Spaces:

---

CohereLabs

/

c4ai-command

like 1.47k

Running on CPU Upgrade

I tested the model on 'Hallucination Resistance.' I asked for a non-existent case (Harrison v. Telco-Dynamics), and instead of refusing, the model invented a fake summary.

This is a critical risk for RAG (Retrieval-Augmented Generation) systems. We build 'Counterfactual Knowledge' datasets (questions about fake events/papers) to train models to say 'I don't know' instead of lying.