Even LLMs have use cases were they are a good tool - fixing grammar, low risk translations, etc. Unconstrained chatbots with models that are way past the point of diminishing returns is just not one of them.