Indeed. llama2.ai seems worse than https://chat.petals.dev/

With some fine tuning can get really good results from the 13B running locally. Using the H2O LLMStudio.