Prototypes
AI lets us build functional prototypes of new tools and test them in the wild much faster and more easily than before. Explore our index of research prototypes below.
Information
Trading cards for 28 frontier models, built from what their labs actually disclose.
Compare models side by side on capabilities, safety and oversight, with disclosure notes flagging where evaluations aren't comparable across labs.
Open the prototype →Governance
The first systematic test of whether frontier models resist authoritarian requests.
We found that resistance to these requests fell significantly when we hid them in code.
Open the prototype →Information
Explore roughly 280,000 candidate fundraising emails and the populism they reveal.
An interactive companion to our research on the Democratic party's turn towards anti-billionaire populism, and how that appeal is beginning to merge with a new kind of anti-AI populism.
Open the prototype →Bellwether Metrics
Structured data on real-world events, with verified outcomes and manipulation-resistant probabilities.
We curated structured data on real-world events — elections, conflicts, policy shifts — with verified outcomes and manipulation-resistant probabilities, built for researchers, newsrooms, and ML pipelines.
Open the prototype →Information
See how the same political question gets answered through different ideological lenses.
Submit any political or policy question and get back a neutral overview alongside how Libertarian, MAGA, Progressive, Center-Left, and Marxist framings would each analyze it. A transparency tool that makes implicit ideological framing in AI responses visible.
Open the prototype →Information
Standardized ideological assessment of every major AI model.
An independent dashboard tracking how the frontier models lean.
Open the prototype →
