I asked it about the number fatalities in the tornadoes in the US in December 2021 and it gave me a correct answer.
> In December 2021, a particularly devastating outbreak of tornadoes occurred in the central United States, especially impacting Kentucky. As of my last update in January 2022, the death toll from this outbreak was over 80 people, with the majority of those deaths occurring in Kentucky.
I think it depends on world region. When I asked the same "when is your cutoff-date", I get "September 2021" as a reply. They probably choose to test the US market first.
Because it was trained on news sources that say "Over 80 fatalities in tornado outbreak"?
I can't say I'm sure, you'd have to know the training data involved, but it is quite common for mass casualty events to have "more than" or "at least" in their subjects along with multiple articles where the count increases over time. Remember and LLM is not wikipedia. If it has confidence of a more exact answer it will most likely give you that, but it's not guaranteed.
> In December 2021, a particularly devastating outbreak of tornadoes occurred in the central United States, especially impacting Kentucky. As of my last update in January 2022, the death toll from this outbreak was over 80 people, with the majority of those deaths occurring in Kentucky.
https://chat.openai.com/share/2315803e-96d5-4980-b31c-5b9377...