Human judgment comes first
AI should help people inspect possibilities and make informed decisions. We want tools whose limits are visible, so people can decide when to use them and when to question them.
Philosophy / What guides the work
Keep people
in the picture.
Our research questions and software builds share a starting point: technology should support human judgment. These principles guide what we investigate, how we explain results, and what we hope to make available.
Let evidence shape the story
Start with a question and be willing to change the answer. Publish methods, explain uncertainty, and give a failed experiment the same careful attention as a promising one.
Build in the open
The aim is to share code, evaluation methods, and the decisions behind the work. An open-source release should help others understand and reproduce a result, with licensing and data constraints made clear.
Efficiency should widen access
Smaller models can make local compute more practical. Measure memory, speed, and quality together so reduced resource use serves an actual human need.
Ambition needs a test.
Questions about cognition and reasoning motivate the work. They do not establish a model's capabilities. A roadmap describes what we intend to investigate; a result needs evidence.