My LLM-powered autocomplete can guess complete functions within 80-95% correctness.

No LSP server autocomplete has managed to do that.