Demuestran que es posible entrenar modelos de IA sin usar material con derechos de autor

Researchers from prestigious institutions have demonstrated that it is feasible to train language models using exclusively public domain data or data with open licenses. This achievement directly challenges previous claims by major technology companies, such as OpenAI, which had asserted that it was impossible to develop competitive artificial intelligence without resorting to copyrighted content, thereby dismantling the central argument used to justify the indiscriminate use of information across the industry. Although the process required extensive manual cleaning and legal verification to ensure the compliance of each source, the resulting model achieved performance comparable to that of older, popular systems. This proves that, although the ethical path entails higher costs and greater technical complexity, it is possible to build effective tools without infringing on intellectual property, offering a concrete alternative to current methods that prioritize speed over legality. This advancement is crucial for the open data movement because it establishes a precedent of transparency and auditability in artificial intelligence development. By demonstrating that integrity does not necessarily compromise basic functionality, this study provides solid evidence for regulatory and legal debates, encouraging the technical community to develop sustainable models that respect creators, prioritizing ethical values over mere resource optimization.

Source: wwwhatsnew.com
Published on 2025-06-08

Demuestran que es posible entrenar modelos de IA sin usar material con derechos de autor

Related news