What training data? Many of these languages have very little digitized literature. Even if we assume they have sizeable extant corpuses (e.g. Tibetic/Bhoti), that's not enough. LLMs are still pretty garbage at English prose, for example.