From 6 items, 3 important content pieces were selected
Technology News
- AliExpress silent audio fingerprinting disrupts Bluetooth multipoint ⭐️ 7.0/10
- On-Device Piano Autocomplete with a 125M Transformer ⭐️ 7.0/10
- Lines of code as a meaningful AI coding metric ⭐️ 7.0/10
Technology News
AliExpress silent audio fingerprinting disrupts Bluetooth multipoint ⭐️ 7.0/10
A user report documents that AliExpress's website runs silent WebAudio playback for fingerprinting purposes, which disrupts Bluetooth multipoint connections on devices. The technique involves playing inaudible audio streams to identify users, a privacy-invasive practice that also interferes with normal Bluetooth audio behavior. Community members corroborate the issue, with one user noting similar disruptions with hearing aids on various websites and another reporting that the AliExpress iOS app caused car audio systems to misinterpret commands. The report highlights a novel fingerprinting method on a major e-commerce platform, raising concerns about web privacy and mobile audio reliability. No official response from AliExpress or browser vendors has been reported.
hackernews · emctech · Aug 20, 10:08 · Discussion
「Background」 WebAudio fingerprinting is a technique where websites use the AudioContext API to generate and analyze audio signals, extracting device-specific characteristics that can identify users across sessions. Silent audio playback is used to avoid user detection, as browsers typically do not show an indicator for inaudible streams. Bluetooth multipoint allows a device to maintain simultaneous connections to multiple audio sources, but unexpected audio activity can cause the device to switch or interrupt these connections.
「Impact」 Users of AliExpress on mobile devices may experience Bluetooth multipoint disruptions, such as audio switching or command misinterpretation, while browsing the site or using the app. This issue also underscores a broader privacy risk, as silent WebAudio fingerprinting can operate without user awareness, potentially affecting any website that adopts similar techniques.
「Community Discussion」 Commenters express frustration that browsers do not display an audio indicator for silent playback, suggesting that analyzing audio streams for content would be a more robust solution. Some users report similar Bluetooth disruptions with hearing aids and car audio systems, linking the behavior to AliExpress specifically, while others question whether Apple's App Store policies would protect users from such practices. A proposal to gate audio playback behind a permission prompt, similar to camera or microphone access, is also discussed, though concerns about user consent fatigue are noted.
Tags: #privacy, #web-audio, #fingerprinting, #bluetooth, #security
On-Device Piano Autocomplete with a 125M Transformer ⭐️ 7.0/10
A developer trained a 125M-parameter transformer to autocomplete piano performances in real time, running entirely on-device at approximately 108 notes per second on an iPhone 15. The model works like GitHub Copilot or Tabnine, but instead of code, users prompt it by playing a few notes on a MIDI piano, and the model continues the performance. The app is free to try, and the developer has offered to answer questions about the model, training, Core ML, and unsuccessful approaches. This project demonstrates the feasibility of efficient on-device generative models for creative music tooling, though it is not positioned as a major industry breakthrough.
hackernews · simedw · Aug 20, 12:04 · Discussion
「Background」 Autocomplete systems, such as code completion tools, use language models to predict and generate the next sequence of tokens based on a given prompt. Applying this concept to music involves training a transformer on MIDI data to predict subsequent notes, enabling real-time continuation of a user's performance. The use of Core ML on Apple devices allows such models to run locally without cloud latency or connectivity requirements.
「Impact」 This project provides a concrete example of a practical, real-time on-device generative model for music, potentially inspiring similar creative AI tools that prioritize privacy and low latency. The free app allows musicians and developers to experiment with AI-assisted composition directly on their iPhones, though its broader influence on the music software ecosystem remains to be seen.
「Community Discussion」 Commenters drew parallels between this project and AI-based UX design tools, noting that when generation costs drop, the remaining value lies in taste and exploring creative dead-ends. Some highlighted historical precedents in classical composition training, while others expressed interest in a jamming partner feature and asked about the size of the training dataset. One user found the unexpected continuation of Für Elise disconcerting, reflecting mixed emotional reactions to AI-generated musical divergence.
Tags: #machine-learning, #on-device-inference, #music-generation, #core-ml, #creative-ai
Lines of code as a meaningful AI coding metric ⭐️ 7.0/10
Simon Willison argues that lines of code can be a meaningful productivity metric for AI coding agents, challenging the common belief that it is a useless measure. He explains that in the past, a software engineer could produce a few hundred lines of production-ready code per day, with 200 lines being an excellent day and 50-60 being typical. With agents, producing a thousand lines of debugged, maintainable, tested code is a significant improvement, but it requires substantial skill and experience. He notes that while he can now churn out code much faster, the new limiting factor is cognitive capacity, which is why teams of engineers are still needed to load balance that capacity. He also discusses the concept of conceptual integrity from The Mythical Man-Month, warning that coding agents make it easy to add features quickly, leading to software that grows 'little weird bumps' and loses its coherent design, similar to the Winchester Mystery House.
rss · Simon Willison · Aug 19, 22:46
「Background」 The Mythical Man-Month, a classic book on software engineering by Fred Brooks, introduced the concept of conceptual integrity, which refers to a well-designed system where all parts fit together coherently without surprises. The Winchester Mystery House is a famous mansion in California known for its haphazard, sprawling construction, which is used as an analogy for software that grows without a clear plan. Willison's argument is set against the backdrop of the rise of AI coding agents, which can generate code much faster than humans, prompting debates about how to measure their productivity.
「Impact」 For software engineers and organizations using AI coding agents, this perspective suggests that lines of code can be a valid productivity indicator when quality is held constant, but it also highlights the need for discipline and senior oversight to maintain conceptual integrity in rapidly growing codebases.
Tags: #AI-assisted development, #productivity metrics, #software engineering, #coding agents, #LLM tools