[Smaug123]

And they retrained it within five minutes, or else Claude has access to real-time traffic logs or something while claiming not to? I state again that the text is completely novel; and I've had friends replicate this result using my text, so it's not metadata.

[pixel_popping]

What they claim is a bit irrelevant tbh, it's on their servers, their DB, they do, they don't, we will never know, so we must assume they do. It goes against their business to trash data for no reason, so why would they.

[twaldin]

its realistic to assume anthropic (and other frontier labs) have already gobbled the entire surface of the public internet into training data, plus so much more llm-generated traning data, that if you have public works and a model has enough parameters, i guess you are now 'in the models knowlege' the same way a figure like michael jackson or elon musk would be 'baked into the weights of the model' scary.