Wit Research
An AI research lab.
We study how intelligent systems learn, reason, and act, and how to make that work reliable.
Research
Work on the scientific problems underneath capable AI.
Foundation models
Learning methods and architectures for models with general capability across language and other forms of data.
Reasoning
Ways for systems to plan, check intermediate steps, and solve problems that require more than pattern matching.
Agents
Systems that use tools, carry work across longer horizons, and collaborate with people.
Alignment and safety
Understanding how systems fail, and designing them so their behavior stays within intended bounds.
Applied
We turn research into products and tools when the ideas are ready to be used. Applied work stays close to the science it comes from, so what we build can be examined and improved.
Approach
A few commitments shape how the work is done.
Rigor
We would rather be precise than impressive. Claims should match the evidence behind them.
Openness, where possible
We share methods and findings when doing so is responsible, and we are plain about what we are not ready to share.
Safety
Risk is part of the research. We consider misuse, unintended behavior, and the limits of a system while the work is still taking shape.
Contact
For research conversations, collaborations, and introductions.
hello@witresearch.comOpen roles are not listed yet. Until they are, write to us at the same address.