NOTE: This site has just upgraded to Forester 5.x and is still having some style and functionality issues, we will fix them ASAP.

Learning diary › Year 2025 › July, 2025 › 2025-07-05 [2025-07-05]
✍️source

     
    #agent #jujutsu #neovim #news #os #software #tla
- #formal
    - A supposedly worthwhile contract I'll never do again
        - "What mattered here wasn’t the weights being moved or the TLA⁺ spec being produced - it was the thinking that writing the TLA⁺ specification required."
            - expectation of TLA⁺ contract: catch issues early, save training time, and get a formal spec
            - real value—deep system thinking during writing—vanishes when the expert leaves
            - the dev team values the 'what do we do in this case' discussions more than the final TLA⁺ specification artifact (risking abandonment)
            - alternative: adopting Hillel Wayne’s approach: "pair-program TLA⁺ specs with team members" to let them grasp the thinking
        - HN discussion summarized
            - success of TLA contracts requires refocusing on knowledge transfer
                - teaching teams to ask critical questions
                - embedding the spec in their workflow (via checks or collaboration)
                - treating the spec as a living document rather than a one-time deliverable
- #anti-bot
    - Anubis Pilot Project Report - June 2025
        - it's interesting to see how serious business sees the adoption of Anubis
    - Anubis now supports non-JS challenges
        - this changelog is very informative
            - weight thresholds in policy is flexible
            - `robots2policy` is convenient if a site has a well-maintained robots.txt
            - no-JS challenge via metarefresh doesn't seem to battle-ready: "many modern scrapers use headless Google Chrome", "its false positive rate is ascertained"
    - How Perplexity is Evading Anti-crawling Measures
    - Robots.txt - Archiveteam #anti-bot
            - On HN
    - Iocaine: The deadliest poison known to AI
    - I was wrong about robots.txt (on HN)
- #agent/idea
    - Everything around LLMs is still magical and wishful thinking
        - Why your vibe coded app only works in your head
    - Enough AI copilots, we need AI HUDs (on HN)
    - Every reason why I hate AI
    - Riff: LLMs are Software Diamonds
    - Identify, solve, verify
    - Operators, Not Users and Programmers
            - the distinction between "user" and "programmer" is an artifact of our presently barely-programmable and barely-usable computing systems.  I would like to use the neutral word "operator" instead. —Stanislav
            - programming to be easy and simple enough to pick up that people can do it without specialized training in the field, so that they can write situated software
            - malleable languages
                - spreadsheets
                    - Every Visual Workflow Tool Is Just Excel for Developers Who Gave Up
                - WYSIWYG editors
                - browser devtools
    - AI promised efficiency. Instead, it's making us work harder
        - AI tools aren’t just saving us time—they’re shifting where we spend our mental energy. And that shift is exhausting in ways we haven’t fully recognized yet.
        - No, AI is not Making Engineers 10x as Productive
    - Lack of intent is what makes reading LLM-generated text exhausting
        - Eleven Music
            - AI is still in the uncanny valley, it still has no soul
    - LLM Inevitabilism
        - New comment by lsy in "LLM Inevitabilism"
        - New comment by mg in "LLM Inevitabilism"
    - Gaslight-driven development
    - Don't Fall for AI: Reasons for Writers to Reject Slop
    - AI is killing the web – can anything save it?
        - from HN
            - Wiby - Search Engine for the Classic Web
            - I2P Anonymous Network
    - Rethinking CLI interfaces for AI
    - Local LLMs versus offline Wikipedia (on HN)
    - Terence Tao: A human metaphor for evaluating AI capability (on HN)
    - Will the Fear of Being Confused for AI Mean That We Will Now Write Differently? (on HN)
    - The vibe coder's career path is doomed (on HN)
    - AI coding agents are removing programming language barriers (on HN) (on lobste.rs)
    - Building better AI tools (on HN)
    - Stop Pretending LLMs Have Feelings Media's Dangerous AI Anthropomorphism Problem (on HN)
    - Developing our position on AI (on HN)
    - GPT might be an information virus (2023) (on HN)
    - #ai-policy
        - You Can’t Pay Me To Prompt!
            - an intereting writeup for (no) AI policy
            - from his AI Policy, I have quite some ideas about my own AI policy, starting by naming it "LM policy"
        - You can now disable all AI features in Zed (on HN)
        - Artificial Intelligence and the Linux Community
        - Generative AI. "Slop Generators, are unsuitable for use [ ]" (on HN)
- #jj
    - Git experts should try Jujutsu
    - Jujutsu For Busy Devs
    - zerowidth positive lookahead | jj tips and tricks
        - earlier: zerowidth positive lookahead | What I've learned from jj
    - rbtools/docs/rbtools/workflows/jujutsu.rst at master · reviewboard/rbtools
    - Jujutsu Strategies :: Reasonably Polymorphic
    - Jujutsu + Radicle = ❤️
    - dtolnay/dissimilar: Diff library with semantic cleanup, based on Google's diff-match-patch #struct
    - Sapling: A Scalable, User-Friendly Source Control System
        - many commands look like `jj`'s
    - The JJ VCS workshop: A zero-to-hero speedrun (on HN) (on lobste.rs)
- #incident
    - Recent Archive of Our Own outage was caused by integer exhaustion
    - Why AO3 Was Down (on HN)
    - GitHub pull requests were down (on HN)
    - Tell HN: 1.1.1.1 Appears to Be Down
    - Cloudflare 1.1.1.1 incident postmortem for July 14, 2025 (on HN) (on lobste.rs)
    - Cloudflare and the infinite sadness of migrations
    - Ask HN: GCP Outage?
    - How we discovered, and recovered from, Postgres corruption on the matrix.org homeserver
    - Starlink is currently experiencing a service outage. Our team is investigating (on HN)
    - Webflow Down for >31 Hours (on HN)
- #agent
    - bgcmd: let AI interact with any REPL
        - a nice example of Claude driving `rr` to debug
    - What I learned building an AI coding agent for a year
        - Codebuff
- #dev
    - What clever tools/workflows do you use to manage development environments?
        - mise
        - Devbox: Portable, Isolated Dev Environments
        - Fast, Declarative, Reproducible, and Composable Developer Environments - devenv (using Nix)
    - Why I got rid of all my neovim plugins
    - Stop using nix-env
- #idea
    - When in doubt, Go for a Walk
    - its ok to have a normal job
- #sec
    - iris
        - a neurosymbolic framework that combines LLMs with static analysis for security vulnerability detection
- #tech/fun
    - Useful utilities and toys over DNS
    - mawkdown, a lightweight, line-oriented(ish) text markup tool implemented in awk