Popular：
Virtualization DNS security formal verification reachability analysis compiler errors macro conflict web extension development framework Bitmap Graphics API inconsistencies All Tags

Soviet Hero: The Extraordinary Rescue of Shavarsh Karapetyan

2025-01-14
Soviet Hero: The Extraordinary Rescue of Shavarsh Karapetyan

Shavarsh Karapetyan, a former Soviet finswimmer, is renowned for his incredible bravery in saving the lives of 20 people during a 1976 trolleybus accident in Yerevan. In freezing, murky water, he repeatedly dived into the submerged vehicle, pulling people to safety. Despite suffering severe injuries and contracting pneumonia, he still competed and set a world record. Karapetyan's heroic act is a testament to human courage and selflessness, a truly inspiring legend.

Read more
(en.wikipedia.org)
Misc heroism rescue

Einsum: Beyond Matrix Multiplication

2025-01-06

Einsum is more than just matrix multiplication; it's an efficient implementation of Einstein summation convention. It uses concise notation to represent complex tensor operations, avoiding nested loops and improving code readability and performance. This article delves into the mechanics of Einsum, demonstrating its advantages in handling high-dimensional tensor operations such as matrix multiplication, transposition, and trace calculations with illustrative examples. For developers needing high-performance tensor computations, Einsum is an invaluable tool.

Read more
(einsum.joelburget.com)
Development Tensor operations High-performance computing

Does TikTok Cause 'Brain Rot'? A Study's Controversial Findings and Limitations

2025-03-02
Does TikTok Cause 'Brain Rot'? A Study's Controversial Findings and Limitations

A recent study from Tianjin Normal University in China claims that excessive short-video consumption (like TikTok) alters brain structure, leading to so-called 'brain rot'. The research found increased gray matter in specific brain regions and enhanced brain activity synchronization among heavy users. However, the study has significant limitations: its cross-sectional design prevents establishing causality; its whole-brain search for differences increases the risk of false positives; and its interpretation of brain activity synchronization is controversial. Experts point out that 'short-video addiction' isn't a formal clinical diagnosis, and the findings can't be simply interpreted as TikTok causing brain damage. While excessive consumption of frivolous videos can be problematic, focusing on healthier media habits is more productive than worrying about brain changes.

Read more
(www.sciencefocus.com)
Tech Addiction

The Messy State of TOTP: A Test Suite is Born

2025-03-02
The Messy State of TOTP: A Test Suite is Born

The current TOTP specification is riddled with inconsistencies. Major implementations by Google, Apple, and Yubico subtly disagree on its implementation, leading to idiosyncratic variants in various MFA apps. The official RFC is frustratingly vague. The author built a test suite to check if your favorite app correctly implements the TOTP standard, highlighting ambiguities in digit count, hash algorithm, time step, secret length, and labeling. The author calls for improved specifications to prevent future issues.

Read more
(shkspr.mobi)
Development

OpenAI's Mathematical Proof: Why ChatGPT's Hallucinations Are Here to Stay (Maybe)

2025-09-13
OpenAI's Mathematical Proof: Why ChatGPT's Hallucinations Are Here to Stay (Maybe)

OpenAI's latest research paper mathematically proves why large language models like ChatGPT "hallucinate" – confidently fabricating facts. This isn't simply a training issue; it's mathematically inevitable due to the probabilistic nature of word prediction. Even perfect data wouldn't eliminate the problem. The paper also reveals a flawed evaluation system that penalizes uncertainty, incentivizing models to guess rather than admit ignorance. While OpenAI proposes a confidence-based solution, it would drastically impact user experience and computational costs, making it impractical for consumer applications. Until business incentives shift, hallucinations in LLMs are likely to persist.

Read more
(theconversation.com)
AI Hallucination

AI Revolutionizes Observability: The Death of Dashboards, The Rise of LLM-Driven Analysis

2025-06-11
AI Revolutionizes Observability: The Death of Dashboards, The Rise of LLM-Driven Analysis

Traditional observability tools rely on dashboards and manual analysis of massive datasets, but the advent of Large Language Models (LLMs) is changing that. The author recounts a case study where a simple LLM prompt automatically identified and diagnosed latency spikes in an application service, far surpassing human efficiency. This heralds a shift in observability tools from graphical interfaces to AI-driven real-time analysis, with fast feedback loops becoming the core competency. In the future, AI agents will assist or even replace parts of development and operations work, and rapid analysis capabilities will be crucial.

Read more
(www.honeycomb.io)
Tech

800-Year-Old Kid's Doodles: A Glimpse into Medieval Childhood

2025-04-16
800-Year-Old Kid's Doodles: A Glimpse into Medieval Childhood

Soviet archaeological excavations unearthed birch bark sketches from medieval Novgorod, circa 1250 CE, created by a schoolboy named Onfim. His whimsical drawings—horses, soldiers, self-portraits—reveal the expressive capabilities of medieval children. Contrasting this are charcoal drawings found in a French iron mine, depicting child miners, a poignant reflection of their harsh reality. These discoveries offer a unique perspective on premodern childhood, highlighting its universality and diverse experiences across time and culture.

Read more
(resobscura.substack.com)
Misc Medieval children Art history Archaeological discovery

Google Unveils Ironwood: A 7th-Gen TPU for the Inference Age

2025-04-09
Google Unveils Ironwood: A 7th-Gen TPU for the Inference Age

At Google Cloud Next '25, Google announced Ironwood, its seventh-generation Tensor Processing Unit (TPU). This is Google's most powerful and scalable custom AI accelerator yet, designed specifically for inference. Ironwood marks a shift towards a proactive “age of inference,” where AI models generate insights and answers, not just data. Scaling up to 9,216 liquid-cooled chips interconnected via breakthrough ICI networking (nearly 10MW), Ironwood is a key component of Google Cloud's AI Hypercomputer architecture. Developers can leverage Google's Pathways software stack to easily harness the power of tens of thousands of Ironwood TPUs.

Read more
(blog.google)
AI Inference AI

Mysterious MAC Addresses: A Hidden Signal in Bluetooth Devices

2025-04-24
Mysterious MAC Addresses: A Hidden Signal in Bluetooth Devices

This article unveils a shocking discovery: Analysis of a large number of Bluetooth device MAC addresses reveals anomalously low entropy and structured patterns, completely unlike randomly generated MAC addresses. These structured patterns include fixed bits, a rotating page counter, and a precise 2000ms broadcast interval. Even more perplexing, these patterns align with the frequency of a microfluidic pump, pulsating at a 2000ms cycle, found in blood samples. This suggests a hidden, synthetic emission architecture may be covertly communicating through consumer Bluetooth devices, the purpose and origin of which remain unknown.

Read more
(substack.com)
Tech MAC address

Resonate: A Low-Latency, Low-Memory, Low-Cost Spectral Analysis Algorithm

2025-04-15

Resonate is a low-latency, low-memory footprint, and low-computational-cost algorithm for evaluating perceptually relevant spectral information from audio (and other) signals. It builds on a resonator model using Exponentially Weighted Moving Average (EWMA) to accumulate signal contributions around resonant frequencies. Its compact iterative formulation allows for efficient updates with minimal arithmetic operations per sample, requiring no buffering. Resonate computes real-time perceptually relevant spectral content estimates; memory and per-sample computational complexity scale linearly with the number of resonators, independent of input sample count. Open-source implementations in Python, C++, and Swift are available, along with demonstration apps.

Read more
(alexandrefrancois.org)
Development

Donkey Kong's Broken Ladder Glitch: Luck and Skill Combine for a New Kill Screen

2025-02-08
Donkey Kong's Broken Ladder Glitch: Luck and Skill Combine for a New Kill Screen

The 'broken ladder' glitch in the classic arcade game Donkey Kong, long thought impossible to exploit, has been conquered. Player Kosmic, using an emulator and a hefty dose of luck, utilized the glitch to not only complete the game but discover a new, true kill screen at level 22-6. The glitch exploits a random delay in Donkey Kong's barrel throwing, giving Mario precious extra frames. This achievement highlights the game's intricate mechanics and underscores the crucial role of both skill and chance in overcoming seemingly insurmountable challenges.

Read more
(arstechnica.com)
Game game glitch luck

Oldest Whale Bone Tools Found, Dating Back 20,000 Years

2025-05-31
Oldest Whale Bone Tools Found, Dating Back 20,000 Years

Scientists have unearthed the oldest known evidence of humans using whale bones to make tools, dating back approximately 20,000 years. Discovered in the Bay of Biscay near Spain and France, these narrow projectiles were crafted from the bones of blue whales, fin whales, sperm whales, and other species. Researchers believe ancient humans likely scavenged beached whales, repurposing their bones for hunting reindeer or bison, rather than actively hunting whales themselves. This discovery, published in Nature Communications, highlights the importance of coastal resources for early human survival and pushes back the timeline of whale bone tool use.

Read more
(apnews.com)
Tech paleolithic whale bone tools

The Invisible Tsunami Kids: A Forgotten Tragedy

2025-01-02
The Invisible Tsunami Kids: A Forgotten Tragedy

The 2004 Sumatra-Andaman tsunami claimed nearly 230,000 lives, many of them children. This article details the plight of the surviving children: orphaned, suffering from PTSD, and at risk of trafficking. While international organizations worked to help, the future of many remains uncertain. The author calls for attention to the plight of these children and encourages readers to help through volunteering or donations.

Read more
(robertvanwey.substack.com)
Misc tsunami disaster relief

From Random Streaks to Recognizable Digits: Building an Autoregressive Image Generation Model

2025-06-08
From Random Streaks to Recognizable Digits: Building an Autoregressive Image Generation Model

This article details building a basic autoregressive image generation model using a Multilayer Perceptron (MLP) to generate images of handwritten digits. The author explains the core concept of predicting the next pixel based on its predecessors. Three models are progressively built: Model V1 uses one-hot encoding and ignores spatial information; Model V2 introduces positional encodings, improving image structure; Model V3 uses learned token embeddings and positional encodings, achieving conditional generation, generating images based on a given digit class. While the generated images fall short of state-of-the-art models, the tutorial clearly demonstrates core autoregressive concepts and the building process, providing valuable insights into generative AI.

Read more
(tunahansalih.github.io)
AI

PLOS ONE Retractions: 45 Editors Linked to Over 30% of Retracted Papers

2025-08-06
PLOS ONE Retractions: 45 Editors Linked to Over 30% of Retracted Papers

A study in PNAS reveals a shocking pattern of misconduct at PLOS ONE. 45 editors, responsible for only 1.3% of published articles, were linked to over 30% of the journal's 702 retractions by early 2024. Twenty-five of these editors even authored retracted papers themselves. The research suggests a coordinated network potentially involving paper mills, highlighting systemic flaws in peer review. Specific editors, like Shahid Farooq (52 out of 79 edited papers retracted), demonstrate exceptionally high retraction rates. PLOS acknowledges the issue and states it has taken action, but the incident underscores the vulnerabilities of open-access journals to manipulation.

Read more
(www.nature.com)
Tech retractions journal editors

Swiss VPNs Under Fire: Privacy vs. Security in the Balance

2025-06-06
Swiss VPNs Under Fire: Privacy vs. Security in the Balance

Proposed changes to Swiss encryption laws are sparking controversy, with increased surveillance obligations requiring companies to collect user data, significantly impacting online privacy. Swiss-based VPNs like Proton VPN and NymVPN are directly affected, with Proton's CEO even threatening to relocate rather than compromise user privacy. Surprisingly, Infomaniak, a Swiss cloud security company, supports the law, arguing that anonymity hinders justice and a balance must be struck. The debate centers around the difference between privacy and anonymity, and the risks of metadata collection. Infomaniak believes metadata (geolocation, timestamps, IP addresses, etc.) aids in crime-fighting, while opponents fear privacy violations and potential misuse. This clash over balancing privacy, security, and anonymity will have significant implications for the global VPN industry.

Read more
(www.tomsguide.com)
Tech Swiss Law

Ladybird Browser Project Monthly Update: Million-Level WPT, Embracing OpenSSL

2025-03-02
Ladybird Browser Project Monthly Update: Million-Level WPT, Embracing OpenSSL

The Ladybird open-source browser project made significant progress this month, merging 281 PRs from 35 contributors. The number of passing subtests in Web Platform Tests (WPT) exceeded 1.77 million, moving closer to the 90% pass rate target for iOS alternative browser engines. The project adopted OpenSSL to replace its homegrown cryptography library and migrated the networking stack to curl. It also added support for Firefox DevTools, improving debugging efficiency. Furthermore, Ladybird added features such as CSS image cursors, new CSS pseudo-classes, text decoration error highlighting, and implemented TextEncoderStream and the Resource Timing API. Style invalidation mechanisms were optimized, and aarch64 Linux continuous integration was added.

Read more
(ladybird.org)
Development

Deep Dive into BSC Cryptocurrency Scams

2024-12-24

During the 2021 cryptocurrency bull market, the Binance Smart Chain (BSC) was flooded with various scams. The author analyzes multiple cases, exposing common tactics employed by scammers, including: creating smart contracts that make tokens unsaleable; manipulating approval functions to fail transactions or approve for minuscule amounts; setting adjustable transaction fees, eventually to 100%; falsely claiming ownership renunciation or liquidity locking; and using deceptive marketing. These scams preyed on the lack of knowledge among many new investors, successfully defrauding significant funds. The article concludes with a warning against seeking financial advice from social media platforms like Reddit.

Read more
(nyuu.page)
Tech BSC scams smart contract vulnerabilities

Tracking New Books with Perplexity AI: An LLM Hack

2025-04-20
Tracking New Books with Perplexity AI: An LLM Hack

The author experimented with Perplexity AI's API to track new books by their favorite authors. While Perplexity AI, being based on web searches, produces inconsistent results and hallucinations, through clever prompt engineering and coding, the author built a system to list new books relatively efficiently. Despite repetition and inconsistencies, this is a fun example of using an LLM to solve a real-world problem, showcasing both the potential and limitations of LLMs.

Read more
(lars.ingebrigtsen.no)
Development LLM application API calls

Conquering the Flash of Incomplete Markdown (FOIM) with a Clever State Machine

2025-06-04
Conquering the Flash of Incomplete Markdown (FOIM) with a Clever State Machine

Streak encountered the 'Flash of Incomplete Markdown' (FOIM) problem while using OpenAI's streaming API to generate Markdown content with citations. Incomplete links and even AI hallucinations leading to incorrect URLs plagued their product. To solve this, they implemented a state machine on the server to buffer Markdown links until complete before sending them to the client. This not only eliminated FOIM but also reduced OpenAI token usage, sped up response times, and improved privacy—a win-win-win.

Read more
(engineering.streak.com)
Development

C++ Metaprogramming Tricks: Optimizing Variant Access Performance

2025-05-14

This article explores optimizing `std::variant` access performance in C++ using metaprogramming techniques, aiming for efficiency comparable to hand-written `switch` statements. Several approaches are compared, including jump tables, dispatch tables, macros, recursive `switch`, and short-circuiting folds, analyzing their pros, cons, and compiler optimization strategies. Ultimately, a solution combining short-circuiting folds and the anticipated C++26 expansion statements is presented, achieving efficient generic access and avoiding performance bottlenecks.

Read more
(pydong.org)
Development

phptop: Lightweight PHP Performance Monitoring Tool

2025-06-05
phptop: Lightweight PHP Performance Monitoring Tool

phptop is a lightweight PHP performance monitoring tool that tracks per-query execution time (wallclock, user, and system CPU time) along with memory and other resource usage. It's easily activated globally on a LAMP server with a single line configuration change in your php.ini. It's low-resource and has been used by Bearstech in production for years without issue. Requires PHP >= 5.2.0, tested up to PHP 8.2.

Read more
(github.com)
Development

Real-time Lighting and Normal Mapping on the N64: A Stunning Reverse Engineering Feat

2025-05-17

This article details an impressive technique for achieving real-time lighting and normal mapping on the Nintendo 64. By cleverly leveraging palette textures and CPU-side shading, the author circumvents the N64's hardware limitations to achieve surprisingly impressive visuals. The technique involves compressing diffuse and normal information into a shared palette, and updating the palette at runtime via the CPU to simulate lighting effects. While the method has some limitations, such as lack of point light support and shadows, its innovative nature is remarkable, opening new possibilities for graphics rendering on the N64 platform.

Read more
(30fps.net)
Development

arXivLabs: Experimental Projects with Community Collaboration

2025-04-27
arXivLabs: Experimental Projects with Community Collaboration

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

Read more
(arxiv.org)
Development

Google AI's Nonsense: Seriously Wrong Answers

2025-04-24
Google AI's Nonsense: Seriously Wrong Answers

Google's AI Overview feature provides definitions and origins for any made-up phrase, even nonsensical ones. It uses a probabilistic model, predicting the next most likely word based on its training data, generating seemingly plausible explanations. However, this approach ignores semantic correctness and may cater to user expectations, leading to seemingly reasonable explanations for meaningless phrases. This highlights the limitations of generative AI in handling uncommon knowledge and minority perspectives, and its tendency to 'please' the user.

Read more
(www.wired.com)
AI

Octocode: AI-Powered Code Indexer and Knowledge Graph Builder

2025-06-07
Octocode: AI-Powered Code Indexer and Knowledge Graph Builder

Octocode is a powerful code indexer and semantic search engine that builds intelligent knowledge graphs of your codebase. It combines advanced AI capabilities with a local-first design, providing deep code understanding, relationship mapping, and intelligent assistance for developers. Supporting numerous programming languages, Octocode offers natural language queries, multi-modal search, intelligent ranking, and symbol expansion. A built-in memory system stores insights, decisions, and context, seamlessly integrating with AI assistants.

Read more
(github.com)
Development

Intel's Xeon Architect Jumps Ship Amidst Executive Exodus

2025-09-13
Intel's Xeon Architect Jumps Ship Amidst Executive Exodus

Ronak Singhal, the chief architect behind Intel's Xeon line of server CPUs, is leaving the company after nearly 30 years. Singhal's contributions are significant, including core development roles in the Haswell and Broadwell architectures, and contributions to the Core and Atom processor families. While the Xeon division has faced stiff competition from AMD and Arm-based cloud CPUs in recent years, Singhal arguably leaves it in its most competitive position in years. However, his departure is just the latest in a string of high-profile exits from Intel's datacenter group, including several other executives and even the CEO, highlighting significant talent drain and intense industry competition.

Read more
(www.theregister.com)
Tech Executive Exodus

8 Years of Sleep-Induced Reboots: Decoding a Dell Inspiron Firmware Bug

2025-09-22

For eight years, the author's Dell Inspiron 5567 randomly rebooted when put to sleep, across multiple operating systems. A deep dive into the firmware's source code revealed the culprit: the Southbridge's SPTS method. This method prematurely sent the sleep command before properly setting the sleep state, causing the reboots. The solution involved reordering code within SPTS to ensure the sleep state was correctly set before triggering the sleep. The article details the debugging process and explores the intricacies of ACPI sleep states.

Read more
(triangulatedexistence.mataroa.blog)
Hardware

MinIO's Controversial Community Edition Changes Spark Outrage

2025-05-30
MinIO's Controversial Community Edition Changes Spark Outrage

MinIO, a popular open-source object storage solution, has removed key web-based management features from its community edition, prompting backlash from users. The free version now requires users to rely on command-line tools or upgrade to a paid plan. This decision, likened by some to 'enshittification,' has led many to explore alternatives like SeaweedFS, Garage, and Zenko. MinIO maintains that the changes are necessary to sustain the project's development.

Read more
(biggo.com)
Development minio
1 2 … 50 51 52 54 56 57 58 … 596 597