[–] 1 point 10 hours ago

If you're not coordinating between time zones, there's no problem. But then, if you're not coordinating between time zones, then there's no benefit to putting all the time zones on one.

Having coordinated issues between multiple time zones on UTC, I agree with the parent commenter. The overhead of stopping to ask "wait how does this fit with their normal work/sleep schedule or the volume of local activity" is actually more friction than simply asking "what time is it over there?" In my experience, coordinating everyone on a single time zone is worse than dealing with a time zone difference.

  • source
  • parent
  • context
  • [–] 12 points 22 hours ago

    I get the impression that first person controls of the first person character hanging drywall and placing one screw at a time to fix it to studs, is a bit more immersive than the Sims, but I've never played Sims. Does it seem fun to me? Well, no, but I also hate hanging drywall in real life too.

  • source
  • parent
  • context
  • [–] 7 points 2 days ago (1 child)

    AI, like all computer programs, is really good at being persistent.

    From the beginning of computer programming, the ability to engage in loops to exhaust every possibility was always going to be faster and more comprehensive than what humans could, or were willing to do. You could naively try every combination for cracking PINs. Then came more sophisticated loops that would try every combination by first going through known password patterns (common dictionary words, common numerical combinations, etc.). There were rainbow tables computed to where certain computationally expensive processes could be pre-computed before a leak so that a leak of hashed passwords could be more quickly exploited.

    And there are tools scanning the web for all sorts of known vulnerabilities, being much more comprehensive and systematic. If you have a vulnerable machine exposed to the internet, an automated hacker might compromise it.

    The current state of AI-assisted hacking includes tools for analyzing patches and CVEs to reverse engineer exploits for unpatched machines. So everyone is now in a hurry to apply patches as soon as they come out, lest they be left behind as hackers adapt their scripts to keep crawling for recently patched vulnerabilities.

    Plus there are more tools for searching for vulnerabilities, either in code or at some sort of other attack surface. The number of vulnerabilities being patched right now is pretty high, which puts just higher pressure on all of the processes to more aggressively develop quick patches for known vulnerabilities, install patches as soon as possible after release, etc.

    Throw in an ecosystem where programmers might have been laid off, or teams are understaffed, or using AI tools themselves and have less of a firm grasp on their own codebase, and that's just a recipe for less reliability.

  • source
  • parent
  • context
  • [–] 10 points 2 days ago

    LLMs overweight the style of this particular context of short explanatory posts (not quite long enough for a full blog post but generally longer than a Facebook status or a tweet) for public consumption. LinkedIn is where a lot of this type of post lives, and LLMs tend to borrow from those conventions and extend them outside of that context.

    But that stuff started there for a reason — em dash clauses are an efficient method to give a broad statement and then elaborate/explain in detail after, or by giving an example after.

    • Bullet points are an easy way to visually organize things into lists
    • Bullet points also are easy to textually encode in markdown, which is a friendly format for both training data and LLM output, because of the way LLMs work

    It's not just that the "it's not X, it's Y" sentence structure comes from LLMs; rather, it becomes some kind of logical negative space, that complements the other type of em dash explainer — the general then specific elaboration.

    Many of us learned to write in specific mechanical styles for corporate communications. Those rules aren't that complex, so it stands to reason that those rules end up getting incorporated into how machines generate text, too. But most of us have been doing it for a while, and oscillate between avoiding sounding like LLMs and defiantly using those conventions and saying stuff like why should I change, he's the one who sucks.

    No, I didn't quit using em dashes, bullets, or "it's not X" sentences. But I do use them more judiciously. And because of that, I recognize when they are still absolutely worth the overhead of people accusing the text of being LLM-written.

  • source
  • parent
  • context
  • [–] 5 points 3 days ago

    Don't forget that percentages still rely on a denominator. Normies stopped using laptops and desktops for normal computing, and all we have left are enterprise workstations, programmers, gamers, creatives, and tinkerers/enthusiasts. Some of the Linux percentage growth comes from an increase in users, but a lot of it comes from a decrease in the users of the other OSes.

  • source
  • [–] 10 points 5 days ago

    So many billionaires remember overcoming naysayers at some point in their journey and incorporate the wrong lessons learned into thinking they can dictate what people want, by sheer force of will, that they slowly learn to ignore signals of public preference. They think that the public is either too stupid to know what they want, or can be persuaded easily into changing their mind.

  • source
  • parent
  • context
  • [–] 20 points 1 week ago (3 children)

    That's the thing from the customer IT perspective. When your Microsoft-administered systems go down at the same time as the rest of the world, nobody blames the CTO or any of the sysadmins. The global outage becomes an acknowledgement that it's not their fault, so nobody in the company actually gets blamed, and nobody gives them a hard time.

    It might not result in more uptime, but it does mean that any downtime doesn't get pinned on them.

  • source
  • parent
  • context
  •  

    In recent years, I've seen people use projectors at night to protest policies: projecting slogans, images, etc., using a normal projector against the wall of the target of a protest (embassies, government buildings, prisons/detainment facilities, corporate headquarters, etc.).

    Here's an example, but I've seen lots of other examples. And even outside of protest, people have been doing ridiculous things with 3D projection, like this guy animating a hotel cuck chair (SFW).

    This got me to thinking: if the image recognition of Flock and similar cameras is just an AI image recognition function (using ImageNet or something like that) being run on captured video, what's stopping people from projecting false data onto streets and sidewalks in a way that tricks the Flock camera into recording false people/vehicles in places where there was no person/vehicle? Including fake license plates, etc.?

    Seems like an attack vector for some kind of denial of service attack, like putting fake faces and license plates all over.

    Is there anyone doing this? Would it be possible, at least at night, for certain camera locations and angles?

     

    I've read some of Ed Zitron's long posts on why the AI industry is a bubble that will never be profitable (and will bring down a lot of companies and investors), and one of the recurring themes is that the AI companies are trying to capture growing market share in an industry where their marginal profits are still negative, and that any increase in revenue necessarily increases their costs of providing their services.

    But some of the comments in various HackerNews threads are dismissive, saying that each new generation of models makes the cost of inference lower, so that with sufficient customer volume, the companies running the models can make enough profit on inference to make up for the staggering up-front capital expenditures it took to build out the data centers, train their models, etc.

    It's all pretty confusing to me. So for those of you who are familiar with the industry, I have several questions:

    1. Is the cost of running any given pretrained model going down, for specific models? Are there hardware and software improvements that make it cheaper to run those models, despite the model itself not changing?
    2. Is the cost of performing a particular task at a particular quality level going down, through releases of newer models of similar performance (i.e., a smaller model of the current generation performing similarly to a bigger model of the previous generation, such that the cost is now cheaper)?
    3. Is the cost of running the largest flagship frontier models going down for any given task? Or does running the cutting edge show-off tasks keep increasing in cost, but where the companies argue that the improvement in performance is worth the cost increase?

    I suspect that the reason why the discussion around this is so muddled online is because the answers are different depending on which of the 3 questions is meant by "is running an AI model getting cheaper over time?" And the data isn't easy to synthesize because each model has different token prices and different number of tokens per query.

    But I wanted to hear from people who are knowledgeable about these topics.

    view more: next ›