Interactive Smart Cake: A Camera Disguised as Dessert

2025-03-22
Interactive Smart Cake: A Camera Disguised as Dessert

This article details the creation of an interactive smart cake, a collaboration between the author and a confectionery roboticist. The cake, designed to resemble a camera, incorporates a camera, LEDs, a thermal printer, and other interactive components. It detects faces, takes photos, and prints personalized receipts. The article provides a detailed walkthrough of the hardware, software, code, and assembly process, culminating in a delicious and photographically capable cake!

Read more

Starlink Poised to Steal $2.4B Air Traffic Control Contract from Verizon

2025-02-28
Starlink Poised to Steal $2.4B Air Traffic Control Contract from Verizon

Starlink, Elon Musk's SpaceX subsidiary, is reportedly on the verge of snatching a $2.4 billion contract for upgrading the US air traffic control system from Verizon. Musk has publicly questioned Verizon's system reliability, claiming Starlink offers a free alternative. While this move could spark accusations of favoritism and conflicts of interest, a Starlink team is deeply involved in FAA modernization efforts, with SpaceX employees even holding FAA email addresses. This highlights Musk's companies' close ties with the government and their reliance on substantial government contracts.

Read more
Tech

Nuanced: Giving LLMs Precise Call Graph Context for AI Coding Assistants

2025-03-12
Nuanced: Giving LLMs Precise Call Graph Context for AI Coding Assistants

Nuanced is an open-source Python library that provides LLMs with precise call graph context by analyzing function relationships and generating a structured representation of code dependencies. It addresses the limitations of current AI coding assistants, which lack understanding of code structure and rely on limited context windows and embeddings. Nuanced leverages static analysis to build a traversable graph of function relationships, offering `init` (to generate the call graph) and `enrich` (to query specific functions) commands. This allows AI tools to access the same structured program understanding developers rely on, improving code comprehension and efficiency. Future development will include function purity analysis, code complexity metrics, and more.

Read more
Development code analysis

Recursive Descent Parsers: Simple Wins Over Complexity?

2025-07-28

The author explores approaches to parsing computer languages, specifically comparing recursive descent parsers to LR parser generators. While LR parser generators handle more complex grammars, the author favors recursive descent parsers due to their ease of use, lack of reliance on external tools, and ability to be written directly in the target language, thus minimizing learning curve and debugging challenges. For developers who occasionally need to build parsers for small languages, the simplicity and ease of use of recursive descent parsers outweigh their limitations in handling complex grammars.

Read more

arXivLabs: Experimenting with Community Collaboration

2025-05-23
arXivLabs: Experimenting with Community Collaboration

arXivLabs is a framework enabling collaborators to develop and share new arXiv features directly on the website. Individuals and organizations involved embrace arXiv's values of openness, community, excellence, and user data privacy. arXiv only partners with those sharing these commitments. Have an idea to enhance the arXiv community? Explore arXivLabs.

Read more
Development

CocoIndex: Building Knowledge Graphs with LLMs

2025-05-13
CocoIndex: Building Knowledge Graphs with LLMs

CocoIndex simplifies building and maintaining knowledge graphs with continuous source updates. This blog post demonstrates using Large Language Models (LLMs) to extract relationships from documents, building a knowledge graph with entity relationships and entity mentions, and exporting it to Neo4j. The process covers adding data sources, collecting data, extracting relationships, building the knowledge graph, and exporting and querying in Neo4j.

Read more
Development

Global Earthquake Early Warning System Leveraging Android Smartphones

2025-07-20
Global Earthquake Early Warning System Leveraging Android Smartphones

A new study demonstrates the effectiveness of a global earthquake early warning system built using the accelerometers in millions of Android smartphones worldwide. The system, called Android Earthquake Alerts (AEA), rivals traditional seismic networks in accuracy, detecting earthquakes globally and delivering timely alerts to users. Even in regions lacking traditional infrastructure, AEA provides crucial early warning to millions, potentially mitigating earthquake damage. By exploiting the speed difference between seismic waves, AEA issues alerts before the destructive waves arrive, buying precious seconds for people to react.

Read more

Evertop: An Ultra Low-Power, Ultra Long-Battery Life Solar PC

2025-04-21
Evertop: An Ultra Low-Power, Ultra Long-Battery Life Solar PC

Evertop is a portable PC emulating an IBM XT with an 80186 processor and 1MB RAM, running DOS, Minix, and Windows 3.0. Its low-power microcontroller, e-ink display, dual 10,000mAh batteries, and power-saving features enable hundreds to thousands of hours of use on a single charge. A built-in solar panel ensures indefinite off-grid operation. It boasts a full array of peripherals including a keyboard, PS/2 ports, various graphics and audio support, serial ports, Ethernet, Wi-Fi, and LoRa radio. Charging is versatile, with options for solar, DC input, and micro-USB, allowing simultaneous charging from multiple sources. A minimized version, Evertop Min, is also available.

Read more
Hardware

Generating Prompts via Activation Maximization: 95.9% Accuracy on Yelp Review Polarity

2025-08-16

This article presents a novel approach to prompt engineering using activation maximization. By optimizing the input rather than the model weights, a 4-token prompt was generated that achieved 95.9% accuracy on the Yelp Review Polarity sentiment classification task using Llama-3.2-1B-Instruct, significantly outperforming hand-written prompts (57%). This method cleverly leverages the LLM's embedding vector space, representing the prompt as a differentiable tensor and using gradient descent for optimization. This technique shows potential for increasing task switching efficiency in large language models, especially under GPU memory constraints.

Read more

Floating-Point Comparisons: Pitfalls and Practical Solutions

2025-05-15
Floating-Point Comparisons: Pitfalls and Practical Solutions

This article delves into the complexities of comparing floating-point numbers. The author highlights the unreliability of simple equality checks due to inherent precision limitations and accumulated rounding errors. Two comparison methods are detailed: relative error (epsilon) and ULP (Units in the Last Place), along with their strengths and weaknesses. The article emphasizes the failure of relative error comparisons near zero, proposing a solution combining absolute error. A compelling example using `sin(π)` demonstrates catastrophic cancellation and how floating-point representation errors can improve π's accuracy.

Read more
Development precision

Avoid the 'Nightmare Bicycle': Systemic Thinking in Product Design

2025-03-05
Avoid the 'Nightmare Bicycle': Systemic Thinking in Product Design

This article criticizes the tendency in product design to oversimplify user experience. Using the 'nightmare bicycle' (lacking numbered gears, only having scenario-specific buttons) as an example, it argues that such designs obscure the underlying system's structure, ultimately hindering user efficiency. Good design reveals the system's structure, enabling users to understand and apply it; poor design replaces systematic understanding with superficial labels, ultimately limiting user learning and application. The author advocates against oversimplification, trusting users' learning ability – just like a microwave only needs time and power buttons, users can figure out how to cook.

Read more

Meta's Data Center Secrets: Scaling to the Extreme

2025-02-11

This collection of papers unveils Meta's cutting-edge research in building and operating hyperscale data centers. From BGP routing to distributed storage (TAO), real-time data processing, cluster management (Twine), global capacity management (Flux), and power management (Dynamo), the papers detail the technologies enabling Meta to handle massive data, global user traffic, and high concurrency. Innovations like MAST for global ML training and RAS for continuous resource optimization highlight Meta's approach to building highly reliable, performant, and efficient data centers. These findings offer invaluable insights for anyone tackling the challenges of hyperscale infrastructure.

Read more
Tech hyperscale

Asahi Linux Founder Steps Down, Project Embraces Sustainable Future

2025-02-14
Asahi Linux Founder Steps Down, Project Embraces Sustainable Future

Following the resignation of founder Hector Martin, the Asahi Linux team announced a new governance structure and funding model to ensure the project's long-term sustainability. Seven developers will share decision-making power, and donations will be facilitated through Open Source Collective, replacing the previous Patreon model. The project will prioritize kernel upstreaming and continuous integration testing to improve stability and maintainability. While support for M3 and M4 chips is temporarily on hold, M1 and M2 users can look forward to features like DP alt mode, sparse image support in the Vulkan driver, and built-in microphone support.

Read more
Development

Hacking the Exception Handler: A Tale of DirectX, Flash, and Code Modification

2025-05-21

A team encountered a frustrating issue while using `SetUnhandledExceptionFilter` for unhandled exception handling: Direct3D and Flash were installing their own exception filters, rendering the team's solution ineffective. To resolve this, they employed a bold strategy: directly disabling the `SetUnhandledExceptionFilter` function via code modification to ensure their exception handler took precedence. The article details the process of code modification, including obtaining the function address, verifying the initial instructions, and utilizing `VirtualProtect` and `FlushInstructionCache` APIs for code replacement. Their solution restored proper exception reporting functionality.

Read more
Development code modification

The Jevons Paradox of Labor: How AI Is Making Us Work More

2025-03-28
The Jevons Paradox of Labor: How AI Is Making Us Work More

The essay explores the unexpected consequence of AI-driven productivity increases: instead of freeing us, it's leading to a 'labor rebound effect,' where increased efficiency paradoxically leads to more work. This is driven by factors like the soaring opportunity cost of leisure, the creation of new work categories, and intensified competition. The author argues that we need to redefine our metrics of progress, shifting from a singular focus on efficiency to a broader consideration of human well-being, to avoid a 'Malthusian trap.' Examples of alternative metrics include employee time sovereignty, well-being indices, and impact depth. Ultimately, the article suggests that in an AI-powered world, the truly scarce resource is knowing what's worth doing—a deeply personal and subjective question.

Read more
AI

UK Govt vs. Apple: A Decade's Encryption Showdown

2025-07-22
UK Govt vs. Apple: A Decade's Encryption Showdown

The UK government's demand for Apple to provide backdoor access to encrypted data has ignited the tech industry's most high-profile encryption battle in nearly a decade. Apple's refusal led to the withdrawal of its most secure cloud storage service from the UK and a legal challenge at the Investigatory Powers Tribunal. The dispute has drawn in the US government, with concerns raised by the Vice President, former President Trump, and the Director of National Intelligence about free speech, privacy, and potential breaches of data agreements. The ongoing legal battle casts a shadow over future UK AI legislation.

Read more
Tech

OpenSSL 3.5.0 Released: Ditching the Three-Way Handshake for 0-RTT

2025-04-09
OpenSSL 3.5.0 Released: Ditching the Three-Way Handshake for 0-RTT

OpenSSL 3.5.0 has been released, featuring support for various post-quantum cryptography methods and 0-RTT connections. The traditional three-way TCP handshake is considered too slow in today's always-on world. 0-RTT (Zero Round Trip Time), integrated into TLS 1.3, lets clients reconnect instantly without the handshake. A full handshake occurs on the initial connection, generating a session ticket used for subsequent connections, allowing the client to send data immediately without waiting for a server response. While security risks like replay attacks exist, 0-RTT's compatibility with the UDP-based QUIC protocol positions it as a key trend in future network connections.

Read more
Development

Floppotron 3.0: A 512-Floppy-Drive Orchestra Upgrade

2025-02-11

The Floppotron 3.0 is here! This massive hardware orchestra, featuring 512 floppy disk drives, 4 scanners, and 16 hard disk drives, has undergone a major upgrade. This enhanced version boasts increased scale and capabilities, incorporating custom electronic circuits and a completely rewritten firmware. The article details its operational principles, construction, and sound generation, explaining the intricacies of the floppy disk drive wall, scanners, and hard drives, and how MIDI controls them to produce music. It also addresses the significant power consumption and the power supply solution, along with plans for future instrument additions.

Read more
Hardware Creation

Ryanair Faces GDPR Complaint Over Mandatory Face Scans

2024-12-19
Ryanair Faces GDPR Complaint Over Mandatory Face Scans

The privacy advocacy group noyb filed a GDPR complaint against Ryanair for forcing users to create accounts and undergo invasive biometric verification, including face scans, during the booking process. This practice, allegedly aimed at preventing online travel agencies from bulk purchasing tickets, violates GDPR principles of data minimization, purpose limitation, and consent. Ryanair is accused of prioritizing competitive advantage over user privacy.

Read more
Misc biometrics

Corebooting My Thinkpad T420: A Tale of Woe

2025-04-09

This post details the author's arduous journey of installing Coreboot on a Thinkpad T420. From the painstaking disassembly of the robust laptop, to wrestling with finicky IC clip connections and battling UEFI compatibility issues during Coreboot compilation, the process proved exceptionally challenging. While the author ultimately succeeded in flashing Coreboot and achieved faster boot times, several functionalities, including Windows XP booting, the hardware clock, and other payloads, are now broken. The author expresses uncertainty about continued Coreboot usage unless a CPU upgrade is pursued.

Read more
Hardware

Beyond RAG: LLM Tool Calling Ushers in a New Era for Semantic Search

2025-05-22
Beyond RAG: LLM Tool Calling Ushers in a New Era for Semantic Search

This article explores methods for implementing semantic search, particularly using LLMs for vector embedding search. While directly embedding user search terms and documents sometimes yields suboptimal results, new techniques like Nomic Embed Text v2 improve embedding methods, bringing questions and answers closer together in vector space. Furthermore, LLMs can synthesize potential answers, then use those embeddings to search for relevant documents. The article also introduces LLM-based Retrieval-Augmented Generation (RAG) systems, emphasizing that RAG doesn't rely on vector embeddings and can be combined with keyword search or hybrid search systems. The author argues that despite the emergence of long-context models, RAG won't disappear because the amount of data will always exceed model context capacity. The author favors the LLM tool-calling approach, exemplified by o3 and o4-mini, believing it's more effective than traditional RAG (single retrieval followed by direct answering).

Read more
AI

Russia Engulfed in Hundreds of Arson Attacks: A Calculated Campaign of Psychological Warfare?

2025-01-14
Russia Engulfed in Hundreds of Arson Attacks: A Calculated Campaign of Psychological Warfare?

Hundreds of arson attacks have swept across Russia, targeting banks, post offices, and police cars. The Kremlin attributes this to a campaign of psychological warfare aimed at destabilizing the country. However, some cases present puzzling details. An elderly man arrested for setting fire to an ATM claimed he was acting under the instructions of unidentified telephone scammers. These incidents raise questions about the perpetrators and their motives, highlighting potential vulnerabilities within Russian society.

Read more

Tesla's RoboTaxi Trademark Rejected

2025-05-09
Tesla's RoboTaxi Trademark Rejected

Tesla CEO Elon Musk's vision of an AI and robotics future, including a sub-$30,000 driverless two-seater dubbed "robotaxi," has hit a snag. The USPTO rejected Tesla's trademark application, citing the term's generic and descriptive nature, lacking originality. This setback complicates Tesla's marketing plans, requiring them to demonstrate their product's unique features to avoid renaming or further legal challenges. This isn't Tesla's first trademark dispute; they previously faced a lawsuit over vehicle design similarities with Blade Runner 2049.

Read more
Tech

Minimum Bipartite Matching via Riemann Optimization

2025-01-08

This paper presents a novel approach to solving the minimum bipartite matching problem using Riemann optimization. The author transforms the combinatorial optimization problem into an unconstrained optimization problem on the manifold of doubly stochastic matrices and solves it using Riemann gradient descent. Experimental results show that this method can effectively find the optimal solution and has good convergence. This research cleverly combines combinatorial mathematics, differential geometry, and computer science, providing a new perspective for solving such problems.

Read more

PlanetScale Metal: Ditching the Cloud Database IO Bottleneck

2025-03-13
PlanetScale Metal: Ditching the Cloud Database IO Bottleneck

This article explores the history of computer storage technologies, from tape to hard disk drives to solid-state drives (SSDs), and the IO performance challenges brought about by cloud computing. Traditional cloud database services typically use network-attached storage (NAS), resulting in high latency and IOPS limitations. PlanetScale's Metal product uses local NVMe drives, directly connecting compute and storage, to achieve extremely low latency, unlimited IOPS, and high data durability, completely solving the IO bottleneck problem of cloud databases.

Read more

Apertus: A Fully Open, Multilingual LLM

2025-09-06
Apertus: A Fully Open, Multilingual LLM

Apertus is a fully open, multilingual large language model with 70B and 8B parameters, supporting over 1000 languages and long context. Trained on 15T tokens of fully compliant, open data, it achieves performance comparable to closed-source models. Apertus uses a novel xIELU activation function and the AdEMAMix optimizer, undergoing supervised fine-tuning and QRPO alignment. Its weights, data, and training details are publicly available, respecting data owner opt-out consent and avoiding memorization of training data. Integrated into the transformers library, Apertus supports various deployment methods. While powerful, users should be aware of potential inaccuracies and biases in its output.

Read more
AI

SeedBox Lite: Stream Torrents Instantly

2025-08-29
SeedBox Lite: Stream Torrents Instantly

SeedBox Lite is a revolutionary torrent streaming platform that lets you instantly watch movies and TV shows without waiting for full downloads. Built with modern web technologies, it offers a Netflix-like experience with powerful torrent capabilities. SeedBox Lite supports multiple formats, features smart caching, subtitle support, and responsive design, working seamlessly across Windows, macOS, and Linux. Deployment is a breeze, taking minutes with either Docker or PM2.

Read more
Development

Guix's G-Expressions: Embedding Lower-Level Code in Higher-Level Code

2025-08-03

Guix uses Scheme for both high-level actions (like defining packages) and low-level actions (like building derivations). To embed lower-level code within higher-level code, it employs G-expressions. For example, in the `start` field of `wesnoth-shepherd-service`, `#~(...)` passes lower-level code, while `#$(...)` escapes higher-level code, which the compiler lowers to lower-level code. The `make-forkexec-constructor` function creates and executes child processes, offering features like setting user, group, umask, and environment variables.

Read more
Development G-expressions

Critical Vulnerabilities Found in Copeland Controllers Threaten Global Supply Chains

2025-09-03
Critical Vulnerabilities Found in Copeland Controllers Threaten Global Supply Chains

Ten critical vulnerabilities (Frostbyte10) have been discovered in Copeland controllers, widely used by major supermarket chains and cold storage facilities worldwide. These flaws could allow attackers to remotely manipulate temperatures, potentially spoiling food and medicine and causing significant supply chain disruptions. The vulnerabilities affect E2 and E3 controllers, impacting critical systems like compressors and condensers. Copeland has released firmware updates, and CISA has issued advisories urging immediate patching. Exploitation of these vulnerabilities could lead to unauthorized remote code execution.

Read more
Tech
1 2 221 222 223 225 227 228 229 596 597