Interesting. Are there any other worthwhile recent papers on using LLMs to improve computer vision?