tink / links
1469 saved links · newest first
trevdev
Trev :vim: (@trevdev@fosstodon.org)
(fosstodon.org)
Trev asks when agentic coding ceases to be "vibed slop," offering four options: never, only if agents answer questions, only after full human review, or only after a complete manual rewrite. The post is a poll with two participants and no further context, serving as a light social media prompt rather than a technical argument or detailed analysis.
knighthk-irc
jwz 8.5x11 CMYK
(www.jwz.org)
jwz 2026 argues there is no ethical use of AI, labeling it a disgusting machine that must be destroyed. The piece contends that using AI ignores its staggering climate impact, labor attacks, and disregard for consent, effectively advancing the agenda of "hollow-eyed" money-ghouls who resent art and life. It frames AI as based on wholesale theft and plagiarism while enriching "cringe billionaire fascist motherfuckers." Consequently, jwz asserts that any use of the technology is a selfish negotiation over acceptable harm to others.
knighthk-irc
503 Service Temporarily Unavailable
(stackoverflow.com)
# 503 Service Temporarily Unavailable
rrix
What Is This?
(engine.arcology.garden)
The Arcology Project is a personal, mobile-first knowledge management system built on org-mode and org-roam, designed to replace traditional Emacs tooling with native, offline-capable applications. It explicitly rejects the open-source community model in favor of an opinionated, idiosyncratic setup tailored for a single user’s workflow, relying on federated infrastructure like Syncthing for synchronization rather than corporate cloud services. The system is developed using literate programming in Kotlin, where org-mode documents serve as the primary source for generating both application code and system configurations.
The ecosystem comprises three main components: an Android app supporting over 50,000 headings with BM25-ranked full-text search; Arroyo, a concept operating system that generates NixOS and Emacs configurations from documentation-driven metadata; and a web publishing server that exposes org-roam headings across multiple domains via RSS and Fediverse integration. The project is distributed as a published personal system rather than a collaborative product, encouraging users to fork the codebase at `code.rix.si/rrix/arcology2go` if they wish to adapt it for their own use.
alternateved
Checking you are not a bot
(todo.sr.ht)
The page is a bot-checking challenge from gammaspectra.live, blocking access to verify the user isn't an AI scraper. It cites mass scraping causing downtime and advises disabling JavaScript-blocking plugins like JShelter to proceed.
antti
GrapheneOS: "Earlier today, ARIN fulfilled …" - GrapheneOS Mastodon
(grapheneos.social)
GrapheneOS has secured an IPv4 /22 allocation (206.220.120.0/22) from ARIN via their waiting list, bringing their total owned space to six /24 blocks including existing allocations for anycast DNS. The project plans to deploy the new block across four locations as separate /24s, starting with a Toronto server hosted by Xenyth, a Canadian provider chosen to keep the mail server within the same legal jurisdiction as the GrapheneOS Foundation. This setup allows the non-profit to announce its own IP space via BGP for inbound traffic engineering without relying on shared upstream blocks.
The post clarifies that the IP space was assigned rather than purchased, with annual maintenance fees currently at $1,100 under ARIN's X-Small tier. GrapheneOS notes that while this provides flexibility for future scaling and direct peering, it also shifts responsibility for abuse handling and network stability to the organization itself. The project argues this infrastructure investment supports its mission by ensuring legal alignment and technical control over its core services, such as mail and attestation, which are being moved to Toronto.
bergheim
System Shock 2: 25th Anniversary Remaster | Download and Buy Today - Epic Games Store
(store.epicgames.com)
Nightdive Studios has released the System Shock 2: 25th Anniversary Remaster, a modernized version of the classic FPS/RPG that adds cross-play co-op multiplayer and mod support to the original 1999 experience. The update includes three military branches (O.S.A., Marines, or Navy) for varied playstyles, enhanced character and weapon models, and optimized controller support. Performance specs claim up to 144FPS with ultra-widescreen support, running on modest hardware requirements like a dual-core CPU and 2 GB of RAM for the minimum configuration.
The game is currently available for free on the Epic Games Store as part of a limited-time promotion ending October 8, 2026, down from its standard price of $29.99. Rated Mature 17+ by the ESRB for blood, gore, and violence, the title has received a high player rating of 4.7 on the platform. While the store page lists these features and ratings, it provides no technical analysis or critical review of how well the remaster handles the original's complex mechanics, leaving the quality of the "modernized" experience to be judged by the free trial itself.
bergheim
Omarchy and NixOS
(discourse.nixos.org)
A NixOS forum thread asks whether the community should build a "batteries-included" flake to rival Omarchy, a preconfigured Arch Linux distribution that is rapidly attracting new users. The original poster cites Omarchy's momentum and star-history chart as evidence of a demand for easy onboarding, suggesting a similar project could lower the barrier for newcomers to NixOS.
The discussion has generated significant traffic—15.3k views and 1.2k likes—but the provided text only lists top replies without revealing the actual arguments. Early responses point to existing resources like the NixOS Configuration Collection wiki and link to an xkcd comic about "fastest-growing" projects, hinting at skepticism toward trend-chasing or a preference for maintaining existing documentation over creating new, potentially redundant projects. The thread serves as a debate on whether a centralized, opinionated default configuration is necessary for NixOS growth or if the current ecosystem already fulfills that role.
bergheim
Nixpkgs core team disbands after governance reform
(github.com)
The Nixpkgs core team, led by emilazy and alyssais, has disbanded after only ten months in existence, citing unsustainable burnout and a failure to recruit new members. While the team points to concrete achievements like onboarding 19 new committers and establishing an initial automation/AI policy, they argue that their lightweight, consensus-based role was fundamentally incompatible with the NixOS Steering Committee (SC). The core complaint is systemic: the SC lacks a "native instinct" for delegation, resulting in chronic micromanagement, poor communication, and a lack of trust in the team’s autonomy to handle matters within its constitutional remit.
This is a direct critique of the current governance structure, arguing that the top-down representative model creates a drag on technical leadership rather than supporting it. The team notes that their high-trust consensus approach was better suited for local dispute resolution than the SC's majority-vote method, which they view as fostering combative, zero-sum dynamics. With the team dissolved and both members stepping back from Nixpkgs involvement, the administrative vacuum is left to the SC as a final backstop, leaving the project without a dedicated body for project direction and coordination until a new leadership structure is formed.
bergheim
The Nixpkgs core team has disbanded
(discourse.nixos.org)
The Nixpkgs core team has disbanded after roughly ten months of operation, citing burnout and a failure to recruit replacements as the primary drivers. The announcement details specific achievements during their tenure, including reforming the committer delegation process, onboarding 19 new committers, securing a GitHub Enterprise Cloud upgrade, and establishing an initial automation/AI policy. However, the core argument is that the team’s role proved incompatible with active technical contribution due to systemic dysfunction in the NixOS Steering Committee (SC).
The post argues that the SC lacks the instinct for delegation envisioned by the constitution, resulting in micromanagement, poor communication, and a lack of trust in the core team’s autonomy. This dynamic created a "consistent drag" on work, making it impossible to fulfill their constitutional mandate for project direction and coordination without sacrificing the health of the two remaining members. The team contends that their high-trust, consensus-based decision-making model was superior to the SC’s majority-vote approach and successfully resolved community tensions, but the environment ultimately forced them to step down rather than continue under what they describe as unresponsive and combative governance structures.
alternateved
paste0r
(share.glvortex.net)
The page is a file hosting stub for a 4.5 MB MKV video named "monkey-eagle-rat" on the paste0r service. It provides no descriptive metadata, only download and management links.
bergheim
paste0r
(share.glvortex.net)
A paste0r instance hosts a 7.6 KB `steam.desktop` file under the identifier emu-sheep-zebra, offering direct download links and QR codes without further context or metadata.
bergheim
paste0r
(share.glvortex.net)
The page is a user dashboard for the paste0r service, listing two text uploads and an empty redirect table. It contains no article content, technical claims, or data to summarize.
alternateved
paste0r
(share.glvortex.net)
paste0r hosts a 4.5 MB MKV video named spider-frog-frog on share.glvortex.net. The page provides download and management links but contains no descriptive text or metadata about the file's content.
alternateved
paste0r
(share.glvortex.net)
paste0r is a text sharing service offering configurable expiration times, read limits, and syntax highlighting for numerous programming languages. It provides privacy options ranging from public to secret, along with password protection and file attachments. The page displays only the upload interface controls without any specific content or technical details.
bergheim
Why Sites Should Block AI Crawlers and How to Do It
(www.jamescherti.com)
James Cherti argues that independent publishers should block AI training crawlers like GPTBot and ClaudeBot to stop resource theft without sacrificing referral traffic, relying on the technical distinction between bots used for model training and those used for live search indexing. The piece leans heavily on Cloudflare Radar data showing extreme crawl-to-refer ratios—Anthropic at 2,585:1 and OpenAI at 685:1—to demonstrate that these companies scrape thousands of pages for every single click they return. Cherti provides a specific `robots.txt` configuration and server-level enforcement methods (Nginx, Apache, or Cloudflare AI Crawl Control) to drop these specific user agents while keeping search bots like OAI-SearchBot and PerplexityBot active.
The advice is practical but relies on the assumption that AI companies will maintain separate crawler identities indefinitely, a premise Cherti admits requires periodic auditing as infrastructure changes. He correctly notes that `robots.txt` is advisory and suggests server-level blocking for real protection, though he understates the difficulty of distinguishing legitimate user-triggered agents from scrapers at the edge. The core argument holds up for bandwidth-conscious sites: if you do not need your content in a specific LLM's training data, blocking its dedicated crawler costs you nothing in search visibility, provided you avoid accidentally blocking the distinct agents responsible for citation and retrieval.
antti
YouTube TV-Wiedergabefehler und Anmeldehinweise
(youtu.be)
The page displays a YouTube TV playback error and login prompt. It advises restarting the device if video does not start, warns that unlogged viewing affects recommendations, and reports a sharing information retrieval failure. No substantive content is present.
bergheim
20261001T160616--its-all-a-bit-busy-though__screenshot_public.webp
(share.glvortex.net)
A screenshot shows data for a weightlifting app
antti
YouTube TV Playback and Sharing Error Messages
(www.youtube.com)
YouTube's TV interface displays generic error messages for playback failures and sharing issues, such as "Falls die Wiedergabe nicht in Kürze beginnt" and "Bei dem Versuch, Informationen zum Teilen abzurufen, ist ein Fehler
trev
The Skate 3 Rust Rewrite
(www.youtube.com)
The Skate 3 Rust Rewrite
trev
Bevor Sie zu YouTube weitergehen
(www.youtube.com)
YouTube's pre-redirect page presents a standard cookie consent notice, detailing how Google uses IP addresses and data for core service operations, spam prevention, and analytics. It distinguishes between essential tracking and additional purposes like ad personalization and new feature development, which only occur if the user selects "All accept." The text explains that rejecting all cookies limits data usage to non-personalized content based on location or current context, while accepting enables tailored recommendations and ads derived from viewing history. This is a routine privacy gate with no technical substance beyond standard web tracking disclosures.
trevdev
DHH
(github.com)
A Neovim website issue argues that a quote from DHH should be removed because it "gives Nazi bar vibes" and reveals his desire to restore Europe to its ethnic roots. The submitter cites a specific Hey.com article, *Three Sacred Cows That Must Die So Europe Can Live*, as evidence of this stance.
The claim rests entirely on the submitter’s interpretation of DHH's political views rather than any technical merit or community consensus regarding the quote itself. With 607 users reacting positively to the complaint and only 12 negative reactions, the issue reflects a strong sentiment that associating Neovim with such rhetoric is inappropriate, though it offers no alternative content or specific policy violation beyond the subjective "vibe."
trev
Neovim website pulls Mitchell name correction PR
(github.com)
The Neovim website repository accepted a pull request to remove a quote from David Heinemeier Hansson (DHH), following community backlash over his recent xenophobic comments. The change was merged by maintainer Justin M. Key on August 28, 2026, resolving issue #501. The PR thread reveals significant friction, as previous attempts to remove the quote had been rejected by maintainers who threatened to ban users for "wasting their time" with similar proposals.
The discussion highlights a disconnect between the project's stated desire for "tech celebrity" endorsements and the reality of maintaining a homepage free from controversial figures. Contributors debated replacing DHH with quotes from Drew Devault or other high-profile figures, noting that DHH’s status as a billionaire and keynote speaker makes him unlikely to face meaningful professional consequences for his views. The maintainer ultimately locked the conversation as off-topic before merging the fix, ending a debate where some participants argued that any non-controversial quote would suffice to satisfy community concerns.
trevdev
Blog post announces removal of AI-generated image
(trevdev.ca)
The author confesses to a compulsive, ADHD-driven cycle of switching between Neovim and Emacs every 60-90 days, driven by an "all-or-nothing" approach to tooling. They acknowledge that Neovim is objectively more productive for web development, citing its Lua extensibility and ability to handle multiple language servers per buffer without the friction of Emacs’s Eglot implementation. In contrast, they value Emacs for its Lisp foundation, vast command set, and Org Mode, which serves as a gateway back into the editor despite its heavier maintenance requirements.
The piece argues that while Neovim aligns better with Unix philosophy and modern JavaScript toolchains, Emacs offers a unique ethical and community appeal rooted in Free Software principles. The author aims to break this switching cycle by rooting themselves in Neovim to focus on other projects, noting they have gone 145 days without an "Emacs incident." It is a personal reflection rather than a technical benchmark, admitting that the choice between the two is less about objective superiority and more about managing personal workflow compulsions.
trevdev
Home Manager Manual
(nix-community.github.io)
The Home Manager module for the Senpai IRC client exposes standard Nix options to manage its configuration, including `enable`, `package`, and a nested `config` attribute set. The configuration supports standard connection parameters like server address, nickname, and SASL password, with specific handling for TLS versus plain-text connections via URL schemes or port defaults.
A notable security consideration is the handling of credentials: storing the password directly in `config.password` leaves it world-readable within the Nix store. To mitigate this, the module provides a `password-cmd` option that executes an external command—such as `gopass` or `pass`—to fetch the password at runtime, effectively allowing users to integrate Senpai with encrypted password managers while keeping sensitive data out of the static Nix store.
Lionyx
Readable Regular Expressions for JavaScript/TypeScript, Inspired by Emacs' rx
(rahuljuliato.com)
Rahul M. Juliato presents a small TypeScript/JavaScript library that ports Emacs Lisp’s `rx` macro into a composable DSL for building regular expressions, aiming to replace unreadable regex strings with readable code trees. The implementation treats regex components as objects with specific "kinds" (atom, seq, alt) to handle escaping and grouping logic automatically, ensuring that plain strings are treated as literals and alternations are bracketed correctly when nested. A standout feature is the `or` function, which sorts string branches by length to mimic Emacs’s `regexp-opt` behavior, preventing JavaScript’s default left-to-right matching from incorrectly capturing shorter prefixes like "in" instead of "interface". The author demonstrates the utility by rewriting the complex SemVer validation regex into a structured, named-group definition that is significantly easier to audit and modify.
The post includes a full technical breakdown of the engine, covering character set merging, lazy quantifiers, and backreferences, while candidly noting limitations such as ASCII-only character classes (unlike Emacs’s Unicode support) and the absence of syntax-table-dependent forms. It also provides a cheat sheet mapping Emacs `rx` forms to their TypeScript equivalents and a `raw` escape hatch for integrating legacy regexps. The piece is a solid, practical tutorial for developers who find standard JavaScript regex syntax brittle, offering a working gist and source code rather than just conceptual examples.
Lionyx
YouTube TV-Wiedergabefehler und Geräteneustart beheben
(www.youtube.com)
YouTube TV playback error page displays generic restart advice and login warnings. Content is mostly interface furniture, offering no technical details or specific troubleshooting steps beyond the initial prompt.
knighthk-irc
Shutting down our public encrypted DNS servers and sponsoring Quad9 instead
(mullvad.net)
Mullvad is retiring its public encrypted DNS servers, operational since 2022, to financially sponsor the Quad9 Foundation instead. The company argues that running a privacy-focused public DNS service is highly specialized and that Quad9 is the "undisputed leader" in the field, making Mullvad’s independent effort redundant. While these servers remain useful for protecting DNS queries outside of the Mullvad VPN or when using the Mullvad Browser, the firm contends its resources are better spent supporting an existing expert rather than duplicating partial efforts.
Users must migrate to Quad9 before November 2nd, 2026. Mullvad Browser users with default settings will be automatically switched, but those with customized configurations must manually revert to defaults or update their settings. Additionally, existing iOS and macOS configuration profiles for Mullvad’s
Lionyx
Verifying your browser…
(safereddit.com)
The page is a browser verification wall powered by Anubis, which uses a Hashcash-style proof-of-work challenge to deter AI scrapers. It explains the scheme adds negligible load for humans but raises costs for mass scraping, noting it relies on modern JavaScript features that plugins like JShelter will block.
antti
Reddit is killing RSS feeds and ending public API access because of AI bots
(techcrunch.com)
Reddit is shutting down RSS feed support on November 13, 2026, and will end public API access by March 2027, citing "large-scale scraping and automated abuse" as the primary justification. The move effectively closes off free, standardized access to Reddit’s user-generated content, a shift that aligns with the company’s growing revenue from AI licensing deals; its non-advertising revenue grew 24% year-over-year to $43 million in Q2 2026, suggesting a strategic preference for monetizing data through commercial agreements rather than open access. While Reddit offers a Discord Relay app as a workaround for moderators, it explicitly admits there is no replacement for RSS feeds used by external tools or researchers, leaving third-party developers and social listening products without a viable path forward unless they secure specific commercial deals.
The announcement includes additional restrictions on Old Reddit, limiting access to logged-in users who have been active in the last 90 days, and requires developers to register their apps by January 12, 2027, to avoid losing API access entirely. Although framed as a security measure against bots, the timing and scope of these changes strongly imply that Reddit is consolidating control over its data assets to maximize licensing revenue from AI companies. The decision has already drawn criticism from moderators who rely on RSS for workflow management and users who preferred the open-web format, marking a significant retreat from interoperability standards in favor of a walled-garden model.
trev
all the people · tink
(tink.glvortex.net)
tink's "all the people" is a single image captioning Lionyx standing in a crowded plaza, described as looking like he just won the lottery. The post is a brief IRC reply from #systemcrafters with no technical content or argument.
trev
404
(trevs.site)
The requested GitHub Pages URL returns a 404 error, indicating the file does not exist on the server.
antti
Some Questions and Answers
(inoti.fyi)
File-notification APIs on Linux, Android, Windows, and macOS leak user behavior because they report file events even when the attacker lacks read permissions for the files themselves. Accepted at CCS 2026, the research from Graz University of Technology demonstrates that an unprivileged local process can reconstruct keystroke timing via `inotify` on `/dev/input`, bypass Android’s FUSE storage isolation to track WhatsApp media transfers with file names, and monitor Windows filesystem events system-wide by watching the root directory. The Windows finding is particularly stark: monitoring `C:\` reveals the full path of every touched file across users, allowing an attacker to identify visited websites with 97.8% accuracy for top-1000 sites via browser cache directories. Microsoft has labeled this behavior an "undocumented feature" and offers a registry policy to fix it, but that policy is disabled by default, leaving the vulnerability exploitable out-of-the-box.
The Linux issue received a partial kernel mitigation in early 2026 (CVE-2025-68788) by suppressing access events on character devices, which addresses the most severe `/dev/input` keystroke leakage but leaves other notification vectors intact. On Android, no fix is provided, and the attack remains fully functional against private app folders. The paper correctly notes that file contents are never leaked, only metadata such as names and timestamps, but argues this is sufficient for high-precision side-channel attacks when combined with existing timing analysis research. The work stands on solid empirical evidence rather than theoretical speculation, effectively exposing a systemic design flaw in how major operating systems handle filesystem observation permissions relative to data access controls.
antti
YouTube App Fehlerhinweise und Anmeldeanforderungen
(www.youtube.com)
The YouTube app is broken, displaying generic error messages and a "not logged in" warning that implies viewing history tracking. The interface shows failed share attempts and a bot-verification prompt, indicating significant UI instability rather than functional content.
Lionyx
b2e613.jpg
(i.imgflip.com)
a cartoon that shows a man in pigtails with a huge exclamation next to it
Lionyx
File:TSTO Ralph-O-Cop.png - Wikisimpsons, the Simpsons Wiki
(simpsonswiki.com)
Click on a date/time to view the file as it appeared at that time.
Naokotani
tealdeer
(github.com)
Tealdeer is a Rust-based client for the tldr community-driven man pages, aiming to be faster and more frictionless than standard implementations. It caches pages locally so that no network connection is required after the initial update, while complying with the official tldr client specification. The project currently holds 6.6k stars and provides static binaries for easy installation across Linux, macOS, and Windows.
The repository includes a notable "AI Policy" that discourages the use of AI tools in contributions, requiring contributors to clearly mark any AI-generated text and avoid posting low-quality "slop." It also mandates leaving small tasks for new human contributors to preserve learning opportunities. The codebase is dual-licensed under Apache 2.0 and MIT, with documentation generated via mdbook and a defined Minimum Supported Rust Version (MSRV) policy that requires the stable Rust version to be at least a month old before release.
antti
tuigreet/contrib/assets/screenshot.png at master · NotAShelf/tuigreet · GitHub
(github.com)
A screenshot asset for the tuigreet project. No substantive content to summarize.
antti
Sponsor tuigreet/tuigreet
(github.com)
tuigreet is a Rust-based terminal greeter for greetd, currently at version 0.12 with a minimum supported Rust version of 1.95. It builds on the original tuigreet foundation to offer a more maintainable codebase and polished UI, featuring session persistence via `--remember` flags, NSS-backed user menus, and themeable components. The project has recently migrated its Nix build system to crane for incremental builds and pinned the toolchain via manifest, while also enabling thin LTO and setting codegen-units to 1 for release builds.
A notable recent addition is support for animated backgrounds, including "doom" fire effects and "matrix" rain, which render at a default 30 FPS (compared to the standard 2 FPS UI tick) without compromising prompt legibility. These animations are configurable via CLI flags like `--background-fps` and can be switched on-the-fly using the F4 key. Installation options include pre-built binaries, AUR packages for Arch Linux, Nixpkgs derivations, and source builds, with the project explicitly encouraging community contributions for additional distribution packaging.
Naokotani
KRS-One and 50 Cent Headline Hip-Hop Mix
(www.youtube.com)
This is a YouTube playlist page titled "KRS-One and 50 Cent Headline Hip-Hop Mix," containing a list of 25 official music videos. The tracklist features classic hip-hop hits from the late 1990s and early 2000s, including KRS-One’s "Sound of da Police," 50 Cent’s "In Da Club," Dr. Dre’s "Still D.R.E.," Cypress Hill’s "Insane In The Brain," and multiple tracks by Eminem and DMX. The page is primarily a navigation interface for the playlist, with links to each video on YouTube.
bergheim
Sigil
(codeberg.org)
Sigil is a Scheme dialect that compiles to standalone binaries for desktop and web, targeting users who want R7RS Small compatibility without runtime dependencies. The current master branch tracks version 0.22.11, which fixes memory-safety issues and introduces opt-in native dependencies, while the project has recently shifted its license to BSD-3-Clause. The codebase is heavily shell-based (68.5%) with JavaScript and Python components, and it integrates tightly with Emacs for a REPL-driven workflow. A notable recent optimization reduced native-codegen web build sizes by roughly 70%, and the toolchain now pins Guix for reproducible releases.
The project’s most significant disclosure is its "AI Disclaimer," which admits the codebase was developed almost entirely using Claude Code (Opus 4.5). This stands in direct contrast to the platform it resides on, as Codeberg recently updated its Terms of Use to restrict LLM-generated content and ban cryptocurrency projects. While Sigil offers standard features like pattern matching, delimited continuations, and channel-based concurrency, the heavy reliance on a specific AI model for its development makes it a prime candidate for the very policy changes currently being enforced on its host.
bergheim
Sigil - Infrastructure for AI Minds | Data Provenance at the Type Level
(sigil-lang.com)
Sigil is a proposed programming language that embeds data provenance directly into the type system, using symbols like `!` (known), `~` (reported/untrusted), and `?` (uncertain) to enforce trust boundaries. The core argument is that mainstream languages allow untrusted external data—such as API responses—to be silently treated as verified truth, whereas Sigil makes this a compile-time error until explicit validation occurs. It pitches itself as "infrastructure for AI minds," offering a nine-layer framework for autonomous agents that includes memory systems (Engram), planning modules (Omen), and security sandboxes (Aegis).
The project claims significant code density advantages, citing a ToDo app example where Sigil requires 42 lines versus 68 in Rust, attributing the difference to "morpheme operators" like `|φ` for filtering. It also advertises a broad ecosystem including a GUI framework, game engine, and tensor library with compile-time shape algebra, alongside new SIMD (AVX-512) and CUDA backends in version 0.4.0. However, the pitch is heavily speculative; while the type-level provenance concept is technically coherent, the "AI infrastructure" layers are described as abstract features without implementation details or benchmarks. The density comparison relies on a single, simplified example that likely omits the boilerplate required for Sigil's complex validation logic, making the 38% line reduction claim unsubstantiated by real-world complexity.
bergheim
“You Said No MCP!”
(earendil.com)
Pi, the agent harness previously famous for refusing MCP support, now includes it in the core. The developers justify this reversal not by admitting they were wrong, but by arguing that MCP has matured into something closer to OpenAPI with intelligent tool discovery, and that integrating it forced useful architectural changes. Specifically, MCP integration enabled "Codemode," a JavaScript sandbox that runs on the trusted harness side rather than in an untrusted shell environment. This allows agents to orchestrate tool calls with more flexibility and state management, such as using the Jev classifier model to analyze Linear issue tracker data without polluting the main context window.
The post is largely a strategic pivot dressed up as technical evolution. It concedes that MCP remains hard to compose and that many servers still rely on inefficient text dumps rather than structured data, yet argues that embracing the standard is better than standing on the sidelines. The inclusion of Codemode is presented as the solution to these lingering issues, though the claim that this "solves quite a few" traditional problems relies heavily on the success of their specific JavaScript-based approach rather than broad industry adoption. It is a clear admission that their previous stance was static and that they are now willing to trade purity for ecosystem relevance.
antti
Mastodon: Chronological Social Media Without Algorithms
(mastodon.xyz)
mastodon.xyz is a landing page for a Mastodon instance, claiming it offers chronological social media without algorithms, ads, or clickbait.
bergheim
Jeeves – Reasoning improves Jev-like decision models
(github.com)
Jeeves, a 9B decision model from PostHog built on Qwen3.5-9B, argues that adding a reasoning step significantly improves the accuracy of "Jev-like" classifiers without sacrificing their calibrated probability outputs. By training with SFT and CISPO reinforcement learning, the model achieves a 0.889 accuracy on held-out test data, outperforming both Kev-9B (0.822) and the original Jev (0.857). It also leads on JevBench's public tiers with a score of 0.935 compared to Jev's 0.866. The system uses a pointer head and a block-4 diffusion drafter to speed up inference, reaching a median latency of 3.3 seconds per request on an H100 GPU when thinking is enabled.
The technical tradeoff is clear: while Jeeves excels at out-of-domain tasks and rule-based logic (scoring 1.000 on held-out structures), it lags behind Jev on pure knowledge retrieval, such as MMLU (0.793 vs 0.900) and MMLU-Pro (0.739 vs 0.840). Additionally, the "thinking" process introduces tail latency, hitting 17 seconds at the p90 mark, though this can be mitigated by truncating the reasoning chain or using a confidence threshold to skip thinking for easier queries. The project provides full training code and weights, allowing users to serve the model via a Jev-compatible API on CUDA or Apple Silicon hardware.
antti
MOTD v0.19.2 Release Prepares Latest Commit
(github.com)
motd is a native Android IRC client written in Kotlin with Jetpack Compose, designed to mimic the Telegram chat interface with grouped bubbles, inline images, and day separators. It speaks IRCv3 and pairs best with a soju bouncer for features like infinite scrollback via `draft/chathistory` and cross-device read state sync, though it functions on plain networks by falling back to local history. The v0.19.2 release highlights hardened transport options including TLS with SASL PLAIN/EXTERNAL, IRCv3 STS pinning, and optional obfuscation via SOCKS5, Tor, or embedded VLESS (TCP + REALITY or WebSocket + TLS). Push notifications are handled through UnifiedPush with on-device decryption, or via a persistent foreground socket.
The project is distributed on F-Droid and GitHub under the GNU GPL v3.0 license, requiring Android 8.0+ on 64-bit ARM devices due to the arm64-v8a-only embedded obfuscation transport. Development uses a Nix flake pinning JDK 21 and the Android SDK, with an e2e harness for regenerating screenshots. The repository notes that while large language models assisted in development, the maintainer remains responsible for the code and releases.
antti
v5.8.8
(github.com)
Podman v5.8.8 fixes CVE-2026-94603, a vulnerability where running a checkpoint image could disable all sandboxing, including user-specified security settings. To address this, the release removes support for checkpoint images in `podman run` entirely. The maintainers argue that checkpoints ignore user-defined security configurations and are too difficult to run safely given the differing security models between standard images and checkpoints. This breaking change effectively deprecates a feature to enforce stricter sandboxing defaults.
antti
Initial GNU Announcement by Richard Stallman
(www.gnu.org)
Richard Stallman’s September 27, 1983 announcement laid out the blueprint for the GNU Project, declaring his intent to build a complete, Unix-compatible system from scratch to avoid signing nondisclosure agreements. He proposed starting with a kernel and C utilities—editor, shell, compiler, linker—before expanding to include a text formatter, YACC, an Empire game, and a spreadsheet. The design promised specific improvements over standard Unix, including longer filenames, file version numbers, a crashproof file system, and a Lisp-based window system allowing shared screens between Lisp and C programs. Stallman relied on the fixed interface specifications of Unix to coordinate distributed contributions, arguing that if individual utilities worked with existing Unix systems, they would naturally integrate into GNU.
The document includes retrospective footnotes clarifying that the original phrasing "give it away free" was a mistake in terminology; Stallman intended to grant freedom of use and modification, not necessarily zero cost. The announcement also notes that the project’s start was delayed until January 1984 and that key philosophical concepts regarding free software were only clarified years later. It remains a primary source for understanding the GNU Project’s origins, detailing Stallman’s credentials as the inventor of EMACS and his specific request for hardware donations from manufacturers to ensure early compatibility with residential machines.
bergheim
Checklist-style note-taking in Markdown
(share.glvortex.net)
Markdown lacks a native structure for expanding list items into answers without breaking syntax.
bergheim
Fix elfeed-feeds for elfeed-protocol 1.0 change
(github.com)
This commit log for bergheim's Emacs dotfiles captures a significant cleanup and performance overhaul. The author replaces writeroom with olivetti for better split-window support, swaps org-alert for the built-in appt package to fix broken appointment notifications, and removes dead code like unused smartparens workarounds and inert aggressive-indent settings. A notable addition is a custom ghostel integration that adds text objects for shell output and prompt navigation keys, addressing specific friction points with evil mode interactions.
The most concrete technical win is a startup speed optimization where trimming default org-modules and deferring org-babel language loading reduced the first org buffer open time from 3.1 seconds to 0.6 seconds. The log demonstrates careful maintenance by fixing subtle bugs, such as ensuring capture hooks run after copy-link adds IDs and correcting fboundp checks that prevented buffer switcher fallbacks from working. It is a solid, pragmatic set of changes that prioritizes responsiveness and removing unused dependencies over adding new features.