The problem you are running into is that you are overindexing on the fact that LLMs will sometimes be wrong and you are used to using technology that is basically always right. But we are in the early stages of LLM adoption - correctness will improve (see for example citation driven LLM-search) but more importantly, the set of LLM-driven applications that can be probabilistically correct and still wildly useful will grow.

LLMs like ChatGPT are just so damn cheap for the power they provide, it's inevitable