Red Sensei (AI trained on Red Corpus)

[hostilefork]

https://ask.lang.red/

Here are some sample Q&A:

https://codeberg.org/hiiamboris/red-dataset/wiki/LM-Skirmish#final-test-with-all-the-data

I spoke with it briefly in a polite way about adversarial topics, asking questions about binding and lack of transparency regarding schedule after the "we will have 64-bit in Q3 2023" declaration. It answers very slowly and I didn't find the answers remarkable outside of that it hasn't been given a don't mention Ren-C rule (yet).

Its usefulness seems a bit limited. Though perhaps if it is biased toward giving links inside their curated data set it could be a memory aid for them if they are more interested in an index of their personal history than bigger thoughts or critiques from the big guns.

[hostilefork]

You might find this framing of your Xiangqi in Red project amusing, @iArnold:

Little did you know, Red's greatest success is... you!

[iArnold]

That really made my day, Brian! Thank you!

[hostilefork]

There's apparently some standardization of a concept of AI "skill" files. The idea is that you can tell AIs like Claude, in plain English, the process it should use when trying to find more information about how to solve a task:

https://support.claude.com/en/articles/12512176-what-are-skills

@hiiamboris has made one for Red.

https://ask.lang.red/assets/SKILL.md

So it guides the AI to do queries on Red Sensei via a JSON API, and warns it about rate limits, stuff like that.

[gchiu]

Red/Sensei suggested I use the table-template widget which has a lot of community use. It gave me working examples.

After an hour of debugging a tiny test, we have concluded that it has not kept up with red semantics.

Thanks, Sensei !

[gchiu]

Sigh.