• 0 Posts
  • 6 Comments
Joined 2 years ago
cake
Cake day: October 20th, 2023

help-circle
  • Was actually wondering about this the other night.

    Are there any pictures of the current state of it? I kind of assume they just half tore down the east wing and then wandered off.

    But, regardless: This was always a scam. Even if they build something, expect trump and his cronies to pocket the vast majority of the money that was supposed to be dedicated to it. Wouldn’t be shocked if it just collapsed the first time a hurricane reached DC.





  • Vote fixing is not that easy.

    Massive landslide victories will be far too obvious. I am not sure if there is a “Your honor, this is clearly bullshit” clause for recounts but it will trigger them.

    As for the closer stuff? Yes, electronic voting is deeply stupid and dangerous. But it is also worth understanding that many systems are still built around a paper ballot that you feed into it. And those will show massive discrepancies. ESPECIALLY in the “blue states” where these wins would most matter.

    For a presidential election? I could see a world where very targeted election fraud could MAYBE be pulled off, but I stand by: can you imagine the chuds and musk NOT bragging about outright fraud?

    But for midterms and anything down ballot? The places that matter most are the ones where republicans don’t control the state government in its entirety. And when the recount is like 20% different than the reported vote?

    Nah. Election fraud is the distraction. They’ll just perform, ANOTHER, violent insurrection and take the government by force.


  • People are deeply stupid.

    The Left will inevitably start getting pissy that their candidates didn’t win primaries and sit out the generals. And the conservatives will get brainwashed into thinking all the problems were a time bomb left by Barack HUSSEIN Obama.

    And… it is increasingly looking like there might be good odds that trump is dead by then and president vance (and now I threw up in my mouth a bit) is sacrificed as “undoing all the good trump did and why you need to vote for hegseth in 2028”


  • found that with just 250 carefully-crafted poison pills, they could compromise the output of any size LLM

    That is a very key point.

    if you know what you are doing? Yes, you can destroy a model. In large part because so many people are using unlabeled training data.

    As a bit of context/baby’s first model training:

    • Training on unlabeled data is effectively searching the data for patterns and, optimally, identifying what those patterns are. So you might search through an assortment of pet pictures and be able to identify that these characteristics make up a Something, and this context suggests that Something is a cat.
    • Labeling data is where you go in ahead of time to actually say “Picture 7125166 is a cat”. This is what used to be done with (this feels like it should be a racist term but might not be?) Mechanical Turks or even modern day captcha checks.

    Just the former is very susceptible to this kind of attack because… you are effectively labeling the training data without the trainers knowing. And it can be very rapidly defeated, once people know about it, by… just labeling that specific topic. So if your Is Hotdog? app is flagging a bunch of dicks? You can go in and flag maybe 10 dicks and 10 hot dogs and ten bratwurst and you’ll be good to go.

    All of which gets back to: The “good” LLMs? Those are the ones companies are paying for to use for very specific use cases and training data is very heavily labeled as part of that.

    For the cheap “build up word of mouth” LLMs? They don’t give a fuck and they are invariably going to be poisoned by misinformation. Just like humanity is. Hey, what can’t jet fuel melt again?