AIs can generate near-verbatim copies of novels from training data

4 hours ago7 min read

The discovery that large language models can spit out near-verbatim copies of entire novels from their training data isn't just a technical footnote—it's a fundamental crack in the foundation of the AI boom. This capability, moving beyond vague stylistic mimicry to direct, line-by-line replication, throws the industry's legal and ethical playbook into chaos.For years, companies have leaned heavily on 'fair use' doctrines, arguing their ingestion of copyrighted material is transformative. But when the output is a carbon copy, that defense crumbles, forcing a stark reckoning with the economic models of human creators.The crisis is multi-modal, echoing in music where artists like SZA have highlighted the disproportionate impact on Black creators, and in visual arts where hobbyist communities feel overrun. In response, we see a fractured corporate landscape: some, like Grammarly, are in retreat, disabling features that used real writers' names, while others, like Meta, are pursuing costly licensing deals like their multi-million dollar agreement with News Corp to legitimize data use.This isn't merely a copyright skirmish; it's a pivotal test of whether our existing legal frameworks, built for a pre-digital age, can govern the scale of AI ingestion and output. The core question, reminiscent of Asimov's debates on robotic ethics, is whether the industry's relentless growth will ultimately come at the expense of the very human creativity it promises to augment, or if a new, sustainable symbiosis can be engineered.

#Here add the most popular tags for this news

#editorial picks

Stay Informed. Act Smarter.

Get weekly highlights, major headlines, and expert insights — then put your knowledge to work in our live prediction markets.

Related News