Inducing anxiety in large language models increases exploration and bias

Open in new window