Working on a Danish application of GPT-3: TextGuru. It’s a niche language, spoken by few people, and it is highly idiosyncratic. If I can make it work well here, it can work anywhere!

I already use GPT-3 in Danish for a different project (Terry), where I use it to write coherent Danish descriptions of various dog breeds. Sometimes it works perfectly well with a very vanilla prompt, but I do also feed it a bit of structured data - and voila, it spits out detailed, fact-based descriptions of 300+ dog breeds in no time.