cutoff
What if the ChatGPT cutoff date of late 2021 wasn't chosen arbitrary?
What if it was chosen intentionally to avoid training ChatGPT on AI-generated content?
What if the researchers chose this date because they know that in late 2021, the total amount of published content reached a point where AI-generated content exceeded a critical threshold, after which training a large language model produces negative added value?
The thing about these language models is that if you can imagine it happening, it is more than likely that in reality, it is already happening.
In light of this, can it be said with any degree of certainty that we even have a stable, working definition of reality?