…cogito, ergo sum…

  • 10 Posts
  • 56 Comments
Joined 5 months ago
cake
Cake day: 3 December 2025

help-circle



  • I am sorry, but what is the actual reason of this blog post? There’s nothing actually valuable to learn from, and even Gist files with unsafe code shared.

    It looks more like an ad for Rode and Claude Code, not to mention the bragging aloud about “understanding” staff you have no idea about still without your LLM mentioned. You just did steps Claude told you to do, I believe, and there’s nothing mentioned actually valuable, including the underlying HID data.

    Do people really clog up their blogs nowadays with how “they did” LLM-supervised steps and “learned” something clicking buttons?

    The ending as “thanks computer, until next time” felt like a spit to the face, too, where I, as a human, do regret reading it now in the first place.







  • I am sorry, but what a… trolling and clickbait…

    At the exact same time, coding agents helped me iterate quickly and ship software that worked well (after some dutiful testing, of course). They were also, I found, excellent tutors

    In the day to day of shipping agents into production

    Recurse Center (RC) is a self-directed, full-time programming…

    Coming into RC, my goals were the following:
    - Train an LLM from scratch. This includes pre- and post-training, and I want to do this mostly from scratch; not just fork a premade codebase but write a Transformer myself.

    For example, a nice thing I picked up from someone I pair programmed with: when this guy was writing code and didn’t quite remember the syntax or operations, he would often just quickly open up a terminal and type a super simple example to rapidly iterate. He was usually able to work it out and verify if it worked correctly in less than a minute, and he didn’t have to google anything and comb through search results or ask an LLM.
    This technique might seem obvious to some, but making this process muscle memory has helped me become unstuck much faster.

    Soon I’ll be shipping agents to prod and running evals with a whole new bag of tricks and skills. But for now I’ve got 6 more weeks left at RC

    Source

    I do regret opening it, reading it, and increasing their visiting counter…


  • Thank you! Hurray! A yet another brand new credentials “local-first” cloud with very transparent and intuitive brand name, that is going to be “responsible” for “decades” (hopefully years) of storing and maintaining someone’s credentials with undisclosed infrastructure, legal terms in cases of the credentials leaked during a bug (e.g. at E2EE/secure channel session), and no actual comparison stated between the current competitors, including DotEnv cloud service, BitWarden, 1Password, KeePass-based etc.!

    No, sorry, I, nor anyone I know, would trust credentials to any organization with so little transparency and lack of guarantees, also considering audited alternatives.

    Oh, and indeed, where is the key card? Is it in an ASCII art somewhere in documentation?

    The design is nice, however, but… may I ask, how much non-AI work was done, if any?



  • The macOS version can make stronger guarantees because it can have more complexity. On Linux, the foundation is eBPF, which is powerful but bounded: it has strict limits on storage size and program complexity. Under heavy traffic, cache tables can overflow, which makes it impossible to reliably tie every network packet to a process or a DNS name.
    And reconstructing which hostname was originally looked up for a given IP address requires heuristics rather than certainty. The macOS version uses deep packet inspection to do this more reliably.
    That’s not an option here.

    Source [web-archive]

    The above feels like an utter AI slop nonsense, sorry. I believe eBPF, the Linux Kernel feature, is absolutely capable for accuracy and perfect processing of network traffic. Have you ever checked Calico or Cilium, or at least, Oryx?




  • Thank you very much! The article is a great reference to consider and start creating workspaces for teams ready to deal with panic and stay efficient in critical events.

    As a reminder, one of the ten levers you can pull to increase psychological safety in your subculture is to provide structure for difficult conversations.

    Unstructured hard conversations rely on confidence and power. Structured ones rely on shared tools and expectations.

    In the Anxiety Zone, difficult conversations feel personal and unpredictable.

    Source [web-archive]