Peter Salus has died
Community Discussion
The comments collectively express respect and appreciation for Peter H. Salus’s contributions to Unix history, noting his influential books and leadership in early USENIX. Readers use his works both as comprehensive references and as targeted look‑ups, while seeking guidance on which titles remain essential today. There is a shared concern that contemporary developments, such as large‑language‑model research, lack comparable oral‑history documentation, prompting interest in finding or creating similar archival efforts. Overall, the tone is nostalgic, grateful, and inquisitive about preserving current technological narratives.
The last six months in LLMs in five minutes
Summary
- **November 2025 inflection point**: Model leadership shifted five times among Anthropic’s Claude Sonnet 4.5, OpenAI’s GPT‑5.1 (and GPT‑5.1 Codex Max), Google’s Gemini 3, and Anthropic’s Claude Opus 4.5, with Opus 4.5 holding the top spot for most of the following months.
- **Coding agents**: Both OpenAI and Anthropic applied Reinforcement Learning from Verifiable Rewards to their code‑generation agents, raising output quality from “often‑work” to “mostly‑work” and making them reliable for daily development without extensive post‑editing.
- **Personal AI assistants (“Claws”)**: The Warelay repository, initiated in November, rapidly evolved (Warelay → CLAWDIS → CLAWDBOT → OpenClaw) into a widely discussed personal assistant platform. Demand for local hardware grew, exemplified by rapid Mac Mini sales for running these assistants.
- **Model releases**: Google launched the Gemma 4 series, the most capable open‑weight US models to date. China’s GLM released GLM‑5.1, a 754 B‑parameter, 1.5 TB model demonstrating strong performance when sufficient hardware is available.
- **Overall trends**: The past six months are defined by (1) dramatically improved coding agents and (2) laptop‑class open‑weight models surpassing earlier expectations.
Read full article →
Community Discussion
Comments show a blend of optimism about recent incremental gains—especially in producing compilable code and measurable benchmark improvements—and skepticism that these advances represent a true “inflection point” rather than marketing hype. Many note that models still falter on higher‑level tasks such as precise image generation, nuanced documentation, and genuine understanding, emphasizing pattern synthesis over abstraction. References to newer models, community documentation, and hard‑coded progress dates illustrate a desire for concrete evaluation, while overall sentiment remains cautious, recognizing both useful advancements and persistent fundamental limitations.
Click (2016)
Community Discussion
The discussion reflects mixed reactions: many find mouse‑tracking and session‑replay tools invasive and unsettling, emphasizing privacy concerns despite disclosures; others view the demos as eye‑opening, educational, or entertaining illustrations of web tracking capabilities. Technical participants note implementation details and limitations, while some express curiosity and appreciation for the concept. Overall sentiment leans toward discomfort with privacy implications, balanced by interest in the technology’s demonstrative value.
Anyone on the Internet Can Ring Your Doorbell
Summary
The post documents critical security flaws in the low‑cost “Smart Doorbell X3” sold on Temu and backed by Naxclow’s cloud platform (Guangzhou Qiangui IoT).
- The device communicates with the backend over plain HTTP; alerts, authentication tokens and permanent P2P credentials are transmitted in cleartext, allowing any internet user to hijack an owner’s account, redirect live video feeds, and generate fake calls.
- The backend returns a static “confkey” containing device‑wide credentials that survive factory reset and account rebinding, enabling persistent impersonation.
- Media streams (JPEG video, audio) and token exchanges on the P2P leg are unencrypted, exposing live audio/video and long‑lived tokens to passive network observers.
- UART access (easy to reach on the front panel) leaks the Wi‑Fi SSID, PSK, and WPA TK/GTK keys during boot, granting full network compromise from a single doorbell.
- Firmware runs on a Beken BK7252N MCU with RT‑Thread 3.1.0; OTA update is non‑functional, and debug registers are printed on each boot.
Responsible disclosure was made to Naxclow (no reply) and a coordination case opened with CERT/CC for CVE assignment.
Read full article →
Community Discussion
The comment expresses ambivalence toward smart doorbells, weighing potential emotional benefits of receiving unexpected visits against feelings of loneliness that such rings might trigger. It questions whether the technology would genuinely improve security, showing skepticism about its material advantages. The author also underscores a strong preference for limiting smart devices, citing reliance on a traditional printer and a defensive posture toward unfamiliar electronic sounds, indicating a broader reluctance to adopt newer home‑automation products.
Regex Chess: A 2-ply minimax chess engine in 84,688 regular expressions
Summary
The article presents **Regex Chess**, a 2‑ply minimax chess engine realized entirely with 84,688 regular‑expression substitutions. The author builds a “regex CPU” that treats a single string as the machine state, encoding a stack, variables, and a branch‑free, conditional‑execution ISA. Core instructions include **push**, **pop**, **lookup**, **assign_pop**, equality tests, and a conditional that toggles execution by temporarily removing the required “%%” header. Because regex replacements are applied globally, the system supports **SIMD‑style threading**: multiple independent “threads” are created with “%%” markers and operated on simultaneously. Fork instructions duplicate threads, enabling exhaustive exploration of boolean branches (useful for chess move generation). Loops are impossible, so all computation is unrolled; bounded tasks like generating a chess move fit this model. A compiler translates Python‑style pseudo‑code into the regex assembly via symbolic execution, recording operations (push, lookup, binary_add, etc.) and handling conditionals by branching the trace and reactivating tags. The resulting engine reads a board string, runs the regex list, and outputs a legal move.
Read full article →
Community Discussion
The comments express strong enthusiasm for the unexpected capabilities demonstrated, with many noting surprise and admiration for the impressive computation achieved with regex and the novelty of seeing complex concepts like NFAs implemented in practice. Several users highlight personal fascination and a sense of discovery. However, a recurring complaint concerns frequent “illegal move” errors on Chrome for Android, suggesting reliability issues, and some question the engine’s handling of simple checkmates, indicating perceived deficiencies alongside the overall excitement.
Anthropic acquires Stainless
Summary
Anthropic announced the acquisition of Stainless, a company founded in 2022 that specializes in SDK generation and MCP server tooling for API integration. Stainless has produced the official Anthropic SDKs since the Claude API’s launch, automatically converting API specifications into native libraries, CLIs, and connectors for languages such as TypeScript, Python, Go, Java, and Kotlin. The acquisition aims to extend Claude’s ability to connect agents with external data and tools, leveraging Stainless’s platform to improve developer experience and agent connectivity. Anthropic’s Head of Platform Engineering highlighted that agents depend on robust integrations, while Stainless’s founder noted the alignment of their focus on high‑quality SDKs with Anthropic’s platform. The combined teams will continue developing MCP‑based connectivity solutions for the Claude ecosystem.
Read full article →
Community Discussion
Comments reflect a mixed reaction to Anthropic’s acquihire of Stainless. Many acknowledge the talent gain and congratulate the team, while others express disappointment that the SDK generator service is being discontinued, fearing disruption for existing users and potential anti‑competitive motives. Observers note a pattern of large AI firms acquiring developer‑tool startups, questioning whether the moves target technology, talent, or market control. Concerns about reliance on commercial offerings and calls for open‑source alternatives also appear frequently. Overall, sentiment balances appreciation for the acquisition’s staffing benefits with criticism of its impact on the ecosystem.
Pope Leo XIV’s first encyclical Magnifica humanitas to be published May 25
Summary
Pope Leo XIV’s inaugural encyclical, *Magnifica humanitas*, addresses the preservation of the human person amid advances in artificial intelligence. The document, signed on May 15 — the 135th anniversary of Pope Leo XIII’s *Rerum novarum* — will be released on May 25 2026. A presentation will occur at 11:30 a.m. in the Vatican’s Synod Hall, featuring:
- Pope Leo XIV
- Cardinal Víctor Manuel Fernández, Prefect of the Dicastery for the Doctrine of the Faith
- Cardinal Michael Czerny, S.J., Prefect of the Dicastery for Promoting Integral Human Development
- Professor Anna Rowlands, theologian, Durham University
- Christopher Olah, co‑founder of Anthropic and AI interpretability researcher
- Professor Leocadie Lushombo, I.T., political theology specialist, Jesuit School of Theology/Santa Clara University
Cardinal Secretary of State Pietro Parolin will deliver closing remarks, followed by a papal address and blessing. The encyclical focuses on ethical and doctrinal considerations of AI’s impact on human dignity.
Read full article →
Community Discussion
The comments express a mix of caution and curiosity about the Pope’s upcoming AI‑focused encyclical, criticizing the current headline as misleading about authorship and questioning the involvement of an atheist‑identified tech founder. Many hope the papal voice will reaffirm human dignity amid rapid automation, while others doubt the relevance of religious input, suspecting a publicity or power‑consolidation motive. There is interest in how the document will relate to prior Vatican AI guidelines, but overall sentiment is skeptical of the partnership’s necessity and intent.
War game exposed U.S. vulnerability to low-tech warfare
Summary
A declassified after‑action report from the 2002 “Millennium Challenge” war game reveals that a simulated U.S. Navy battle group was defeated in ten minutes by an opponent using commercial vessels and other low‑tech, unconventional tactics. The report—part of a $250 million exercise—identified a critical vulnerability to low‑technology warfare that later manifested in the 2003 Iraq invasion and subsequent conflicts. Marine Lt. Gen. Paul Van Riper, who led the anti‑U.S. forces in the simulation, criticized the exercise as “rigged.” National Security Archive fellow Nate Jones obtained the document after filing a FOIA request in 2013; following a mandatory declassification review by five agencies, the Pentagon released portions of the report in November 2024. The findings underscore the need for the U.S. military to adapt to asymmetric threats that exploit civilian assets and non‑conventional methods.
Read full article →
Community Discussion
The comments convey a critical assessment of the 2002 Millennium Challenge war game, highlighting perceived unrealistic assumptions about asset deployment, communication, and load‑out capacities. They argue that the exercise underscored vulnerabilities in the U.S. Navy, especially the high cost and limited defensive capability of aircraft carriers against swarms of inexpensive drones and low‑cost cruise missiles now being fielded. The discussion stresses that modern threats have amplified these weaknesses, suggesting a need to reconsider force composition and invest in cheaper, asymmetrical weapon systems.
We stopped AI bot spam in our GitHub repo using Git's –author flag
Summary
The Archestra team reports that AI‑generated bots have flooded their GitHub repository with low‑quality comments, PRs, and issues, drowning legitimate contributions and creating security risks. To mitigate this “AI slop,” they disabled issue, PR, and comment creation for users who have not completed an onboarding process, effectively using GitHub’s “Limit to prior contributors” setting. Since GitHub only recognises accounts with a commit on the main branch, the team devised a workaround: during onboarding they generate a commit authored with the user’s GitHub noreply email ( `+@users.noreply.github.com` ) via `git commit --author`, push it to main, and record the handle in an EXTERNAL_CONTRIBUTORS.md file via a GitHub Action. This grants the user prior‑contributor status, allowing them to comment and submit PRs. The process includes an ethical‑AI agreement and CAPTCHA. The post stresses that while AI‑driven activity inflates GitHub metrics, it harms open‑source collaboration and requires active filtering and custom workflows to preserve repository quality.
Read full article →
Community Discussion
Overall sentiment is critical of GitHub’s handling of pull‑request spam and AI‑generated contributions, highlighting security gaps where contributors gain elevated rights without robust verification. Commenters propose stronger identity checks, reputation or ELO‑style scoring, financial deposits, or real‑name requirements, while noting that simple captchas or temporary limits provide only limited relief. Frustration is expressed over the large volume of low‑quality AI PRs and the difficulty of distinguishing genuine work, with occasional acknowledgement that incremental tool improvements can help but are seen as insufficient.