Read-Decompose-Ask approch dropped?

[cforce]

Once there was Read-Decompose-Ask approach which seems have been removed on the way. I am curious why they have been dropped and what was chosen finally as better option ?

In the Read-Decompose-Ask Approach, the question is broken down into individual steps or sub-questions, and each step is answered in turn to arrive at a complete answer. The process involves three steps:

1. "Read": The system reads and understands the question.
2. "Decompose": The system decomposes the question into sub-questions or individual steps.
3. "Ask": The system answers each sub-question or step to arrive at a complete answer.

This was useful for complex questions that require a multi-step thought process to arrive at a complete answer. By breaking down the question into smaller steps, the system can answer each step more easily and then combine the answers to arrive at a complete answer.

[pamelafox]

That approach used the langchain SDK, which was unstable at the time, so the code was in a broken state. We made a few attempts to update it, but it didn't seem worth it based on the popularity of the other approaches. The langchain SDK is now a bit more stable and lighter weight, so perhaps it could be brought back in if someone wanted to make a PR. Or it could be implemented without langchain. It seems also like what the o1 model does internally, so that might be another approach.