Better than Nothing
In late June we noted that the court case involving AI model builder Anthropic’s (pvt) use of copyrighted books to train its models had gone primarily in Anthropic’s favor, as the judge found that using the works to train the LLMs was transformative, essentially pointing out that statistically mapping text fragments is fundamentally different from copying. He compared it to how humans learn to create new works from copyrighted material and deemed it acceptable as a ‘fair use’ of the material. However the judge also noted that Anthropic had used pirated copies of books to fill its ‘central library’ and found that such was not a ‘fair use’ and could decrease the value of the works to its authors.
The judge has now given preliminary approval to a $1.5b settlement to the class-action suit, which is thought to cover ~500,000 works. This amounts to ~$3,000 per work, four times the minimum statutory damage rate, but far below the maximum ($150,000 for ‘willful’ damage), although if the infringer was found to be ‘innocent’ of being aware that they were infringing, the minimum drops to $200/work. This was not the case. While Anthropic initially filed an interlocutory appeal to a higher court concerning the pirated works ruling, it was denied and this settlement was reached, closing the case, although other legal challenges to AI ‘fair use’ remain in the courts.
The judge has now given preliminary approval to a $1.5b settlement to the class-action suit, which is thought to cover ~500,000 works. This amounts to ~$3,000 per work, four times the minimum statutory damage rate, but far below the maximum ($150,000 for ‘willful’ damage), although if the infringer was found to be ‘innocent’ of being aware that they were infringing, the minimum drops to $200/work. This was not the case. While Anthropic initially filed an interlocutory appeal to a higher court concerning the pirated works ruling, it was denied and this settlement was reached, closing the case, although other legal challenges to AI ‘fair use’ remain in the courts.
RSS Feed