Its relevant because those same issues occur with large models. Also, the "crappy small model", was a model trained for safety tasks, and outperformed the frontier lab safety models.
Small models are notoriously unreliable and prone to hallucination in my experience, so that would not surprise me to be an issue there.