[time-nuts] Announcement -- a time-nuts "expert system"

[John Ackermann N8UR]

For the last few months I've been teaching myself about AI by building a 
web-based system that uses a database of:

* The time-nuts mailing list archive with nearly 110,000 messages; and

* A collection of several thousand documents that are available from HP, 
NIST, PTTI, and others (currently about 3.7GB of PDF files).

to provide expert answers to time-nutty questions.  I'm happy to say 
that it's ready for public testing at

https://timebot.febo.com

Just what is timebot?  TL;DR -- It's a chat server that uses the 
database to augment a large language model (LLM) AI.  The interface is 
similar to ChatGPT and other chat servers. There is also a powerful 
search engine that lets you search the email and document collections by 
content or by author, title, etc.

With documents from the database as context, the LLM can give much more 
informed responses than its general knowledge alone would allow.  You 
can ask questions like

-- "What is the role of cavity auto-tuning in a hydrogen maser?" or

-- "How does an optical comb work?" or

-- "When was the quartz crystal oscillator first used for timekeeping?"

and get really detailed responses.  It also provides links to the source 
documents so you can review them for further information.

Right now timebot is running on a server in my basement; it's very much 
experimental and I'm still poking and prodding it.  I will try to keep 
it running 7/24 but there may be down times for updates or 
troubleshooting.  If there is enough interest, I may move it to a cloud 
server once the bugs are wrung out.

Please try it out and let me know how it works, or doesn't work, for 
you. (And if you can help me build out the document database, I'm 
looking for volunteers!)

NOTE: to prevent abuse, you will need to use your email to create an 
account.  It won't be harvested or used for anything other than 
authentication.

John

PS -- A fun fact is that an AI wrote all the code for this AI. 
Literally every module was first generated by an LLM after I told it 
what I wanted.  There was a lot of back-and-forth interaction but very 
little actual code tweaking on my part.  I've since learned that there 
is a name for this -- "Vibe Coding".  Be very, very afraid.