News for hackers. Stuff that matters.
Re:An embarrassingly simple approach to recover unlearned kn... by seanmcdirmid on Monday November 04, 2024 @10:03AM (#42042146)
I thought it was even worse than that: learning any of the corpus verbatim would actually reduce model utility.
Re:An embarrassingly simple approach to recover unlearned kn... by edude03 on Monday November 04, 2024 @10:09AM (#42042215)
Yes although how close to verbatim is debatable. For example there are questions that you’d ask that other people have asked many times before that you’d like the exact answer for (e.g. when does daylight saving time end?)
Re:An embarrassingly simple approach to recover unlearned kn... by startupsfail on Monday November 04, 2024 @11:50AM (#42043400)
> that make it forget specific facts. These are often meant to satisfy copyright claims

Facts are not copyrightable.

To quote copyright.gov: “Copyright does not protect facts, ideas, systems, or methods of operation, although it may protect the way these things are expressed.”

Re:An embarrassingly simple approach to recover unlearned kn... by PittleyDunkin on Monday November 04, 2024 @01:14PM (#42044407)
What is a fact without expression? It's not clear under what interpretation might be necessary to get the quoted sentiment to be considered sensical.
Re:An embarrassingly simple approach to recover unlearned kn... by seanmcdirmid on Monday November 04, 2024 @12:34PM (#42043938)
Wouldn't that be stuffed in the prompt anyways? No reason for the LLM to learn that.
Re:An embarrassingly simple approach to recover unlearned kn... by int_19h on Monday November 04, 2024 @12:30PM (#42043877)
It really depends on which part of the corpus, though. I do expect my LM to be able to reproduce culturally important citations, for example.