That mimics precisely my experience.
When i ask simple and common things, the LLM answers with an adequate response. When I ask more complex and niche things, the response is just random garble.
Also, for code / programming, it is not only quite useless, but also dangerous. The samples that...