AI firms pulping rare books to feed models, outrage grows

The news broke that several AI companies, in their insatiable hunger for training data, have been purchasing and destroying rare, often one-of-a-kind physical books. The books, which include historical manuscripts, out-of-print works, and culturally significant texts, are being cut apart and scanned at high speed to create digital copies, after which the originals are discarded. This practice has drawn sharp criticism from librarians, historians, and the public, who see it as an act of cultural vandalism. The companies argue that digitization preserves the content, but experts counter that the physical artifact itself holds value—marginalia, binding, provenance—that is lost forever. The backlash has been intense, with many calling for legal restrictions on such destruction. For the AI industry, this is a public relations disaster, and it raises ethical questions about the lengths companies will go to in order to secure data. For readers and researchers, it's a stark reminder that the AI boom has real-world consequences beyond the digital realm.

An investigative report has revealed that several major AI companies are using complex financial structures to hide billions in debt, painting a rosier picture for investors than reality. The debt is tied to massive compute costs, data licensing fees, and acquisitions, and companies are reportedly shifting liabilities to off-balance-sheet entities. This news has sent ripples through the investment community, as it suggests that the AI boom may be built on shakier ground than previously thought. The report details how companies are 'kicking the can down the road' by refinancing and leveraging future revenue projections, which may not materialize. For the industry, this could mean a coming correction, with layoffs and consolidation. For startups, it means that funding may dry up as investors become more cautious. The report urges regulators to scrutinize AI companies' financial disclosures more closely. While some insiders dismiss the claims as alarmist, the sheer scale of the debt, if confirmed, could have systemic implications.

Mozilla's inaugural 'State of Open Source AI' report, released today, provides a comprehensive look at the open-source AI landscape. The report finds that while open-source models like Llama and Mistral have made impressive strides, they still trail proprietary models like OpenAI's GPT-5.6 and Google's Gemini on complex reasoning tasks. However, open-source models are closing the gap in efficiency and customization, and they dominate in local deployment and privacy-sensitive applications. The report also highlights the growing ecosystem of tools and frameworks that make open-source AI more accessible, which is driving adoption among startups and enterprises. Mozilla calls for more investment in open-source AI to ensure a healthy, competitive market. The report has been praised for its depth and neutrality, but some critics argue it underplays the commercial pressures that favor closed models. For developers, the report is a valuable resource for deciding which models to build on, and it underscores the importance of community-driven AI development.