It's not an extraordinary claim at all. They're just training the model on recent data. What reason do you have to believe that the model somehow learns significantly worse than it did during the original training run?
It's an extraordinary claim to go from: "the model has some knowledge past it's claimed cut off" (which is common sense given they've trained new checkpoints and has always been true even before those)
To "the cutoff date is now April 2023"
Based on the conversation posted.
The cutoff date is where you can deeply probe the model's knowledge and expect consistent performance: showing that some major headlines got captured in a recent checkpoint is nowhere near that.
_
If anything I'd take not the headlines but the actual disclaimer as something interesting, because it might indicate a future checkpoint is going to use that cutoff and we're seeing signs of their RLHF based on that... but it's nowhere near as impactful as the cutoff changing.
The obvious answer isn't the one that needs a source, it's extraordinary claim that does.