News for hackers. Stuff that matters.
Re:OPT: Open Pre-trained Transformer Language Models by londons_explore on Tuesday May 03, 2022 @12:15AM (#31244533)
It's super easy to watermark weights for ML models.

Just add a random 0.01 to a random weight anywhere in the network. It will have very little impact on the results, but will mean you can identify who leaked the weights.

Re:OPT: Open Pre-trained Transformer Language Models by dotnet00 on Tuesday May 03, 2022 @01:40AM (#31245034)
It should be easy enough to make that sort of signature very difficult to trace by simply adding a bunch of small noise to the network overall, or even simply training for a few iterations.
Re:OPT: Open Pre-trained Transformer Language Models by zelphirkalt on Tuesday May 03, 2022 @03:13AM (#31245593)
The person leaking it might not do so intentionally. Their computer might be compromised. Are we going to punish people for not being cybersec experts?
Re:OPT: Open Pre-trained Transformer Language Models by ipaddr on Tuesday May 03, 2022 @12:30AM (#31244632)
Compare two copies.
Re:OPT: Open Pre-trained Transformer Language Models by vladf on Tuesday May 03, 2022 @01:35AM (#31245013)
Slightly modify a million random weights by changing the least significant bit up or down.
Re:OPT: Open Pre-trained Transformer Language Models by ipaddr on Tuesday May 03, 2022 @01:40AM (#31245029)
Compare three copies.
Re:OPT: Open Pre-trained Transformer Language Models by ALittleLight on Tuesday May 03, 2022 @01:56AM (#31245140)
Or slightly randomly modify all the parameters on the copy you distribute, then it will be a match for nobody.
Re:OPT: Open Pre-trained Transformer Language Models by ipaddr on Tuesday May 03, 2022 @12:52PM (#31250544)
You compare all three and average the variance of each value. So the more copies the better.
Re:OPT: Open Pre-trained Transformer Language Models by bogwog on Tuesday May 03, 2022 @05:11PM (#31253644)
...or just steal it so that even if it can be traced, it's not your problem.
Re:OPT: Open Pre-trained Transformer Language Models by vladf on Wednesday May 04, 2022 @11:39AM (#31262007)
In fairness to ipaddr, this can result in worse performance at this point.
Re:OPT: Open Pre-trained Transformer Language Models by vladf on Wednesday May 04, 2022 @11:31AM (#31261893)
OK, this is a fun game. I think your counterattack assumes I'm picking these million weights uniformly randomly among the 175 billion. I modify my original answer: s/a million/half the weights in a deterministic subset of 2 million weights/

Select the deterministic subset by just hashing some identifier for each weight.

For any reasonable number of copies, there's a pretty unique subset between all your copies sharing a large amount of bits flipped in the same direction among this subset.