Cryptography nerd

Fediverse accounts;
@Natanael@slrpnk.net (main)
@Natanael@infosec.pub
@Natanael@lemmy.zip

Lemmy moderation account: @TrustedThirdParty@infosec.pub - !crypto@infosec.pub

@Natanael_L@mastodon.social

Bluesky: natanael.bsky.social

  • 0 Posts
  • 421 Comments
Joined vor 1 Jahr
Cake day: 18. Januar 2025
  • Natanael@infosec.pubtoLinux Gaming@lemmy.world•BitCraft Open Sourcing - Phase 1 Announced!English
    5·
    vor 3 Tagen

    Real-time in computing usually either means a real-time OS with guaranteed low latency response (typically for stuff like microcontrollers regulating machines) or streaming live data (low latency delivery of the most recent value)

    This sounds like the latter, and a typical SQL database don’t guarantee real-time updates (you can have “atomic writes” to prevent inconsistency but usually this would make it slower) but some databases like this one are designed to ensure you can read out updated correct and consistent values much faster. Also with standard databases you usually make scheduled individual requests, but a real-time database could often send a stream of updated values to a “subscribing” program

  • Natanael@infosec.pubtoTechnology@lemmy.world•Judge orders Anna’s Archive to delete scraped data; no one thinks it will complyEnglish
    3·
    vor 4 Tagen

    It’s actually kinda easy. Neural networks are just weirder than usual logic gate circuits. You can program them just the same and insert explicit controlled logic and deterministic behavior. To somebody who don’t know the details of LLM training, they wouldn’t be able to tell much of a difference. It will be packaged as a bundle of node weights and work with the same interfaces and all.

    The reason that doesn’t work well if you try to insert strict logic into a traditional LLM despite the node properties being well known is because of how intricately interwoven and mutually dependent all the different parts of the network is (that’s why it’s a LARGE language model). You can’t just arbitrarily edit anything or insert more nodes or replace logic, you don’t know what you might break. It’s easier to place inserted logic outside of the LLM network and train the model to interact with it (“tool use”).

  • Natanael@infosec.pubtoWorld News@lemmy.world•BREAKING: Death toll from US operation in Venezuela has risen to 80, according to New York Times report.English
    32·
    vor 16 Tagen

    That argument works when the difference is small.

    That argument doesn’t work when one option is a cataclysmic disaster

    But most importantly - when you look at what policies that Trump voters said they voted for, you could divide that into a fraction of voters voting for evil (plain stupid racists), and a large fraction voting for something positive which they had been told Trump would deliver - yet which he was objectively worse at. Most people voted Trump for the economy while told he was a great businessman, or for healthcare while told he’d make health insurance more affordable (but now he made it less), etc…

    Almost every positive impact in the last decades that his voters attributed to him was delivered by his opposition.

    This wasn’t an election lost to attrition. Your quote explains nothing about what happened.

    There was more votes than ever. It was lost to propaganda and people being idiots, not seeing through the fraud. Trump’s policies lost every poll when names were taken off. Everything he wanted to do kept being rejected. But the propaganda machine made people distrust the people who delivered all the things they said they were grateful for, and to trust the liar instead.