I have been trialling the use of ChatGPT to provide help with TW programming. It works a treat and I seem to get better results than using the TW documentation directly.
Give it a try it works a treat!
Bobj
I mostly use Claude3.7 thinking on Github Copilot. It performs better on programming task than GPT4o.
But agent mode is worse than simply using edit mode. Agent mode usually end up with tons of garbage files, still need me to review and refactor one by one.
1 Like
It would be ideal to train a simple language model specifically for TiddlyWiki and host it online. I am uncertain whether LAMA could be applicable in this case: How to Fine-Tune Llama-3.2 on your own data: A detailed guide | by Alexandros Chariton | Medium
2 Likes
Thanks for the information @linonetwo I already started to feel stupid because I am too lazy to start using agent-mode.
I agree with @linonetwo, When it comes to TiddlyWiki, so far Claude seems to be the only AI who seems to be “knowing” what it is doing. Other AIs pretend to but it ends up in a mess fast. It is a problem to get the right mix between js and wikitext.
Finding bugs in existing code is also viable with other programs.