results, or co-author a paper.
  4. 04
    A name to publish under. When the work is ready, it goes through review. Whatever the outcome, you come away with detailed feedback that makes it stronger.

Not running a project yet? You can contribute by reviewing, or by lending your expertise to someone else's.

Get started →
Our community

People doing research, reviewing it, and supporting it.

Researchers

You might be an engineer exploring a research question in your spare time, a researcher between positions, someone working outside the usual hubs, or someone who has chosen to work independently.

Bring your own projects, find collaborators, and get serious feedback before you submit. We ask members to share what they learn and to contribute to others' work when they can.

Members also get the Idea Engine, the lab's in-house tooling for staying on top of the literature.

Join as a researcher →
Reviewers

The standard behind the name is only as good as the people holding it.

We're looking for people with experience in AI research and scientific publication. Reviewers assess methods, challenge conclusions and decide whether work is ready to carry the nimblelab.ai name. Sometimes that means saying no. Reviewers also help shape the standard itself.

We ask for two reviews a quarter. Contributions are publicly acknowledged unless reviewers prefer otherwise. Members are encouraged to review too.

Become a reviewer →
Supporters

Independent researchers often have good ideas but limited access to compute, funding or other support.

We welcome organisations and individuals who want to help: by providing compute credits, funding specific projects, supporting shared infrastructure or connecting researchers with other institutions.

Research decisions and scientific review remain independent of financial support.

Support us →
The Idea Engine

A small lab can keep up, with the right tools.

Independent researchers don't have a reading group down the hall or an engineering team behind them. We think that's largely a tooling problem, so we treat it as one.

The Idea Engine is the lab's own research software, built and run in-house. It keeps us on top of a field that moves faster than anyone can read, so the hours we have go into thinking and experiments rather than sifting.

It reflects how we work more broadly: we take AI tooling seriously, build what we're missing, and share what works. Every member gets it, and it gets better the more the lab uses it.

Member login →

What it gives members

  1. 01
    Signal, not noise. The work that matters to the lab, every morning, with a reason to read each one.
  2. 02
    A shared memory. Everything the lab has come across, in one place, alongside what other members found worth reading.
  3. 03
    Room to go deep. When your own question needs a proper look at the literature, the Engine gets you there in minutes rather than weeks.
Our research

Learning and self-improvement in small language models

Alongside our members' work, nimblelab.ai runs its own research programme. We study how language models can improve their own learning, particularly when compute is limited.

Can a model recognise which new training data would help it improve? Can it learn from its own feedback? Can it adapt its training process without relying on ever-larger datasets or models?

Working with small language models lets us repeat experiments, isolate variables and examine failures in detail. We want to understand when self-improvement works, when it fails, and whether the mechanisms we observe generalise beyond the original experiments.

Curious about these questions? Members can join the programme: propose ideas, run experiments, or challenge our results. You'll work alongside us from the first experiment, and share in the results when we publish.

Follow our research →
Who's here

A lab made by the people in it.

nimblelab.ai is small on purpose. Everyone here does research, reviews it, or helps make it possible, and the lab is shaped by what its members bring: their questions, their expertise and their standards.

Niko Groeneweg

Niko Groeneweg

Founder — Paris

15+ years experience building AI products. MSc in artificial intelligence, MA in neuropsychology.

Vitaly Charushin

Vitaly Charushin

Researcher & contributor — Paris

AI researcher and strategist.

You?

Researcher, reviewer or supporter

Doing AI research on your own or in a small team, or want to help others do it? There's room for you here.

Get started →

Join the conversation

Working on something interesting, have experience to share, or simply want to contribute? Tell us a little about yourself and what you're interested in. We read everything.