← All work
Project / AI / Interpretability

Jewish Qwen

Experiments in steering Qwen3.5-4B with neurons, activation directions, and optimized vectors.

Open the original ↗Original cover artwork for Evil Qwen’s Jewish Cousins

Steering Qwen3.5-4B

The project experiments with individual neurons, activation directions, and optimized vectors to change the model’s behavior. Kosher-Qwen and Rabbi-Qwen are two of the experimental variants described in the repository.

The source includes the steering methods and example outputs, and links to the accompanying “Evil Qwen’s Jewish Cousins” article. These are experiments in model behavior, not a general capability benchmark.

Related work.