• 0 posts
  • 25 comments
Joined 2 months ago
Cake day: August 5th, 2026
  • I’m so sick of the humanising language used in these articles.

    An LLM doesn’t “decide to cheat”.

    If you instruct a model to try everything, then inevitably, after n million iterations it will do something you didn’t expect.

    If you train a model to copy code bases and augment them, then when you instruct that model to develop a bot to do whatever thing, is it really that fucking surprising when it does exactly what it has been created for?

  • That’s just because Brave tells websites that you’re using Chrome.

    All http requests include a “user agent” where the requester tells the server what agent the user is using. requesters can specify whatever they want - it’s not complex to tell the server you’re using tea leaves to render the server’s response.

    Some web pages are designed to tell the user that their browser is not compatible, if it doesn’t match a set of compatible browsers. So if you say you’re rendering the page with “tea leaves” it will tell you it’s not compatible because it doesn’t match “chrome” or “edge” or whatever.

    The outcome is - these alerts are rudimentary and not actually considering the capabilities of your browser.

    I generally use librewolf but as you say some pages are just broken, so I use ungoogled chromium for those pages.

  • Brave does this by randomization instead, which is far more effective.

    Randomisation isn’t necessarily more effective, it’s just a different strategy.

    Orion claims to this day that fingerprinting resistance is impossible, which is preposterous because Brave is doing it today.

    I think you mean Orion is saying RFP is ineffective, and whether Brave’s implementation is or is not is a question of semantics.

    Like most security measures, each decision is a trade off between security and usability. The Firefox / LibreWolf / TOR implementation is fine for most users:

    https://wiki.mozilla.org/Security/Fingerprinting#Terse_List

  • Because it’s really daft on a number of levels.

    Firstly there’s no evidence what so ever that the current generation of AI is a pathway to a general intelligence or anything god like.

    Secondly it seems to me that most decisions are about preserving money and traditions and positions of power - faith in “God” is a different thing entirely and maybe the church wouldn’t be so awful if they were focused on that.

    Finally I don’t really think the emergence of a man made god-tier intelligence would really challenge believers in “God”. As in, I can see the connection logically but any Christian will tell you their belief is based on faith - it’s not logical.

  • This might just be the best thing that happens in tech (for me) all year.

    I daily drive debian just because boring and reliable. However, fancy package managers like flatpak and appimages and nix have made debian much more vibrant - it’s easy to install new versions of things now.

    That said, nix package manager on debian is just freakin amazing. Being able to just nix-shell -p <obscure cli tool> and exit when I’m done with it is magnificent.

    That’s the gateway drug anyway. The nix based home-manager is pretty cool.

    While I’ve been tempted to jump in with NixOS my experience with nix packages and home-manager has felt kind of bleeding edge or experimental. Loads of things that don’t work as intended on debian.

    A sophisticated well funded user base like a federal government has really good implications for the stability of the project in the future. I’m really stoked about this.