alt Hacker NewsHa, explains why AI is at a loss trying to figure out the meaning behind that sentence - to recent.
Perhaps that's the ultimate AI detector? Information too recent, too obscure or too useless to have been used to train language models?
Ha, explains why AI is at a loss trying to figure out the meaning behind that sentence - to recent.
Perhaps that's the ultimate AI detector? Information too recent, too obscure or too useless to have been used to train language models?
Not a bad idea, though 'too recent' I think would be the only benchmark - for now, until they get faster at training. OpenAI has thrown away most of their safety team so that shouldn't be a reason anymore to delay model releases "for safety checks".
What would be too obscure or useless, when every new model boasts increasing parameter count (as if having more parameters would make the models better after a certain threshold)?