Comment by greyjoyduck

They had to scale back all the instances when they used openai as the llm (GitHub copilot) and now they are using their own proprietary model which is just ok at best.