Research finds fine-tuning the MLP of some AI models lessens catastrophic forgetting during the fine-tuning process.