AI for foreign language conversation practice?

A user on reddit.com is asking for opinions from those knowledgeable in computer science or linguistics regarding the use of a sustainable, net environmental positive LLM for foreign language conversation practice. They are curious if such a system would be reliable for this use case, where they could speak to it, receive responses, and get feedback on their mistakes, contrasting it with research applications where they believe LLMs are fundamentally unreliable.