9.19.26 | Android 17 adds new APIs without AOSP, Cloudflare Quick Tunnels, OpenJev

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 Android 17 adding new APIs, Cloudflare enabling public tunnels from laptops, and OpenJev allowing decision modeling and browsers. Let's jump in.

Title, Android 17 is the first since three. X to add new APIs without releasing to the AOSP. Source, grafinius.social. The news story discussed how Android version 17 introduced new APIs that are only available on Google's Pixel devices, bypassing the standard Android open source project releases. This move marks the first time since version 3 that new APIs were added exclusively to Pixel hardware without being made available via the broader AOSP. Community members expressed concern that this creates an unfair advantage for Pixels and undermines open source principles, with some criticizing Google's increasing control over Android updates and features. In the comments, the community largely supported the view that Google's move to restrict new APIs to Pixels is anti-competitive and damaging to the open source ethos of Android.

Users debated whether this helps or harms device diversity, security, and user choice. There was concern that Google's strategy favors Pixels at the expense of other OEMs and could lead to a future where Android's open source foundation becomes sidelined. A common theme was skepticism about Google's claims of supporting open source and the hope that alternative projects like Graphene OS and custom ROMs can counterbalance this trend. Overall, the tone was largely critical of Google's increasing control of Android development and their departure from open source commitments.

Title, Cloudflare Quick Tunnels. Source, tridotcloudflare.com. The post described Cloudflare's new feature called Quick Tunnels, which allows users to turn their local servers into public URLs without needing an account, DNS setup, or open ports. The process involves a simple command that connects your local machine to Cloudflare's edge network providing an encrypted DDoS filtered URL accessible worldwide within seconds. The story emphasized that no configuration or registration is necessary and highlighted the service's ease of use and fast setup time, making it appealing for testing, demos, and quick sharing. In the comments, the community mostly expressed skepticism about the product's marketing and its actual utility. Many noted that Quick Tunnels are not new and have existed for a few years, with some pointing out they could be misrepresented as a new feature. Several users raised concerns about security, potential abuse, and the centralization risks associated with routing traffic through Cloudflare's infrastructure.

Others discussed technical limitations such as the lack of persistent URLs, bandwidth caps, restrictions on streaming, and the possibility of traffic being easily discoverable or blacklisted. A recurring theme involved distrust toward Cloudflare's business model, with worries about surveillance, data security, and reliance on a large corporation for personal or small-scale hosting. Overall, the consensus was that while the service is convenient for quick testing or temporary projects, it raised questions about security and the implications of centralizing such traffic under Cloudflare's control.

Title, OpenJev introduces a new approach for decision-making with fast, calibrated, non-generative classification models. Source openjev.com. The Post discussed a project called OpenJev, which offers a model that responds to questions by quickly providing probabilities instead of generating text. The aim is to improve decision tasks like classification and ranking, making them faster and more reliable than traditional language models. Users debated whether OpenJev is truly novel or just an optimized classifier, with most agreeing it significantly improves speed and certainty, but lacks transparency about its architecture. There was also skepticism about the hype. With some pointing out it resembles existing schemes like schemas and classifiers, but with better calibration.

In the comments, the community mostly viewed OpenJev's approach as a useful evolution rather than a groundbreaking innovation. Most agreed it's capable of very fast calibrated answers that could replace slower, more expensive language models for decision tasks. Some questioned the technical novelty, comparing it to established methods like classification schemas and BERT-based approaches. Several users discussed the architecture, stressing that it resembles a classifier rather than a generative model and expressed interest in open-source implementations and benchmarking. There was consensus that the community appreciated the speed and calibration, but skepticism about whether it could fully replace more complex models or handle ambiguous scenarios. Overall, the sentiment was largely positive about its practical value, but cautious about claims of novelty.

Title, Saving Another 100TB of RAM. Source, Cloudflare Blog. The article detailed how Cloudflare reduced the memory footprint of a service called Pingora by making small algorithm improvements. They focused on optimizing how consistent hashing works, especially for their internal load balancer, Pingora backend router. Adjustments included storing hashes more efficiently and reducing the number of hashes used per server, which led to reclaiming over 100TB of RAM globally. The story emphasized that small mathematically-backed tweaks at large scale can lead to massive resource savings. In the comments, the community was mostly supportive, appreciating the deep technical analysis and math. Some debated whether these optimizations are only practical at Cloudflare scale or applicable broadly. Others discussed the potential impact of such RAM savings on cloud costs and the future of software engineering, with some suggesting that the focus on tiny improvements could push companies toward more efficient, lean systems. There was also curiosity about alternative hashing methods and schemes that might outperform the approach described. Overall, the community appreciated the technical depth and practical insights, but expressed varied opinions on the scalability and general applicability of these optimizations.