I am arguing LLMs should not be used for anything, since in my opinion their downsides outweigh their upsides.

Also, I don't consider LLMs a tool, because I can trust my tools, and I cannot trust anything an LLM outputs.