Do we need large language models to translate a language spoken by not more than 500 individuals worldwide?

There seems to be an obsession going on in the field of computational linguistics: How many neglected and rare languages an LLM can translate!

It maybe exciting, but this pursuit seems to be kind of silly! Do we not have more important tasks to solve?

Here is just one recent example mentioned by Andrew Ng:

"[Google Gemini Pro 1.5] ... Given 500 pages of documentation that describes Kalamang, a language spoken by fewer than 200 people in West Papua, it translated English text into Kalamang as well as a human who had learned from the same materials. ..."

Google's Troubled Gemini Launch, OpenAI's Next Act, Groq's Blazing Inference Speed, Faster Network Pruning