When writing a document in VSCode, with Copilot enabled (i.e., help with code completion), I notice that it often gives me latex commands back instead of Typst code. I am guessing this is probably because the data for training Typst specific code is lacking and it looks “similar” enough to latex that copilot suggests latex code.
Has anyone switched to another LLM that may be better suited (I believe the VSCode uses Copilot, which under the hood is OpenAI)
How can we help the next iteration of LLMs train on Typst data?
This is most probably that the trained model doesn’t have enough Typst code in the training data set. Do note that training an LLM does take quite a lot of time and unfortunately a lot of energy (computation and cooling systems), so it might take a while until Typst code gets to have a part in the models.
nope, mostly just raw dogging the Typst syntax and the documentation (and do plan on helping out there at some point)
Create and publish Typst code into the world wide web and await for Baby Jesus in form of millions of AI crawlers