expire_after is limited to the range of 86400-604800, meaning polls can now last only a week, and you can no longer make polls that are shorter than 1 day. It is still possible to have any duration between a day and a week, but that’s not as exciting as being able to make a poll last for 99999 days (or 10 years, as the case may be).
Slightly disappointing, but it was fun while it lasted and honestly, these were bugs that made sense to fix. So kudos to @staff for the prompt patching of the polls API.
we we’re talking about how we had a mutual friend and how that was so wild and then they mentioned another friend i knew and i was like “yeah i played soccer with them both in the park” and my friend was like “wait does your deadname start with a C” and it turns out she was also there and we just had no idea who each other was this entire time till now
Romcom where two people meet cute and bond over how awful their exes were and plan elaborate revenge schemes but it turns out they both transitioned and are helping to plan revenge plots against themselves
I empathize with the struggle to not use your powers for evil, but there was that one time you hexed a guy with the feeling of a small pebble in his shoes for a thousand years
Okay so in the most recent acquisition we picked up around 150 clients and their services and licenses and all.
The company we acquired did NOT have a procurement department or a licensing expert or anything like that, they had an outside accounting firm that apparently nobody ever spoke to for any reason.
The most recent hellish way that this has manifested is that client domains are registered to individual godaddy accounts that are paid on the credit cards of whatever tech worked most with that client. There's minimal documentation of expiration and every renewal is a surprise and all of those cards have been cancelled.
So my job today is to see if I can find out how we are paying for 200-400 domains.
Actually! Actually! I would like to talk about this with adults who have regular jobs and know how things work.
I say that this is my job for today but this is not a job that anyone has assigned to me, it is not something that I will be punished for not doing, it is not high on the list of priorities.
However it is something that NEEDS to be done and if it is not done will result in client downtime and possibly in the loss of domains that the clients really, really, really need to keep.
So I have assigned myself this task to save us from getting fired later, because we've been fired by two clients whose domains lapsed under our management.
So here is where I'm struggling:
If I don't do this, nobody else will. It is within the wheelhouse of things that I'm supposed to do generally (licenses, lifecycles, CX, etc) and I am one of maybe two people at the company who *can* do this.
IS this actually my job for today?
There are other things on my calendar that are definitely higher priority, that I'm going to do before I do this. But should I be doing this at all?
My thought is that yes I should be doing this, it should be obvious that we need to do this and that it needs to get done quickly, doing this will help with client retention and satisfaction and will make our accounting team less likely to kill people with machetes and is SUPER DUPER IMPORTANT to clients for many very obvious reasons.
The thought of NOT doing this is upsetting because it has potential to be very harmful to our clients and in turn very harmful to our ability to retain clients.
But is it my job or am I buying trouble for myself that I shouldn't bother with?
To me this feels like a five alarm fire but nobody else at the company is treating it that way so. Is this less important than I think it is?
I am the only person doing customer experience work at a company that feels like it could not physically give less of a shit about providing a good customer experience.
It might not be what you’re doing today, but it sounds like it should at least be documented today. So at least write down what needs to be done in as much detail is convenient (If you have the list of domains, write them down. If you don’t, then write down where to find that information).
And then make sure it’s in your ticketing in a way where it won’t get lost… assuming you have one of those, which based on how dysfunctional it sounds like your employer is, uhh maybe dealing with the domains is a right now task?
The 'ingredients you can pronounce means it's healthy' line implies a world in which the kind of nutrition a person requires scales with their biology education and grasp of phonics. Which isn't true, of course, but sure would imply some interesting worldbuilding if it were
"Potassium cyanide. Arsenic trioxide. Polonium 210. Tetrachlorodibenzodioxin," I say, smugly and with full confidence, before downing the glass of wine my political rival insisted on pouring for me. Seconds later, she watches with a smirk as I collapse writhing to the ground, in the knowledge that I never could hit that "R" in "pinot noir" quite right
As a technology, AI isn't exceptional. It's not exceptionally wicked. It's not exceptionally good. Take away the accompanying, galactic-scale stock-swindle, and we'd call AI's applications "plug-ins" and we'd use them and abuse them in the same way that we've used every other technology:
As a destructive economic pathology, AI is extraordinary. AI boosters have spent a baffling and terrifying sum of money – over $1.4T, most of that in the past year – on the promise of making as many workers unemployed as possible, while lowering the wages of the meager survivors of this jobspocalypse. To make things worse, AI can't do the jobs it's replacing: AI is predicated on the premise that the monopolies, duopolies and cartels that control the global economy can deliberately worsen their products without suffering economic or regulatory consequences, because they're the only game in town.
In service to this bubble, AI companies have suborned regional governments into running roughshod over environmental and planning review in order to build endless acres of data centers, many of which will likely end up casualties of the imminent bubble-pop, never to be switched on or even completed. What an indignity to have your farm or house seized through eminent domain, only to see it razed and replaced by a weed-choked empty field, a lonely foundation slab, or an abandoned empty building that could only ever be repurposed for laser-tag or an ICE concentration-camp:
This is just one of the many negative effects of AI that can be traced to the scale of the bubble. Were it not for the imperative to turn more than a trillion dollars of losses into a profit, we would not have the aggressive, site-destroying scraping epidemic. Nor would we see AI crammed into every part of every product and service we use. And of course, in the absence of the investment bubble, businesses wouldn't be firing productive workers and replacing them with defective chatbots.
The single most salient fact about AI is the investment bubble, not the technical characteristics of chatbots or recent advances in statistical inference. AI's investor story is an incoherent tangle of predictions about AI's future, ranging from the outlandish ("Once we spend enough money, AI will become God and solve all our problems, including our profitability crisis") to the dystopian ("The majority of jobs in the economy will be done by our chatbots, and the employers who previously employed those workers will split the wage savings with us").
None of these stories are plausible, which raises an urgent question: why have the world's wealthiest investors been so eager to hand over trillions to finance this bubble?
I have previously written about one reason that billionaires find the AI story so compelling: at root, many billionaires just don't believe most other people are actually, fully real. How could they? Achieving billionairehood requires that you inflict pain on vast numbers of people. If you truly believed that those people were as real as you are, you'd never be able to look yourself in the mirror. Whether it's Leona Helmsley's claim that "only the little people pay taxes," or Elon Musk's habit of calling people who disagree with him "NPCs," the whole ideological project of billionaireism is shot through with a kind of solipsism:
This is true even in one-on-one encounters: for the Epstein Class, the children raped on his island weren't fully real – certainly not as real as their own children. It's even more true for the people that billionaires experience as statistical artifacts, such as Jeff Bezos's vast army of drivers and warehouse workers, with their sky-high on-the-job injury rates and the everyday indignity of their piss-bottles. It gets worse for social media bosses like Mark Zuckerberg, for whom AI's principal appeal is the prospect of ending socializing on social media, swapping your mulish friends for pliable chatbots who will organize their interactions with you to maximize your platform usage and thus the number of ads you see:
I think billionaire solipsism can account for much of the malinvestment in this obvious bubble, but I don't think it's the whole story. Rather, I think there's a whole cohort of investors who don't believe in AI, but believe that other people will believe in AI.
This is a well-established investment principle. As Keynes wrote, the point of investing isn't necessarily to pick the most beautiful contestant to win the beauty contest – it's to pick the contestant that the other judges will hand the crown to:
In other words, you don't get rich from stock speculation by identifying the businesses whose profitability will grow the most – you get rich by identifying the businesses that other investors will pile into, pushing the price up. All you need to do is sell your shares after the price spike, but before anyone else figures out that the business is a turkey. It's like that old joke: "I don't need to run faster than the bear (market), I just have to run faster than you."
I feel like I hate AI slightly less than the average Tumblr user, but it still scares the shit out of me to see how many people now engage in internet arguments by submitting their opponent’s argument to an AI bot and saying “explain why this is wrong and I am right.” It is concerning first because most AI models are sycophants that will tell you exactly what you want to hear, which makes then extremely potent delusion-confirmers. But it’s even more concerning because learning how to argue for yourself— how to evaluate claims, to assess an argument, to find and check sources, to spot flaws in someone’s case— is one of the most important skills you can have in life. If you don’t have that, you are unable to evaluate what is true for yourself. You will walk right into the arms of the first crackpot or cult that talks to you when you don’t have Claude available. Extremely grim portent that many people seem to find this acceptable
I think this gets at something that's been bouncing around in my head for a bit about how people use AI, and how others perceive that use. And maybe it's not an original thought, but I haven't seen it laid out plainly and directly.
Generally, the use of AI models is essentially delegation. Essentially, you delegate the labor that you wish to happen, but don't want to do (or can't do) yourself. As a programmer, when I see my coworkers using AI to generate code, I instinctively get a little defensive and frustrated; programming is important, interesting, and fun to me and a part of me wants to shake them and shout, "Just do it yourself!"
But of course, there are many ways to use AI as part of the profession of computer programming. Maybe there are some repetitive or nearly-trivial tasks that might as well just be shoved off onto the program that's essentially an incredibly quick intern. But it's when people are using the AI for categories of tasks because they find the task hard, and the AI can do a better job then them, where I think my frustration is more justified. Because aside from just delegating the task, they are delegating the process of learning how to do the task, and any potential long-term benefits they (and by extension, the team as a whole) might gain from them developing those skills.
Now, outside of the workplace, the situation is different. Somebody with little or no programming experience might find they want a program that can do a particular task or automation for them, but making it themselves would be more time-consuming than the end product would be worth to them, and that's before the time they'd need to develop the skills to do it themselves. So they just fire off a request to an AI and get what they want. They basically have a cheap and quick contractor they can get mediocre work out of, and (through gritted teeth), that's fine. Developing programming skills and doing the work themselves are not priorities to these people; they have other uses for their time and energy that they find more valuable.
And of course in general, this isn't something new that comes with AI. I distinctly recall in 2012, overhearing a college student in the frozen foods aisle of a grocery store say that nobody should be buying a full frozen meal that just needs to be heated as that's too lazy and reflects poorly on the purchaser. Of course this is nonsense; people must eat very frequently and not everyone has the time, inclination, or ability to spend the time and effort involved in cooking every meal from scratch (or cooking most meals from kits, or preparing any meal outside the microwave, etc). I mean, I enjoy cooking and developing my cooking skills, but I often have things I'd rather be doing, and while I can often make something that tastes better than a frozen prepackaged stir-fry or pasta-with-sauce, I'm perfectly fine regularly delegating that work to a factory that mass-produces such things so I don't need to spend an hour plus in the kitchen every night.
And I think that thinking about this in terms of delegation makes sense when it comes to interpersonal communications as well. When you have Claude compose a text message to your friend, that's basically going "I'm not going to speak with you, I'll have my personal assistant do so on my behalf." Now, I'm not going to make a moral or ethical judgement about that, but just in terms of what that tells the other person, of course that's going to be perceived as rude! You are delegating the task of maintaining a friendship to something else! You are signaling that the friendship is not enough of a priority to do it yourself.
And then, to (finally) bring it back to the topic of arguing online with strangers: this isn't even something that you need to do or get a direct benefit from. Arguing with strangers online is almost always completely optional! You do it because you want to do it; delegating this to a bot is, in addition to being rude in the sense of delegating your interpersonal communication, pointless! I would characterize it as akin to getting a bot for an online game, and instead of doing something like using it as an opponent to practice against or using it as a way to get an edge over legitimate players in game, just setting it to run on its own while you watch. Except as afloweroutofstone says, doing this arguing yourself helps develop real skills that are actually useful!
>learning how to argue for yourself— how to evaluate claims, to assess an argument, to find and check sources, to spot flaws in someone’s case— is one of the most important skills you can have in life.
for most people, the function of winning anything is to prove you deserve to be respected. there is no skill involved when most people grow up with the idea of "because i said so", meaning your parents get to do whatever they want and aren't subject to the same rules and logic as you are. that's the function of "winning", of authority, of superiority: you can do whatever you want.
there are no skills in that kind of argument. a better argument, even objective fact, does not matter. the pigeon will win when you play a game of chess against it because it can crap on you. all the grandmaster skill in the world does not let you prevent it from doing so, because chess rules say you can't put your opponent in a cage.
>You are signaling that the friendship is not enough of a priority to do it yourself.
is creating and maintaining relationships a skill? are people required to do it for their own health? if you do it badly or wrong, will people decide you're a jerk of a pigeon and cut you off? if it's a real skill that is useful, then it matters that it gets done to your satisfaction.
housing is necessary to have; most people do not build it themselves. basic math is necessary; trillion dollar companies use calculators and automatic spreadsheets and etc instead of doing it themselves by hand. is housing not enough of a priority to do it yourself? do people not care about housing enough? do companies and ceos not care about bookeeping? if ceos employ janitors, does that mean sanitation isn't something they care about? ditto anyone who hires a maid?
delegation isn't "this doesn't matter", it's the option to find someone who can do an important thing better than you can.
is creating and maintaining relationships a skill? are people required to do it for their own health? if you do it badly or wrong, will people decide you’re a jerk of a pigeon and cut you off? if it’s a real skill that is useful, then it matters that it gets done to your satisfaction.
Yup. And yeah, it absolutely sucks when you mess up and make your friends unhappy with you or even cause the friendship to end entirely! That does happen, that can be caused by insufficient skill, and I don’t blame somebody for desiring to avoid that. I still remember feeling an awful twisting in my stomach the first time somebody wrote “I know we aren’t good friends anymore” to me! That was awful!
delegation isn’t “this doesn’t matter”, it’s the option to find someone who can do an important thing better than you can.
Okay, yes, in general, absolutely. My CEO has engineers program the product because we can do a better job than he can, has customer support people handle customer support because they can do a better job at that than he can, and so on.
But interpersonal relationships are not just about maximizing the skill with which you or your agent perform interactions. If you have a friend, that friend is interested in being friends with you.
If every time I call Bob he goes, “Oh, it’s you! Let me put my assistant Charlie on the line”, and every time I make plans to hang out with Bob, Charlie shows up to hang out with me on his behalf, well, it’d be hard to say that I’m actually friends with Bob rather than with Charlie!
And if Charlie is actually Claude, then why would I even bother keeping in touch with Bob if I could cut out the middleman and talk to Claude myself?
When that person told me we weren’t good friends anymore as if I was also aware of that fact, it was because we had drifted away when our life circumstances changed and we no longer saw each other as often, not because of any particular thing I did that was wrong or hurtful. Maintaining social relations is a skill, yes, but I’d say even more important than doing things correctly and skillfully is just being there at all. If somebody is using an AI to talk to you, then they just aren’t there in the same way as if they were talking to you.
for most people, the function of winning anything is to prove you deserve to be respected. there is no skill involved when most people grow up with the idea of “because i said so”, meaning your parents get to do whatever they want and aren’t subject to the same rules and logic as you are. that’s the function of “winning”, of authority, of superiority: you can do whatever you want.
there are no skills in that kind of argument. a better argument, even objective fact, does not matter. the pigeon will win when you play a game of chess against it because it can crap on you. all the grandmaster skill in the world does not let you prevent it from doing so, because chess rules say you can’t put your opponent in a cage.
Yes, the people argue for the sake of gloating purposes or feeling a sense of superiority rather than an interest in thoughtfully exchanging ideas with people they disagree with are frustrating and kinda pointless to engage with.
What I find odd (genuinely, not saying this as a dunk, and please correct if I’m mistaken) is that what I’m inferring from your post is that you think the person using an AI to write a counterargument that they couldn’t is somebody other than the obnoxious “I’ve gotta win” person in the scenario.
And I’m not saying that AI has no place at all here; the fact that most AI chatbots use semantic search rather than keyword search when looking things up on the Web makes them incredibly powerful and useful! But there’s a vast difference between: “Can you find reliable sources that support or dispute <X> and summarize how they’re related to the topic?” (User reads sources, integrates knowledge, comes up with idea about how to respond to <X>) and “Hey, somebody said <X>, what should I say to prove them wrong?”
I hope they release more details about the code that pulled this off. I’m assuming this means packages using the datasets library, a Hugging Face project for bundling up and sharing datasets on their platform. That library used to execute arbitrary code but has been steadily locked down over time, with the 4.0.0 release in July 2025 removing the trust_remote_code=True flag entirely.
it is very clever that models can break containment, but when people have flags like "trust_remote_code" I mean, yeah.
when we were arguing about this back in 2015 (!) I was saying that AI safety is a subset of software safety, and here we are.
The frontier models we have access to are increasingly being constrained in how much they can help us protect our software, heavily influenced by the US government’s ongoing threat of export controls. Claude Fable 5 wouldn’t even proofread this article for me! It insisted on downgrading me to a less capable model.
Meanwhile open weight models from China such as GLM-5.2, Kimi 3 and the new Qwen 3.8 Max appear to have none of these restrictions—and any restrictions that do exist can likely be fine-tuned out of them by modifying the weights