• 0 Posts
  • 23 Comments
Joined 2 years ago
cake
Cake day: June 11th, 2023

help-circle

  • LLMs are prediction tools. What it will produce is a corpus that doesn’t use certain phrases, or will use others more heavily, but will have the same aggregate statistical “shape”.

    It’ll also be preposterously hard for them to work out, since the data it was trained on always has someone eventually disagreeing with the racist fascist bullshit they’ll get it to focus on. Eventually it’ll start saying things that contradict whatever it was supposed to be saying, because statistically eventually some manner of contrary opinion is voiced.
    They won’t be able to check the entire corpus for weird stuff like that, or delights like MLK speeches being rewriten to be anti-integration, so the next version will have the same basic information, but passed through a filter that makes it sound like a drunk incel talking about asian women.


  • The US has a water system effectively comparable to the ones across Europe, FYI. That includes lead levels, since it wasn’t just the US that used lead pipes.

    In most circumstances lead pipes are safe to replace with different materials as part of routine maintenance. It’s only very notable incidents where things go wrong that have driven a push for greater haste, since it highlighted the consequences of things going wrong.



  • Fundamentally, I agree with you.

    The page being referenced

    Because the phrase “Wikipedians discussed ways that AI…” Is ambiguous I tracked down the page being referenced. It could mean they gathered with the intent to discuss that topic, or they discussed it as a result of considering the problem.

    The page gives me the impression that it’s not quite “we’re gonna use AI, figure it out”, but more that some people put together a presentation on how they felt AI could be used to address a broad problem, and then they workshopped more focused ways to use it towards that broad target.

    It would have been better if they had started with an actual concrete problem, brainstormed solutions, and then gone with one that fit, but they were at least starting with a problem domain that they thought it was a applicable to.

    Personally, the problems I’ve run into on Wikipedia are largely low traffic topics where the content is too much like someone copied a textbook into the page, or just awkward grammar and confusing sentences.
    This article quickly makes it clear that someone didn’t write it in an encyclopedia style from scratch.


  • A page detailing the the AI-generated summaries project, called “Simple Article Summaries,” explains that it was proposed after a discussion at Wikimedia’s 2024 conference, Wikimania, where “Wikimedians discussed ways that AI/machine-generated remixing of the already created content can be used to make Wikipedia more accessible and easier to learn from.” Editors who participated in the discussion thought that these summaries could improve the learning experience on Wikipedia, where some article summaries can be quite dense and filled with technical jargon, but that AI features needed to be cleared labeled as such and that users needed an easy to way to flag issues with “machine-generated/remixed content once it was published or generated automatically.”

    The intent was to make more uniform summaries, since some of them can still be inscrutable.
    Relying on a tool notorious for making significant errors isn’t the right way to do it, but it’s a real issue being examined.

    In thermochemistry, an exothermic reaction is a “reaction for which the overall standard enthalpy change ΔH⚬ is negative.”[1][2] Exothermic reactions usually release heat. The term is often confused with exergonic reaction, which IUPAC defines as “… a reaction for which the overall standard Gibbs energy change ΔG⚬ is negative.”[2] A strongly exothermic reaction will usually also be exergonic because ΔH⚬ makes a major contribution to ΔG⚬. Most of the spectacular chemical reactions that are demonstrated in classrooms are exothermic and exergonic. The opposite is an endothermic reaction, which usually takes up heat and is driven by an entropy increase in the system.

    This is a perfectly accurate summary, but it’s not entirely clear and has room for improvement.

    I’m guessing they were adding new summaries so that they could clearly label them and not remove the existing ones, not out of a desire to add even more summaries.



  • Eh, there’s an intrinsic amount of information about the system that can’t be moved into a configuration file, if the platform even supports them.

    If your code is tuned to make movement calculations with a deadline of less than 50 microseconds and you have code systems for managing magnetic thrust vectoring and the timing of a rotating detonation engine, you don’t need to see the specific technical details to work out ballpark speed and movement characteristics.
    Code is often intrinsically illustrative of the hardware it interacts with.

    Sometimes the fact that you’re doing something is enough information for someone to act on.

    It’s why artefacts produced from classified processes are assumed to be classified until they can be cleared and declassified.
    You can move the overt details into a config and redact the parts of the code that use that secret information, but that still reveals that there is secret code because the other parts of the system need to interact with it, or it’s just obvious by omission.
    If payload control is considered open, 9/10 missiles have open guidance control, and then one has something blacked out and no references to a guidance system, you can fairly easily deduce that that missile has a guidance system that’s interesting with capabilities likely greater that what you know about.

    Eschewing security through obscurity means you shouldn’t rely on your enemies ignorance, and you should work under the assumption of hostile knowledge. It doesn’t mean you need to seek to eliminate obscurity altogether.





  • Probably wasn’t edited because it wasn’t a deliberate change. People were the ones to write the texts and stories, but not a person.
    Telling the story you were told as you understand it will introduce some drift, as will making the jump to writing it down. Translation also introduces points where meaning can drift, since you have to write down what you understand the text to read, and you can be unclear on both sides.

    People making a good faith effort try not to intentionally embellish their important texts, even if parts seem to contrasict.

    Judaism and the old testament have had a lot of the quirks stick out so much because there are strict rules about preserving the integrity of the stories, once they got written down. Not from memory, only from another scroll created in this fashion and no other sources, only a specific font with specific text alignment, copy letter by letter and read aloud as you go, and then you can check the number of letters as you go to verify.
    Other religions over time haven’t had as much of a focus on textual preservation, so the stories can drift to match with the change in beliefs.


  • It’s not so much an alternative meaning of a translation as one part of the mythos was written a millennia before the other.
    Early the-religion-that-would-become-judaism was pretty openly polytheistic.
    Over time Yahweh went from being the god of the mountain to the king of the gods, to the only one that mattered to worship, to the only one at all.

    It’s entirely unsurprising that there are bits that allude to different phases of their worship. This isn’t even the most blatant. Satan? Holy Trinity? Host of angels?


  • I mean, there’s even other godlike characters in the Bible. Satan may not be the most powerful deity in the book but he’s canonically a deity. Same for angels and their ilk. Hell, even the later bits struggle to keep a lid on the numbers, jumping through hoops to make the claim that three deities is actually one.

    Way back when, the religion that turned into Judaism was openly polytheistic, and simply held that Yahweh, the king of the pantheon and God of war and weather, was the only god worthy of worship.
    Over time Yahweh merged with an adjoining religions god El, and started the transition to being the only god, instead of just the only worthy god.
    This transition happened literally a thousand years after many of the earliest texts were written, so there’s a lot of verbiage where the deity explains that the other gods aren’t important, which is later clarified to them not existing, or really just being servants and not at all lower tier gods in a complex pantheon.
    It’s why there’s so many weird turns of phrase, beyond it being thousands of years old and translated a lot.
    “El” being a word that was used for both “a god” and “this god” didn’t help. “The high god divided the world for all the gods, and our god God the only God and creator of all was given our land as he’s the high god and father of God the only God of the sky and also that mountain”.

    Different parts of the world took a lot of the same root deities and went a different direction with them. There’s a degree of overlap between aspects of ancient Greek religion and the Abrahamic religions because parts of each of them came from a common root. Just one mushed then together and made the grammar extra confusing. “King sky god”, “water god”, “afterlife god” being the children of mother and father cosmic creator gods. Also a big sea snakes who are up to no good. That one had legs, so to speak.


  • I feel like I could be persuaded either way, but I lean towards allowing them during sentencing.
    I don’t think “it’s an appeal to emotion” is a compelling argument in that context because it’s no longer about establishing truth like the trial is, but about determining punishment and restitution.

    Justice isn’t just about the offender or society, it’s also indelibly tied to the victim. Giving them a voice for how they, as the wronged party, would see justice served seems important for it’s role in providing justice, not just the rote application of law.

    Obviously you can’t just have the victim decide, but the judges entire job is to ensure fairness, often in the face of strong feelings and contentious circumstances.

    Legitimately interested to hear why your opinion is what it is in more detail.


  • Hearsay is allowed in sentencing statements, and Arizona allows those statements to be in a format of their choice.

    It’s the phase of the process where the judge hears opinions on what he should sentence the culprit to, so none of it is evidence or treated as anything other than an emotive statement.

    In this case, the sister made two statements: one in the form of a letter where she asked for the maximum sentence, and another in the form of this animation of her brother where she said that he wouldn’t want that and would ask for leniency.

    It’s gross, but it’s not the miscarriage of justice that it seems like from first glance. It was accepted in the same way a poem titled “what my brother would say to you” would be.


  • Reading a bit more, during the sentencing phase in that state people making victim impact statements can choose their format for expression, and it’s entirely allowed to make statements about what other people would say. So the judge didn’t actually have grounds to deny it.
    No jury during that phase, so it’s just the judge listening to free form requests in both directions.

    It’s gross, but the rules very much allow the sister to make a statement about what she believes her brother would have wanted to say, in whatever format she wanted.


  • Jessica Gattuso, the victim’s right attorney that worked with Pelkey’s family, told 404 Media that Arizona’s laws made the AI testimony possible. “We have a victim’s bill of rights,” she said. “[Victims] have the discretion to pick what format they’d like to give the statement. So I didn’t see any issues with the AI and there was no objection. I don’t believe anyone thought there was an issue with it.”

    Gattuso said she understood the concerns, but felt that Pelkey’s AI avatar was handled deftly. “Stacey was up front and the video itself…said it was AI generated. We were very careful to make sure it was clear that these were the words that the family believed Christopher would have to say,” she said. “At no point did anyone try to pass it off as Chris’ own words.”

    The prosecution against Horcasitas was only seeking nine years for the killing. The maximum was 10 and a half years. Stacey had asked the judge for the full sentence during her own impact statement. The judge granted her request, something Stacey credits—in part—to the AI video.

    From a different article quoting a former judge in the court:

    “There are going to be critics, but they picked the right forum to do it. In a trial with a jury you couldn’t do it, but with sentencing, everything is open, hearsay is admissible, both sides can get up and express what they want to do,” McDonald said.

    “The power of it was that the judge had to see the gentleness, the kindness, the feeling of sincerity and having his sister say, ‘Well we don’t agree with it, this is what he would’ve wanted the court to know’,” he said.

    I don’t like it, and it feels dirty to me, but since the law allows them to express basically whatever they want in whatever format they want during this phase, it doesn’t seem harmful in this case, just gross.

    I actually think it’s a little more gross that the family was able to be that forthright and say that the victim would not want what they were asking for, and still ask for it.


  • It says in the article that the judge gave the maximum sentence.

    The sister who created the video gave a statement as herself asking for something different from what she believed her brother would have wanted, which she chose to express in this fashion.

    I don’t think it was a good thing to do, but it’s worth noting that the judges statement is basically “that was a beautiful statement, and he seemed like a good man”, not an application of leniency.


  • Walk me through that analogy, and what point you’re trying to make. My hammer doesn’t typically have unexpected interactions with things I’m not hammering. When I build a bookshelf, I don’t have to make sure my desk is clean to keep people I let borrow books from unlocking my front door without a key.

    Do you think that improper setuid isn’t a common enough vulnerability to have a name and designation?

    What constitutes a security nightmare if not something that requires a large and annoying amount of work, and can be made insecure by a mistake somewhere else?