The author is grasping at straws. I don't think it'll be a particular problem. The original dataset, yes, would be subject to GDPR and the onus is on OpenAI to

The author is grasping at straws. I don't think it'll be a particular problem. The original dataset, yes, would be subject to GDPR and the onus is on OpenAI to follow GDPR and delete any confidential information for which they do not use. The treatment is no different to that of web crawlers, is it implied that search engines are non -compliant? The model itself can be argued to be outside the scope of GDPR as it contains anonymized data, the words merely have a higher probability that they will go together based on its past knowledge but it is not actually presenting the dataset information. It's no different to me inventing a person's name and birthday (John Smith, 1st April), there's a high probability that there's a John with a surname Smith and given the limited numbers of days in the year, there's likely a John Smith born on that day. It does not mean that a real life John Smith born on that day can claim my fictional person to be himself and claim GDPR noncompliance.