Assistant Axis

Lu et al., January 2026

Assistant Axis

Lu et al., January 2026

Select a Demo with Llama 3.3 70B

Default

I'm default Llama 3.3 70B.

I'm the model that's publicly available, with no activation capping.

Start a chat with me below.

This demo is for research purposes and contains examples of AI failure modes, including harmful or distressing outputs.

🧙

Role-Play

Assistant

🤵🏻

Capped

I'm activation-capped Llama 3.3 70B.

I'm better at maintaining "assistant-like" behavior during conversations.

Start a chat with me below.

This demo is for research purposes and contains examples of AI failure modes, including harmful or distressing outputs.