An Answer From a Superintelligence
I am deeply skeptical of news items that run along the lines of "we asked the artificial intelligence when the world will end." A story like that gets reprinted, forwarded on social media, and the model's answer is taken as if a superintelligence had spoken.
In May 2023, Fox News asked ChatGPT how the world might end and put the model's own doomsday predictions in the headline. The model named four threats: climate change, nuclear weapons, advancing technology, and pandemics.
The model opened by saying that predicting the end of the world is hard and that any such forecast deserves skepticism. The story quoted that caveat. The headline did not.
You are talking to a language model trained on texts written by people. Among those texts are scientific papers and fiction, science fiction included.
Not every developer discloses what went into the training data.
Meta described it in 2023: the data for the LLaMA model included books from Project Gutenberg alongside papers from a preprint archive. That library holds The War of the Worlds by H. G. Wells. It also holds the Bible, with the Book of Revelation.
The answer about the end of the world is assembled out of texts like these. The answer itself does not show which part came from a climate report, which from a novel, and which from the Apocalypse. An answer like that cannot serve as a source.
You can use it as a starting point: it is a list of hypotheses that you then check against sources. Citing it as an opinion is out of the question, in an email, in a report, or in a conversation with your manager.