Recently, gpt-4 has responded that it cannot provide detailed summaries of more than 90 words. It says this because of its guidelines and copyright concerns. But the documents I ask it to summarize are publicly available government materials. They are not copyrighted. 90-word summaries are worthless because they are too general to say anything meaningful about the information in a document. My recommendation: allow gpt-4 to make exceptions for publicly available government documents. Nuance the guidelines so clearly legitimate uses of gpt-4 are not curtailed.
There is no reliable way for the model to know what is and what is not copyrighted. It’s not possible for them to provide this exception.
Or better: Give API developers an AI that is not pretrained on producing nothing.
Could it not verify the source of the document from the official website of the government agency? Of course nothing is certain. But it would be highly likely that the source and content would make it reasonable to conclude the content is not copyrighted.
No.
No, as in it cannot do it or OpenAI won’t take the risk? It’s unfortunate in either case. It forecloses a lot of basic uses. I guess LLMs wont replace some human abilities.
No, as in,