Really depends on the context.

Think of these LLMs as black boxes that predict the most likely next word based on the thousands of previous words. So, without this history, it's impossible to answer your question.

Now, modern versions can search the internet and thus can get accurate information, which goes beyond the "most likely next word" paradigm. But these web searches only get triggered with the right keywords. I just tried, and it gave me what seems to be an accurate exchange rate.

For some reason, your model did not get triggered to search for up-to-date information.

Understanding how LLMs work, even at a high level, also gives you a decent intuition as to why they likely won't achieve AGI within this current paradigm. It does not know if what it tells you is correct or not.