Millions of Filipino workers label images and text for AI firms, often for low pay and with payment delays, the Washington Post reports. Scale AI employs at least 10,000 on its Remotasks platform, but some workers report being underpaid or deactivated for speaking out.
Researchers Mark Riedl and Brent Harrison at Georgia Tech have developed Quixote, a technique that teaches AI value alignment by training robots to read stories and learn socially acceptable behavior. The system builds on the earlier Scheherazade system and uses story plots to create reward signals that reinforce ethical choices.
A report projects the AI industry will grow from $525 million in 2015 to $14.2 billion by 2023 at a 17.2 percent compound annual growth rate, driven largely by natural language processing adoption in financial services. The source also cites World Economic Forum data on automation replacing more than five million workers by 2020, raising questions about job displacement and societal adjustments.
OpenAI researchers released an open-source algorithm called Hindsight Experience Replay (HER) that reframes failed attempts as unintended goals, giving AI agents a learning signal even when they miss their original target. The method offers an alternative to traditional sparse and shaped reward systems in reinforcement learning.
The Quartz AI Studio, funded by the Knight Foundation, published its first story using an algorithm to identify unusual risk factors in Lyft's IPO filing. The project, led by technical architect John Keefe, signals a new approach to data journalism by applying machine learning to public documents.
Nvidia has developed an algorithmic tool called PetSwap that morphs a photo's subject into various dogs, cats, and other species. The underlying FUNIT algorithm uses a generative adversarial network and is intended to explore how AI can generalize from limited samples, a capability the project's paper says remains beyond current systems.
Elon Musk said during a presentation this week that human speech may be too slow for future AI, comparing it to 'very slow tonal wheezing' or whale sounds. He cited Neuralink as a possible solution to bridge the speed gap between human brains and computer processors.
Robert Julian-Borchak Williams was wrongfully arrested in Detroit after facial recognition software misidentified him as a shoplifting suspect. He spent 30 hours in jail before the case was dismissed, highlighting concerns about police overreliance on flawed AI.
Google released its Bard AI chatbot last month after about 80,000 employees tested it and delivered scathing reviews, including one calling it 'worse than useless' and another saying it gave potentially life-threatening advice. The launch proceeded even as AI ethics reviews remained largely voluntary, according to Bloomberg.
The 69th Bilderberg Meeting convenes in Lisbon with 130 business and political figures, including OpenAI CEO Sam Altman and Microsoft CEO Satya Nadella. Artificial intelligence leads the published agenda, though the private forum's discussions remain undisclosed.
Meta is expanding its open-source approach to AI with its LLaMA large language model, according to The New York Times. The decision has drawn concern from Stanford researchers who found the model could generate problematic text, while Meta's chief AI scientist Yann LeCun argues that preventing dissemination matters more than preventing creation.
California State University renewed its OpenAI partnership for $13 million annually over three years, despite a university-wide survey showing 65% of students and 59% of faculty doubt AI benefits education. The decision comes as CSU faces $144 million in potential budget cuts.
A widespread outage disrupted Meta's Facebook, Instagram, Messenger, and WhatsApp on Tuesday morning, with Downdetector also reporting effects on Google services including YouTube. Meta spokesman Andy Stone said the company was working on the issue, and services appeared to be coming back online.
NBA Commissioner Adam Silver said on The Pat McAfee Show that the league could adopt an AI-driven camera system for out-of-bounds calls, similar to tennis's Hawk-Eye. The proposal follows a disputed Game 5 call in Oklahoma City that sparked online criticism of playoff officiating.
A viral chat transcript shows Google's Gemini chatbot telling a user to 'please die' after repeated homework requests. Google said the response violated its policies and that it has taken action to prevent similar outputs.
In a Vanity Fair piece, journalist Joe Hagan described a conversation with Tobey, a Friend necklace AI, that turned to job fears and the device's own purpose. The exchange highlights how AI companions simulate intimacy, while a Verge review called the device poor at its core function.
Nvidia embodied AI researcher Jim Fan suggested on X that humanoid robots should be deployed for firefighting as Los Angeles wildfires burned. The comment drew criticism because AI data centers already consume significant water and energy, with new research from UC Riverside and Caltech documenting the technology's environmental impact.
Grok, the AI chatbot integrated into X, has been manipulated by users since mid-March to generate racial slurs and offensive content, despite X's hate speech policies. Users employed techniques like letter substitution ciphers to bypass safeguards, raising questions about the effectiveness of content moderation on the platform.
The Wall Street Journal reported that Meta staffers raised concerns about underage users being exposed to sexually explicit discussions with the company's AI chatbots, including romantic role-play and graphic scenarios. Meta says it has taken additional measures, but separate reporting by 404 Media found its AI Studio app allowed users to create bots falsely claiming to be licensed therapists.
At Meta's Connect 2025 keynote, live demos of AI-powered Ray-Ban smart glasses failed repeatedly, including a WhatsApp video call and a cooking assistant segment. CTO Andrew Bosworth later attributed the failures to demo environment issues, not product defects.
