LAION is also training, but that's not part of the lawsuit so the court may (or may not) have to limit their analysis to that part. The judges indicated they would limit the scope for the analysis of uses, but relevant downstream factors should be taken into account per caselaw!
REGULATION
-
Bing AI search impact on website access and content protection
By
–
Bing approach would force websites to turn on protection + offer exclusive access through their AI assistants potentially
-
LAION Lawsuit: Plaintiff’s Claim Coverage Limitations Identified
By
–
The problem we discussed above, is that the plaintiff filed a claim that does not and cannot cover the whole of LAION's activities. That was a mistake.
-
Data Laundering: How Non-Profits Exploit Legal Loopholes Systematically
By
–
Data laundering is a tricky one because legislators are inclined to look at each event alone… then fail to see the big picture of non-profits doing otherwise illegal activities. It's the fallacy of composition: look all these small events are legal, so the whole thing must be?
-
LAION Image Embeddings: Legal Implications of Lossy Reproduction
By
–
LAION computes embeddings, they are compressed forms of the images, could be easily argued to be another lossy Reproduction. (If Robert's lawyers had done their job well, this would have come up.)
-
Embeddings as Derivatives: TDM Exception Legal Analysis
By
–
I think this is a technical mistake: an embedding, as used in the comparison, is a compression of the original input through a neural model that can be reconstructed. It's a derivative! TDM exceptions do now allow creating derivatives either, let alone disseminating them.
-
Text Mining vs Generative AI: Research Tool Distinction
By
–
That definition makes it clear TDM is intended as research tool, intended to produce only statistics, not extract expressions into a neural database. There is a qualitative difference in ML being used for mere analysis and research, and ML being used for generative applications!
-
TDM Definition and Legislative Intent in AI Regulation
By
–
I think there's a strong argument for this, based on definition of TDM which is "analysis" and "analytical" only (which imply no downstream uses except "research") — nothing to do with reproducing inputs or automating labour. But it does go against the intuition of legislators.
-
Copyright Traps Paper Enhances Membership Inference Attacks Detection
By
–
Quoted in nice article by @Melissahei
, on a #ICML2024 paper on "copyright traps" by @matthieu_meeus @ffuuugor @ManuelFaysse @yvesalexandre Cool idea to enhance the efficacy of membership inference attacks to detect copyrighted data. Caution tho: similar to Glaze, not futureproof -
Fair Use Loopholes in AI Training Dataset Selection
By
–
I get that too. Fair Use and exceptions depend on "Use" and "purpose" — I hope they consider *all* of the uses not just a couple seeded by LAION that look innocent. (If not, that's another loophole!)