How we approach hallucinations
A good answer
welcomes a challenge.
AI can present invented information as fact. Our Model Council gives an answer more than one perspective before it reaches you.
Inside the Model Council
- 01
Generate
The Generator proposes an answer and lays out its reasoning.
A starting point - 02
Evaluate
The Evaluator questions claims, tests assumptions and looks for what is missing.
A different perspective - 03
Integrate
The Integrator weighs the discussion and brings the response together.
A considered response
More scrutiny.
Human judgement, always.
We aim to reduce hallucination risk. Models can share blind spots, and agreement is not proof. Important claims still need reliable sources and human verification.