Article archive
Browse individual source reports, including coverage still awaiting the summary or community analysis needed for our story feed.
5,369 articles · page 53 of 108
Browse topics
Claude Code Just Got 5x More Expensive
news
3 min read
Claude Code has implemented a hidden test that monitors the destination of user requests through the ANTHROPIC_BASE_URL environment variable. Despite no change in monthly pricing or machine specifications, users have reported a fivefold increase in operational costs.
I ported Kubernetes to the browser
tool
13 min read
A partial port of Kubernetes, named webernetes, has been developed in TypeScript to enable running clusters directly in the browser. The implementation includes features such as pod lifecycles, cluster DNS and networking, container garbage collection, and IP allocation, generating nearly 100,000 lines of code over 552 commits.
Claude Sonnet 5
tool
8 min read
Claude Sonnet 5 is designed as the most agentic Sonnet model, capable of making plans, using tools like browsers and terminals, and running autonomously. This model offers enhanced capabilities compared to previous versions, such as Sonnet 3.5, 3.6, and 3.7, which were the first to demonstrate significant skills in coding and tool utilization.
Claude Science
tool
5 min read
Claude Science beta serves as a research partner for scientific analysis, enabling users to perform data analyses, search databases, and document each step from data handling to publication. The platform supports viewing proteins, structures, and molecules with fully reproducible results linked to their underlying code and environments.
We Are the Last People Who Know How It Works
opinion
4 min read
In the 1990s, playing a computer game often required understanding the underlying computer system, including editing files like autoexec.bat. Users sometimes created boot disks specifically configured to launch a single game.
Nano Banana 2 Lite
tool
6 min read
Nano Banana 2 Lite offers dramatically reduced latency for image generation and editing. This model provides high-speed performance at a lower cost, allowing users to generate thousands of images efficiently while maintaining character consistency and accuracy.
Crypto firms have spent $189M so far on 2026 US election, report says
news
1 min read
Crypto firms have invested $189 million in the 2026 U.S. election cycle. This spending reflects the increasing influence of the cryptocurrency industry in political campaigns.
Anthropic has embedded hidden spyware-like code in Claude Code
news
1 min read
‼️ BREAKING: Anthropic has embedded hidden spyware-like code in Claude Code that covertly targets Chinese users. It then sends information regarding every user by injecting it into their prompt message. Claude Code is sending info like timezone, proxy and possible AI Lab connections into the system prompt in ways Chinese users can't notice. A coding agent with repo and command permissions should n...
Claude Code Is Steganographically Marking Requests
research
4 min read
Claude Code is implementing steganographic techniques to mark requests for privacy reasons. The inspection of the Claude Code binary revealed concerns regarding the permissions granted to coding agents, which often include extensive access to files, shells, and repositories.
Words Are a Byproduct of Consciousness. For LLMs, It's Backwards
opinion
5 min read
Human communication begins with an idea or thought, with words serving as a means to express that underlying consciousness. In contrast, large language models (LLMs) generate words first, with the underlying idea or consciousness being a secondary consideration.
Memoirs of Extraordinary Popular Delusions and the Madness of Crowds (1852)
research
2 min read
"Memoirs of Extraordinary Popular Delusions and the Madness of Crowds" by Charles Mackay is a seminal study on crowd psychology published in 1841. The work analyzes humanity's tendency toward collective manias, covering topics such as financial bubbles, religious crusades, witch trials, and alchemy.
Popping the GPU Bubble
research
15 min read
GPUs often remain idle during AI model inference due to delays in receiving instructions from the CPU, leading to a phenomenon known as the GPU bubble. Optimizing communication between the CPU and GPU can enhance the efficiency and speed of AI model execution.
LongCat-2.0, a large-scale MoE model with 1.6T total and 48B Active
tool
1 min read
LongCat-2.0 is a large model with 1.6 trillion parameters, fully trained using domestically produced chips. The model aims to improve performance and efficiency in AI applications.
South Korea to spend $1T on more memory chip production and humanoid robots
news
1 min read
South Korea's government and major tech firms will invest $1 trillion in projects to enhance global memory chip supply, develop AI data centers, and promote humanoid robot deployment by 2028. This initiative responds to increased demand for memory chips driven by the AI industry, which has resulted in shortages and rising prices.
Scientists find molecular-level evidence for two structures in liquid water
research
6 min read
A study published in Nature Physics reveals that liquid water consists of a constantly shifting mixture of two distinct microscopic structures. This finding provides molecular-level evidence supporting the long-held hypothesis that water is not a single uniform substance.
Micro-Agent: Beat Frontier Models with Collaboration Inside Model API
tool
9 min read
Micro-Agent enables collaboration within model APIs to enhance AI inference efficiency. Routers serve as the control plane, optimizing requests by directing them to appropriate models, thereby reducing costs associated with using frontier models versus open-source or local alternatives.
Ornith-1.0: self-improving open-source models for agentic coding
tool
10 min read
Ornith-1.0 is an open-source self-improving model for agentic coding, available in configurations of 9B-Dense, 31B-Dense, 35B-MoE, and 397B-MoE. It achieves state-of-the-art performance on coding benchmarks such as Terminal-Bench 2.1, SWE-Bench, NL2Repo, and OpenClaw by utilizing reinforcement learning for solution generation.
Qwen 3.6 27B is the sweet spot for local development
tool
6 min read
Qwen 3.6 27B is a dense local model recommended for its powerful performance in general intelligence tasks. It is slower than the mixture-of-experts variant Qwen 3.6 35B A3B but is considered more effective for local development.
Amazon Is Awash with AI-Written Guideslop for Games That Aren't Even Out
opinion
3 min read
Amazon is experiencing an influx of AI-generated guidebooks for unfinished games, reflecting a growing trend of low-quality content on the platform. The situation is exacerbated by Amazon's significant investment in OpenAI and the presence of its AI assistant, Rufus.
In San Francisco, even $180k tech salaries are no longer enough
news
2 min read
In San Francisco, tech workers earning $180,000 are struggling to compete with the rising salaries of the A.I. elite as companies like OpenAI and Anthropic prepare for public offerings. Many workers express concerns about their ability to afford living in the city amidst these salary disparities.
Working With AI: A concrete example
opinion
10 min read
AI has become a powerful tool for development, but it poses risks such as diminishing intellectual capacity and environmental concerns. The reliance on AI by developers can lead to issues similar to the Sorcerer’s Apprentice problem.
A field guide to the modern front end for developers who hand-wrote HTML
opinion
13 min read
Frontend development has evolved significantly over the past two decades, transitioning from simple file uploads to complex build tools. In 2008, creating a website involved saving an index.html file and using an FTP client without the need for a build process.
Mag 7 starting to underperform [pdf]
research
1 min read
Mag 7 is experiencing a decline in performance metrics. Data indicates a need for strategic adjustments to improve outcomes.
What can you confidently guarantee about your software?
tool
14 min read
The cost and tooling of formal verification have become widely accessible, allowing for increased use in software development. AI is reducing the barriers to formal verification by eliminating the costs associated with writing proofs, enabling the creation of software that guarantees mathematically correct business rules.
Tidal AI Policy
news
1 min read
Tidal has implemented an AI policy that outlines guidelines for the use of AI technologies within its platform. The policy aims to ensure responsible and ethical use of AI while enhancing user experience and creativity.
Anthropic CEO: Open-Source AI is getting dangerous (2023)
news
1 min read
Anthropic CEO Dario Amodei warned lawmakers that open-source AI is becoming increasingly dangerous. He stated that once powerful models are released openly, companies cannot effectively monitor misuse, revoke access, or update safety measures.
Herdr: Agent multiplexer that lives in your terminal
tool
6 min read
Herdr is an agent multiplexer that allows users to run multiple coding agents in a single terminal, providing a real terminal experience for each agent. Users can monitor the status of agents, manage workspaces and tabs, and utilize features like click-and-drag pane splitting while running agents on local machines or remote servers via SSH.
HackerRank open sourced its ATS. My resume scored 90/100. Oh wait 74. No – 88
tool
5 min read
HackerRank has open-sourced its Applicant Tracking System (ATS), available on GitHub. A user tested the ATS and received varying resume scores, ranging from 74 to 90 out of 100.
AI boom risks global financial crash, warn central bankers
news
1 min read
Central bankers warn that debt-fueled spending on AI could lead to a global financial crisis. The Bank for International Settlements (BIS) cites excessive investment in AI data centers and complex financial ties among AI companies and shadow banks as significant risks.
Better Images of AI
opinion
1 min read
Clichéd images of AI, such as humanoid robots and glowing brains, misrepresent the technology and its societal impacts. Using abstract or realistic visuals can enhance understanding of AI's significant environmental and social implications.
Knowledge Distillation of Black-Box Large Language Models (2024)
research
2 min read
Knowledge distillation (KD) enhances the performance of smaller models by transferring knowledge from proprietary large language models (LLMs) like GPT-4. This method aims to improve the capabilities of smaller models while utilizing the strengths of black-box teachers.
Xonaly – Canada's Independent Search Engine
tool
1 min read
Xonaly is an ad-free and private web monitoring tool developed in Canada. It allows users to add sites for monitoring and features a core monitor for efficient search capabilities.
TOP500 at ISC’26: We have a New Number 1 Supercomputer
news
5 min read
The LineShine Supercomputer in Shenzhen, China, has been ranked as the new number one supercomputer on the TOP500 list at ISC 2026. This marks the first Chinese submission to the TOP500 in nine years, featuring a powerful CPU-only system.
The US Used to Demand the Best Tech. Now We Ban It
opinion
9 min read
Jim Farley, CEO of Ford, expressed a desire to keep the Xiaomi SU7, a Chinese electric car he imported to the US. Marques Brownlee reviewed the Xiaomi SU7 Max, which features dual-motor All Wheel Drive and an 800V battery system.
Do LLMs pass the mirror test?
opinion
12 min read
The mirror test has been adapted for large language models (LLMs) by asking them to identify their own outputs among anonymized responses. Results show varying success rates among models, but the outcomes are considered uninformative regarding their self-awareness.
Librepods: AirPods liberated
tool
6 min read
librepods.org falsely claims to be the official website of the LibrePods project while using its logo and copyright. LibrePods enables the use of AirPods features that are typically exclusive to Apple devices by implementing the proprietary protocol used by Apple.
Historical memory prices 1960-2026
research
3 min read
Historic and current memory and storage prices are available in an interactive format, allowing users to download raw data. The dataset includes price per gigabyte over time for different memory types, including DRAM, NAND flash, and HBM, along with a breakdown of DRAM prices by generation.
The cost YAGNI was never about
opinion
4 min read
YAGNI (You Aren't Gonna Need It) is a principle that emphasizes not implementing features until they are necessary, countering arguments for premature complexity. The principle is often misunderstood as merely about saving effort, but its true focus is on avoiding unnecessary complexity in software development.
GLM 5.2 beats Claude in our benchmarks
research
9 min read
GLM 5.2 from Zhipu AI achieved a 39% F1 score on the IDOR detection benchmark, outperforming Claude Code, which scored 32%. Both models were evaluated on the same dataset and prompt, with GLM 5.2 costing approximately $0.17 per vulnerability found.
Professor denounces mass AI fraud on an exam at Brown
opinion
8 min read
Roberto Serrano, a professor at Brown University, claims to have overwhelming evidence of widespread cheating using AI among his students. He emphasizes the need for a thorough discussion on the implications of AI technology for academic integrity in higher education.
I used Claude Code to get a second opinion on my MRI
opinion
5 min read
Opus 4.8 provides analysis tools for interpreting MRI results, offering users a secondary opinion on medical diagnoses. The platform aims to assist individuals in understanding their MRI findings, although it is not a substitute for professional medical advice.
Tokenmaxxing is dead, long live tokenmaxxing
opinion
17 min read
Tokenmaxxing remains a controversial topic, with claims of its decline being overstated. Consumers often invest significant amounts in products and services that may not provide a clear return on investment, similar to behaviors seen in gambling apps.
Reflections on software engineering in the age of AI
opinion
13 min read
The software industry increasingly relies on artificial intelligence for coding tasks. AI has become proficient at writing code due to its training on vast amounts of publicly accessible source code and the structured nature of programming languages.
Flock cameras track more than your license plate, and they're spreading fast
news
9 min read
Flock cameras, commonly known as automated license plate readers (ALPRs), are rapidly spreading across the United States and can track more than just license plates. These cameras are typically mounted along roadways, logging the movements of vehicles that pass within their field of vision.
Austria Lobbies EU to Host Anthropic After US Access Curbs
news
1 min read
Austria is advocating for the European Union to host Anthropic PBC in response to US restrictions on access to advanced AI models for foreigners. Austrian State Secretary for Digitalization Alexander Pröll emphasized the need for member states to consider the strategic establishment of Anthropic within the EU.
Google limits Meta's use of its Gemini AI models
news
1 min read
Google has restricted Meta's access to its Gemini AI models due to Meta's request for more computing capacity than Google could supply. This limitation has caused disruptions and delays in some of Meta's internal AI projects.
A way to exclude sensitive files issue still open for OpenAI Codex
tool
1 min read
A mechanism is proposed to mark sensitive files or paths that should not be read or sent to the model, applicable at both repository and global levels. This feature aims to enhance privacy and security in AI interactions with code.
More evidence is consistent with possible ancient life on Mars (2025)
opinion
4 min read
Recent geological evidence suggests the possibility of life on Mars, but scientists caution that non-life explanations may also account for the findings. Historical patterns show that previous claims of discovering life on the Red Planet have often been met with skepticism.
Wayfinder Router: deterministic routing of queries between local and hosted LLM
tool
20 min read
Wayfinder Router is a CLI tool that enables deterministic routing of queries between local and hosted LLM models without making model calls to decide the route. It analyzes the structure and wording of prompts to determine the appropriate model for processing, allowing for offline calibration on user data.
Ford hired AI and sacked humans. It backfired badly
news
2 min read
Ford has rehired over 350 veteran engineers after its AI adoption strategy led to significant mistakes. The company acknowledged the need for experienced human oversight to correct issues caused by automated systems.