OpenAI or ChatOpenAI when using RetrievalQA ?

[SleepingSkipper]

I see the following two types of sample code for instantiating RetrievalQA, what is the difference? Which one is recommended?

from langchain.llms import OpenAI
qa = RetrievalQA.from_chain_type(llm=OpenAI("gpt-3.5-turbo"), chain_type="stuff")

vs

from langchain.chat_models import ChatOpenAI
qa = RetrievalQA.from_chain_type(llm=ChatOpenAI("gpt-3.5-turbo"), chain_type="stuff")