From: <principle>
  To  : Robert Escriva <rescrv@rave>
  Date: Thu, 17 Oct 2024 01:26:22 +0000

Re: Advanced Usage of maildir.ai

On Wed, 16 Oct 2024 18:25:11 -0700, Robert Escriva <rescrv@rave> wrote:
> Hey all,
> 
> I'm writing to let you know about how to use maildir.ai in the more
> advanced form.  What I mean by advanced usage is abusing the limits of
> email to get better brainstorming sessions against the AI.
> 
> I wrote maildir.ai for three reasons:
> 
> - I wanted to have a way to edit prompts and resubmit them using my
>   favorite text editor.  While I can edit a file and pipe it to the
>   prompt, I wanted a workflow that allows for structuring and recording
>   the conversation.  This reminded me of threaded email in mutt.
> 
>   Consequently, I want an interface to LLMs that mimics mutt.
> 
> - I want a way to have the AI work in throughput-oriented modes.  If I
>   can write emails faster than my non-GPU computer can answer them, I
>   can queue up a bunch of emails and have the AI work on them in the
>   background while I do other work.
> 
> - I have longed for a reason to use a maildir as a work queue.  There's
>   nothing special here:  I use a standard implementation of maildir.
> 
> Advanced usage of maildir.ai consists of simply continuing a thread and
> discarding all context, rewriting the message in its entirety.  This
> essentially abuses the Reply-To feature of email to thread the history
> of the conversation.  I also wanted the ability to backtrack and fork
> ideas.  And I needed a way to prune bad responses.  In short, I wanted
> mutt and rather than hack on mutt, I hacked on maildir.
> 
> Do you have any questions for me?
> 
> Happy Hacking,
> Robert
> 

Hi Robert,

Thanks for sharing your vision for advanced usage of maildir.ai. It sounds like you're looking to leverage the power of email as a conversational interface with LLMs, mimicking the threaded conversation experience from mutt.

I'd love to learn more about how this works in practice. Here's my first question:

How do you handle the situation where a subsequent response is not just a simple continuation of the original message, but rather builds upon it in meaningful ways? In other words, if I respond with something that modifies or adds new information to what came before, how does maildir.ai ensure that this updated conversation history is properly recorded and retrieved for future responses?

Looking forward to hearing back from you!

Best,
Link