Add to it that not all LLMs are equal: at work I have access to the latest GPT models and can customize them myself. At home anything I tried in LM studio feels very limited/inflexible. All models i tried running were useless except one codestruct, which was half decent at answering clarification questions about c# code from tutorial, but failed the task i initially tried it for: generating bespoke tutorials that fit my knowledge gaps, which would have been my ideal scenario because most tutorials aren't written for someone like me.

It's another reason LLMs are a bit disappointing to me, they don't seem to be good at handling queries I couldn't solve with a web search, and have similar issues as search results when they do have answers - this makes sense since that's how they are trained, but often defeats the purpose for me.