Chat as a process

[enricoros]

Consider chat itself as an AGI operation.
This improvement would abstract the chatting function itself, to enable different interactions (UX), agi products (Product), easier development (DX), and composability.

Abstracted Components: Context, Timeline, Operators (incl. Pre/Post processors).

• the process for an assistant response may include: gathering inputs, injecting context, recursive context pre-processing, generation, post-processing, all while rendering w/optional devtolling
• example new functionality: squash: f(Purpose, Log, Context) -> (Log1, Context1). "Compression of a chat up to this point (discarding potentially irrelevant information) into a single message in the Log, and potentially smaller context"
• other example: Auto Personas - Meta-analysis of chats and baby steps to AGI: Hyperlink in chat to launch new agents for specialized tasks. #78, referral to a new chat (with different purpose)

There would be endless new useful functionality once the technology is refactored from this over-fit chat paradigm (and implementation).

High stress on reusability, zero-lag impact on the base case, auto-progressive rendering, and composability.

Me and gpt4 can take care of the refactoring. @fredliubojin, #78 - heavy inspired.

[jondwillis]

https://github.com/hwchase17/langchainjs -- I just found out that the JS version of LangChain exists. Highly relevant to anything "AGI"

[enricoros]

Interesting. Thanks for the share. I believe we can do something better also :)