About AI Potluck

Potluck: A model for collective abundance

The dominant metaphor for AI is the race. Races require hoarding resources, moving in secret, treating collaboration as risk. The AI race metaphor ends up being prescriptive: it shapes strategy, funding logic, and the behaviour of institutions. For the public interest movement, this is a trap. By race logic, we are way behind. We should not bother trying.

We propose a different metaphor: Every culture in the world has a version of the shared meal. The Lebanese mezza, the Indian thali, the Spanish tapas. When people bring their best and break bread together, something emerges that no single person could have done alone.

We're making a bet that this oldest model of collective abundance is also the right one for the moment we are in. A host issues an invitation to like-minded entities who have something to contribute: a foundation model, fine-tuning capabilities, a benchmarking framework, safety tools, some compute. Our first versions do not need to be perfect. They just need to get us up and running.

Photo of a shared table with many dishes

From a fragile stack to a resilient one

AI resilience matters enormously in AI infrastructure. The current proprietary stack is fragile: one company's policy change, one geopolitical shift propagates instantly to millions of dependent users. Meanwhile the potluck model is distributed, open, and has no single point of failure.

A government makes a focused, high-leverage contribution (compute, data, evaluation, a domain-specific application layer) and that contribution connects to a coherent global technical effort.
A government can own the parts that matter most to them without building everything else from scratch. When Finland improves the evaluation framework, India's deployment benefits immediately. When Kenya adds Swahili training data, Japan's multilingual fine-tuning improves. The contribution culture of open-source software—fork, improve, merge—is the agile international cooperation model that governments have been looking for.

Our vision: AI that puts people before profit

We want to build better AI. AI that doesn’t vacuum your data. That doesn’t want you to lose your job, AI that wants you to thrive, doesn’t want you to be a zombie. AI that doesn’t thirst after your water, but runs efficiently. AI that respects your time off. AI that’s beside you until you don’t want it to be. AI that measures itself, keeps itself accountable. AI that gives you controls. AI that thrives off our differences and doesn’t flatten the richness of our languages, our cultures, our writing. AI that respects and remembers our ancestors and our pre-industrial ways of life. AI that doesn’t make you shrink yourself. AI that doesn’t perpetuate bias, or skew towards the loudest or the most digitized. That doesn’t sell you sycophancy. That models good behavior. That tells you it’s enough when it’s been too long. For real. Not pretend to get more engagement. AI that wants you to turn it off. AI that doesn’t try to sound human. That doesn’t need to have a pronoun and a feeling and a consciousness. AI that just acknowledges that it’s a robot, and that the most beautiful thing about life is being a real human.

A. Bdeir
Ayah Bdeir, CEO of Current AI

Beyond words: this is how we do it

The manifesto is our why. Our roadmap is how we do it. It is where our vision and values turn into new partners, new product releases, new deployments, a new model for what open source AI that puts people before profit can look like.

  1. Shipped Jul 2026

    Gap Map

    The Gap Map tracks more than 58,319 open source AI products across the stack and assesses them for openness, capability, and adoption.

    Read release
  2. Shipped AI For Good – Jul 2026

    Alpha v0.1

    An ethical chatbot built with 10 partners, out of available open source parts in just 7 weeks. Alpha is the proof that an alternative is possible.

    Read release
  3. Next up MozFest – Oct 2026

    Alpha v0.2

    Make AI Yours: your language, your data, your rules. New partners join the potluck as we build an AI that anyone can speak with, and that is in your control.

  4. FOSDEM – Feb 2027

    Alpha v0.3

    With a focus on safety, security and community use cases.

  5. World Summit on AI – Jun 2027

    Release Candidate: Alpha v1.0

    Alpha will be live for adoption across the public interest ecosystem.