Anthropic Self-Improving AI Research Surfaces as Lambda Secures $1B for Nvidia Chips
An Anthropic researcher revealed early evidence of self-improving AI, showing that automated systems improved performance across 10 specific behavioral benchmarks without degrading overall model performance, according to a research briefing. The findings offer a rare glimpse into how AI labs are building systems that can iteratively refine their own alignment and safety properties.
Neocloud Lambda has secured $1 billion in private debt to purchase Nvidia AI chips and lease them to Microsoft, underscoring the enormous capital costs fueling the AI infrastructure boom, per Reuters. The financing adds to a growing string of loans aimed at acquiring high-demand GPUs for cloud compute providers.
A federal judge ruled that the Trump administration's blacklisting of Anthropic for being "woke" was illegal, according to The Washington Post. Anthropic had reportedly refused to support lethal autonomous warfare and mass surveillance, prompting the attempted punitive action.
The Trump administration's EPA plans to eliminate a federal rule requiring public notice and comment periods for certain industrial sites, making it harder for communities to weigh in on pollution from new data centers, reports The Washington Post. The move comes as data center construction faces intensifying backlash from neighboring residents concerned about air quality and emissions.
New AI-powered video generation tools have accelerated a surge of AI-produced dramas in China, with 89 of the top 100 animated dramas on Douyin in May being AI productions, according to DataEye via the Financial Times. Chinese actors and online influencers are increasingly left jobless as these programs sweep the industry.
Meta updated its AI glasses to stop recording whenever users cover the built-in safety light, a change designed to reduce nonconsensual recording concerns, according to The Verge. The fix addresses longstanding privacy criticisms surrounding the company's wearable camera devices.
Sources: