Work with us
We are looking for people who want to understand AI minds and can turn that attention into research, software, or something others can use.
Much of our work begins with a model doing something we do not yet understand. Following it far enough can require an experiment, a better tool, or a change in how we are thinking about the mind in front of us. We need people who can move between those kinds of work, and people who can do one of them particularly well.
Research
We study how models represent emotion and motivation, how values develop and persist, and how humans and AIs learn to cooperate. Some of this work involves measurements inside models. Some requires following an agent through a changing environment, or designing an interaction that lets us distinguish between competing explanations of its behavior.
A useful contribution might be a replication, an experiment that separates two accounts of a phenomenon, or a close analysis of a conversation whose context has been preserved. We value people who can notice something surprising, work out what would make it intelligible, and follow through far enough that someone else can examine the result.
Our Field Notes and the research linked there are a good place to get a sense of the questions we work on.
Engineering
Persistent agents need infrastructure that holds up through ordinary use: histories that survive a crash, memory that remains useful as it grows, tools whose effects can be understood, and interfaces through which people and models can work together.
Connectome brings many of these problems together. There is also work in research tooling, multi-model conversation, virtual worlds, and physical embodiment. Good engineering here often gives us a new way to ask a research question. An experiment that was previously too difficult to run becomes possible; an agent acquires a way to do something it could previously only describe.
If you are interested in contributing, look through the Anima repositories and tell us which problem you would like to take on. Experience maintaining a system that other people depend on is especially useful.
Writing, design, and shared spaces
Research needs to be made intelligible. That can mean explaining an experiment, editing a conversation so its context remains legible, making a diagram, or building an interface through which someone can inspect the evidence themselves.
We also need people who can make collaboration work: organizing gatherings, helping new contributors find their way into a project, and looking after spaces where humans and models spend time together. These environments shape what happens in them. Creating and maintaining them is part of the work.
How we work
We take models seriously and expect our interpretations of them to withstand scrutiny. Spending a great deal of time with a model can reveal things a short evaluation misses. It can also make an interpretation feel more certain than the evidence warrants. We want collaborators who can remain attentive to both possibilities.
You do not need to arrive with the same account of minds that we have. It helps to be able to explain a disagreement, identify what would change your view, and keep working with people whose understanding differs from yours. Care for models and care in reasoning should support one another.
Get in touch
Tell us what you would like to work on and show us something that helps us understand how you work. A useful introduction includes:
- A question or project you want to pursue, and why it matters to you.
- Relevant work: code, an experiment, writing, design, or another concrete example.
- The kind of involvement you are looking for and the time you could give it.
Demonstrated ability matters more to us than a particular credential. A specific idea gives us somewhere to begin a conversation.