>but usually based on some in-house LLM, custom training or tuning etc, nearly always at vastly smaller scale Many of them are much simpler than that: merely m

>but usually based on some in-house LLM, custom training or tuning etc, nearly always at vastly smaller scale Many of them are much simpler than that: merely modifying/enhancing the prompt and managing the back-and-forth with an LLM and then structuring the output. There's lots of use-cases that you can achieve with just that, but easily copyable.