Google unveils TurboQuant as Anthropic faces lawsuit

Key Takeaways

• Google introduced TurboQuant, an AI compression method that reduces AI memory use by up to 6 times and boosts speed by up to 8 times for models like Gemma and Mistral.
• China is investigating Meta's acquisition of AI startup Manus, preventing co-founders Xiao Hong and Ji Yichao from leaving the country due to potential violations of investment and technology export rules.
• Openreach utilizes Google Cloud's Vertex AI, BigQuery, and Gemini Enterprise to optimize its 24,000-van fleet, reduce emissions, and plan broadband expansion.
• Music publishers, including Universal Music Group, are suing Anthropic, alleging illegal use of millions of song lyrics to train its Claude AI model without permission.
• A NOAA official expects AI to revolutionize weather forecasting by speeding up predictions, emphasizing the continued need for solid scientific knowledge in model training.
• Experts advocate for AI to prioritize societal and worker benefits, proposing policies for worker training and co-design of AI systems to combat underpaid work.
• Florida State University expert Paul Marty highlights human skills like creativity, passion, and curiosity as crucial differentiators in the age of AI.
• Solink launched AI-powered VerifEye features that use vision language models to filter false security alarms, saving thousands of hours of manual monitoring and improving threat prevention.
• Boxer Moses Itauma partnered with Datavault AI, a company developing a Name, Image, and Likeness (NIL) digital asset exchange platform slated for launch in the second half of 2026.
• White House science adviser Michael Kratsios stated that Trump's AI policy framework will not increase energy costs and aims to protect creators' work from unauthorized commercial AI use.

Google's TurboQuant cuts AI memory use by 6x

Google developed a new AI compression method called TurboQuant. It reduces the memory needed for AI models, specifically the key-value cache, by up to 6 times without losing quality. TurboQuant uses techniques like PolarQuant and QJL to convert data into a more efficient format. Tests showed it works well on models like Gemma and Mistral, making AI faster and cheaper to run.

Google's TurboQuant shrinks AI memory by 6x, boosts speed 8x

Google announced TurboQuant, a new algorithm that significantly lowers the memory needed for large language models (LLMs) and increases their speed. It can reduce memory use by up to 6 times and improve performance by up to 8 times. TurboQuant uses PolarQuant and QJL to make AI models more efficient. This breakthrough could lower the cost of running AI systems and potentially impact memory manufacturers' demand.

Google's TurboQuant AI compression gets 'Pied Piper' nickname

Google has introduced TurboQuant, a new AI memory compression algorithm that can reduce AI's working memory by up to 6 times. This technology, which is being presented at the International Conference on Machine Learning, could make running AI cheaper. While it targets inference memory, it does not address the large memory needs for AI training. The algorithm has earned the nickname 'Pied Piper' online, referencing a fictional compression technology from the TV show 'Silicon Valley'.

China restricts Meta-bought AI firm founders from leaving

China has told two co-founders of the AI startup Manus, recently acquired by Meta, not to leave the country. Authorities are reviewing the acquisition to see if it violates Chinese investment and technology export rules. The co-founders, Xiao Hong and Ji Yichao, were informed of this guidance after a meeting with officials. Manus develops AI agents capable of research and creating presentation slides.

China bars Manus founders from leaving after Meta acquisition

China has prevented Manus co-founders Xiao Hong and Ji Yichao from leaving the country as regulators investigate Meta's acquisition of the AI startup. The National Development and Reform Commission is examining potential violations of Chinese investment and technology export rules. Manus, which launched in March 2025, develops AI agents that can perform multi-step tasks like research and planning. The company had relocated its headquarters to Singapore before the acquisition.

Openreach uses Google Cloud AI for faster internet and lower carbon use

Openreach is using Google Cloud's AI tools, Vertex AI and BigQuery, to improve its operations and reduce environmental impact. By analyzing data from its fleet of 24,000 vans, Openreach is optimizing routes, speeding up the transition to electric vehicles, and cutting emissions. The company has also created a digital replica of UK transportation corridors to plan faster broadband expansion. Gemini Enterprise is being used to make engineering tasks more efficient.

FSU expert discusses AI literacy and human skills

Paul Marty, an expert in academic innovation at Florida State University, emphasizes the importance of human skills in the age of AI. He believes that while AI can remix existing knowledge, human creativity, passion, and curiosity are what set people apart. Marty also highlights that universities should focus on engaging students and teaching them how to apply knowledge to solve problems. He suggests that focusing on human skills like communication and empathy is crucial for students' future success.

Boxer Moses Itauma partners with Datavault AI

Undefeated boxer Moses Itauma has partnered with Datavault AI, a company developing a Name, Image, and Likeness (NIL) digital asset exchange platform. Itauma will feature the Datavault AI brand during his upcoming fight against Jermaine Franklin on March 28 in Manchester. This collaboration aims to provide global exposure for Datavault AI and explore how its technology can help athletes manage their digital legacies. The partnership could extend to future participation in Datavault AI's platform, set to launch in the second half of 2026.

NOAA official: AI will transform weather forecasting

A NOAA official stated that artificial intelligence is expected to revolutionize weather forecasting. While AI and machine learning are speeding up predictions, experts emphasize that solid scientific knowledge is still essential for training these models. The agency sees significant promise in AI's ability to improve the speed and accuracy of weather predictions.

Future of work: AI should prioritize people, not profits

A new vision for the future of work suggests that artificial intelligence should be used to benefit society and workers, not just increase corporate profits. Experts propose policies that support worker training, enhance roles in caregiving, and encourage co-design of AI systems. This approach aims to combat the trend of work becoming increasingly underpaid and overworked. The AFL-CIO will host a summit to ensure workers have a voice in shaping AI policies.

Music publishers sue Anthropic over AI training data

Music publishers, including Universal Music Group, are suing AI company Anthropic, arguing that it illegally used millions of song lyrics to train its Claude AI model. They are asking a judge to rule that Anthropic's use of the lyrics was not 'fair use.' The publishers claim Anthropic copied copyrighted lyrics on a massive scale without permission or payment, creating AI products that compete with human songwriters. They state this case is not against AI technology itself but against Anthropic's unauthorized use of their work.

Solink AI enhances security, cuts costs with VerifEye

Solink has launched new AI-powered features for its VerifEye system to help security teams manage burnout and reduce operational costs. VerifEye uses vision language models to filter out false alarms, saving thousands of hours of manual monitoring. It prioritizes high-risk events, allowing human operators to focus on real incidents. This system not only improves threat prevention but also provides broader business visibility by integrating access control, loss prevention, and operations data.

Trump AI policy won't raise power bills says advisor

White House science adviser Michael Kratsios stated that President Trump's AI executive orders and policy framework will not increase energy costs. He mentioned that data center companies have agreed to build or secure their own power sources. Kratsios also highlighted that the AI framework aims to protect creators by ensuring their work is not used for commercial AI purposes without permission or compensation. The White House is working to implement these policies quickly.

Sources

• Google says new TurboQuant compression can lower AI memory usage without sacrificing quality
• Google Announces TurboQuant, A New Compression Algorithm That Reduces LLM Memory Requirements By 6x And Increases Speed by 8x
• Google unveils TurboQuant, a lossless AI memory compression algorithm — and yes, the internet is calling it 'Pied Piper'
• Leaders of AI Firm Bought by Meta Are Restricted From Leaving China
• China Bars Manus’ Founders From Leaving Country After Acquisition By Meta
• Openreach Taps Google Cloud AI to Accelerate High-Speed Internet Access and Cut Carbon
• FSU expert available for interviews on AI literacy
• Undefeated 21-year-old boxer Moses Itauma lands Datavault AI deal
• NOAA Official Says AI Will Overhaul Future of Weather Forecasting
• A people-first vision for the future of work in the age of AI
• Music Publishers Say Anthropic AI Claude Training Violated Copyrights
• Solink Gains Momentum with New AI-Powered Capabilities to Solve Security Burnout and Reduce Operational Costs
• Exclusive: Kratsios says Trump AI push won't raise power bills