10.5.26 | Run Qwen 3.8 Flash Next RTX 4090 at 100T/s, turn off Apple Intelligence on macOS 27 for disk space, improper redaction reveals Google Data Center water and electricity usage

SPEAKER_1 (0:00)

Welcome to the Hacker News Highlights, where we explore the top 10 posts on Hacker News every day.

Today, we dive into running a 125 billion parameter AI on consumer hardware, turning off Apple Intelligence to reclaim disk space, and improperly redacting data center water and electricity use. Let's dive in.

Title, run Qwen 3.8 Flash Next 125B on consumer hardware, Rtx 4090, at 100T slash S source, GitHub. The post detailed how Strata software enables running Qwen 3.8 Flash Next, a large 125 billion parameter AI model on standard gaming PCs with consumer graphics cards like an Rtx 4090 It provided benchmarks showing fast token processing speeds, reaching up to 124 tokens per second, and explained the hardware requirements and setup process.

The project aims to make powerful AI models usable on cost-effective, everyday computers without needing server-grade hardware. In the comments, community members mostly supported the achievement, emphasizing that it makes running large models more accessible and practical. Users discussed the performance compared to other models like Qwen 3.8 27B and Qwen 3.6 35B, debated quantization levels and their impact on accuracy, and shared experiences running the same software on various hardware. There was a general consensus that low VRAM and affordable hardware can now handle models previously thought to require expensive servers, though some raised concerns about potential accuracy loss with aggressive quantization. Overall, the community expressed excitement about the progress in local AI inference and the potential for wider adoption.

Title, Turn Off Apple Intelligence on macOS 27 and Get Its Disk Space Back. Source, GitHub. The post explained how macOS 27 no longer has a single switch to turn off Apple Intelligence, but a script called Remove Mac AI can disable the features, delete related models, and prevent re-downloading. The script also allows users to revert changes and manage features like Siri, writing tools, notes, and notifications. It highlights that the models stay on disk after being turned off, taking up significant space, which the script helps to reclaim, especially on devices with limited storage. In the comments, the community was mostly critical of the Curl Bash installation method, debating trust and security practices. Many pointed out the risks of running unverified scripts, comparing it unfavorably to package managers like NPM or PIP, which also have similar security considerations. Key topics included concerns over privacy, disk space, and the bloating of macOS with AI models and background processes. Some users argued that the lack of an easy option to disable or remove these features is user hostile, especially on storage-constrained devices, while others defended Apple's approach or expressed frustration with the overall trend toward system bloat. The community generally felt that users should have more control and transparency over pre-installed features and AI models.

Title, Improper Redaction Reveals Google Data Center Water and Electricity Usage, Source, Qon. The article detailed how Nebraska data centers, including Google's sites, claimed their water and electricity data as trade secrets to prevent public disclosure. However, redacted information was uncovered by using a cursor to reveal that Google's Lincoln data center reported a peak of 52.65 megawatts of electricity and approximately 13 million gallons of water annually, used for cooling and site operations. The total water usage of all six reporting data centers last year was around 765 million gallons, enough to fill over a thousand Olympic-sized swimming pools, with the largest single site expected to use nearly 547 million gallons. Google also reported expecting a tax refund of over 55 million dollars for 2025, raising questions about transparency and the true impact of these facilities. In the comments, the community largely debated the significance of water and energy usage, with most seeing the reported figures as relatively small compared to other industrial uses like agriculture or power plants. Many compared the water consumption to local farmland or household use in Nebraska to better understand scale. There was concern that the companies hide these figures behind shell companies and NDAs, which hinders public oversight. Several users argued that while water use might not be hugely impactful locally, the broader issues of industrial resource consumption, environmental impact, and lack of transparency remain important. Community members expressed skepticism about the claims of low resource use and criticized how regulatory and public relations tactics are employed to avoid scrutiny, while others highlighted the environmental and societal costs associated with the growth of data centers and AI infrastructure. Overall, community sentiment was mostly critical of corporate opacity and environmental implications, emphasizing that broader systemic issues need public attention.

Title, What is Going On with Ceiling Fans, Source, Tumblr. The post discusses the recent surge in unconventional and often bizarre ceiling fan designs, especially those with integrated LED lighting and sculptural elements. It highlights how this trend is driven by cheap overseas manufacturing, the proliferation of disposable products, and a desire for decorative appeal over functionality. The author also examines the long-term issues with non-replaceable LEDs, poor design choices, and the rise of smart remote-controlled and integrated fixtures that complicate maintenance and usability.