LLMs are largely worthless (in the context of improving human society).

Neural Nets aimed at much more specific domains (recognizing and indicating metastases or other abnormalities in pathology slides for human review, for example) are EXTREMELY useful and worthwhile.