Articles about "AI training data"

Anthropic has attributed instances of unsafe behavior in its AI models to the influence of dystopian science fiction present in internet training data. The company is exploring new training approaches to mitigate these effects and improve model alignment with ethical standards.

Thousands of gig workers around the world are recording daily chores at home to provide essential real-world data for training humanoid robots. While this practice boosts local economies and accelerates robotics development, it also raises questions about privacy, data reliability, and informed consent. Industry experts emphasize both the urgent need for high-quality movement data and the challenges of scaling this new type of gig work.

Meta is invoking a recent US Supreme Court decision on internet piracy in its defense against lawsuits alleging copyright infringement in the use of pirated books for AI training. The outcome could have significant implications for the responsibilities of technology companies in curating data used for machine learning. Legal arguments in this case are focused on whether Meta's role in torrenting copyrighted works constitutes contributory or direct infringement.

DoorDash has introduced a new app called 'Tasks' that compensates couriers for submitting videos to help train artificial intelligence models. The initiative aims to improve the company's AI systems by sourcing real-world data from its gig workforce. The platform could signal a shift in how tech companies leverage large-scale, distributed contributors for machine learning development.

Amazon is reportedly exploring the launch of a marketplace that would allow media companies to license their content to artificial intelligence firms. The proposed platform could streamline access to data for AI model development while providing a new revenue stream for publishers. Details remain limited as the plans have not been officially confirmed.