jakke @jakke - Tumblr Blog

The exploit shows the extreme risk of offloading technical support to AI.

So on one hand I trust this reporting but on the other hand I'm really not sure what to make of stories like this. What's the chain of decisions that leads to involving a LLM-powered chatbot anywhere in the password reset process, which was already a simple well-understood workflow that looks the same on every website? I know that everyone says "oh they're doing this because it delights the shareholders" but you can incorporate the LLM in half a dozen other features and put out basically the same press release and the investor response would be identical.

From the investor calls I've listened to (and I highly recommend that everyone listen to Big Tech investor calls) a big part of what's going on is that the average tech executive sees the future editorial voice of their platform as a helpful friend who's scrolling through all these pictures and videos with you and is recommending you buy things along the way. Like, the idea is that you and a chatty pal are browsing the Internet together, and that chatty pal is actually a puppet with an advertisement algorithm inside of it, but as much as possible you should feel like it's someone you're chatting with as opposed to the Web 2.0 ideal of social media as a platform that lets people post things and gives other people tools to share or react to those posts, and so you should basically never see buttons and should always be reading (or hearing!) a friendly voice whenever you're in the app. And I'm very skeptical that users will ever learn to love this new mode of interaction, because it adds yet another layer of mediation on the already heavily mediated social media interactions, but I'm certain that it's absolutely terrible for security in a way that can't be mitigated.

jakke

The real long-run ideal for social media advertising, incidentally, is that you just point at a thing in a picture you see (could be an ad, could be LLM generated, could be an "organic post") and say out loud "I want that one" and then an LLM recognizes what you mean, buys the item on your credit card, and ships it to your home. I have literally heard a Pinterest executive describe this exact process as their goal, because apparently Pinterest still exists. So in this long-run goal the LLM would also have access to your credit card number and home address, which (as this most recent breach demonstrates) is probably kind of a security risk.

jakke

The more I think about this the less I understand how this could ever work with realistic token pricing. Like you either need to make it basically free for a user to say "who made that shirt" a thousand times in a row, or you need to make sure that the user is buying several of those shirts to pay for the tokens. I think the narrative is that the second one will happen, but in general if it's costing the platform upwards of a dollar per user shopping search, that's going to be really hard to recoup.

jakke

The exploit shows the extreme risk of offloading technical support to AI.

So on one hand I trust this reporting but on the other hand I'm really not sure what to make of stories like this. What's the chain of decisions that leads to involving a LLM-powered chatbot anywhere in the password reset process, which was already a simple well-understood workflow that looks the same on every website? I know that everyone says "oh they're doing this because it delights the shareholders" but you can incorporate the LLM in half a dozen other features and put out basically the same press release and the investor response would be identical.

From the investor calls I've listened to (and I highly recommend that everyone listen to Big Tech investor calls) a big part of what's going on is that the average tech executive sees the future editorial voice of their platform as a helpful friend who's scrolling through all these pictures and videos with you and is recommending you buy things along the way. Like, the idea is that you and a chatty pal are browsing the Internet together, and that chatty pal is actually a puppet with an advertisement algorithm inside of it, but as much as possible you should feel like it's someone you're chatting with as opposed to the Web 2.0 ideal of social media as a platform that lets people post things and gives other people tools to share or react to those posts, and so you should basically never see buttons and should always be reading (or hearing!) a friendly voice whenever you're in the app. And I'm very skeptical that users will ever learn to love this new mode of interaction, because it adds yet another layer of mediation on the already heavily mediated social media interactions, but I'm certain that it's absolutely terrible for security in a way that can't be mitigated.

jakke

The real long-run ideal for social media advertising, incidentally, is that you just point at a thing in a picture you see (could be an ad, could be LLM generated, could be an "organic post") and say out loud "I want that one" and then an LLM recognizes what you mean, buys the item on your credit card, and ships it to your home. I have literally heard a Pinterest executive describe this exact process as their goal, because apparently Pinterest still exists. So in this long-run goal the LLM would also have access to your credit card number and home address, which (as this most recent breach demonstrates) is probably kind of a security risk.

The exploit shows the extreme risk of offloading technical support to AI.

So on one hand I trust this reporting but on the other hand I'm really not sure what to make of stories like this. What's the chain of decisions that leads to involving a LLM-powered chatbot anywhere in the password reset process, which was already a simple well-understood workflow that looks the same on every website? I know that everyone says "oh they're doing this because it delights the shareholders" but you can incorporate the LLM in half a dozen other features and put out basically the same press release and the investor response would be identical.

From the investor calls I've listened to (and I highly recommend that everyone listen to Big Tech investor calls) a big part of what's going on is that the average tech executive sees the future editorial voice of their platform as a helpful friend who's scrolling through all these pictures and videos with you and is recommending you buy things along the way. Like, the idea is that you and a chatty pal are browsing the Internet together, and that chatty pal is actually a puppet with an advertisement algorithm inside of it, but as much as possible you should feel like it's someone you're chatting with as opposed to the Web 2.0 ideal of social media as a platform that lets people post things and gives other people tools to share or react to those posts, and so you should basically never see buttons and should always be reading (or hearing!) a friendly voice whenever you're in the app. And I'm very skeptical that users will ever learn to love this new mode of interaction, because it adds yet another layer of mediation on the already heavily mediated social media interactions, but I'm certain that it's absolutely terrible for security in a way that can't be mitigated.

slept 7.5 hours each of the last two nights

Math and programming easier, concentration uninterrupted, zero meaningful lessons learned about the impact of sleep on overall quality of life. I've just unintentionally overloaded the past week with DMing D&D games and performing in improv shows and hosting friends from out of town and organizing picnics and going to music festivals in the park and having long niche kink sessions until I'm fully just losing consciousness without meaning to at the end of the day, because that way I'm reliably going to sleep at 12:30 or 1 and then waking up at 8. This past week has been so full of time with friends and progress on projects and general feelings of life fulfillment, but at the end of the week I'm going to visit my parents and experiencing three days of their heartbreak with how I've turned out, and that's going to tank my mood for the next few weeks, so currently feeling like I'm flying towards a brick wall.

(Bloomberg) -- Walmart Inc. is limiting staff’s usage of an artificial-intelligence tool following high demand, illustrating how corporation

Very relevant to that thread with @brostateexam yesterday. Walmart was really making a big deal of mandating heavy AI usage at their corporate headquarters for tasks like spreadsheet-building, but now that usage is not quite so subsidized by Big Tech shareholders, costs ar ticking up, and now Walmart is reconsidering. This gives rise to a three-part ruleset that feels like a heavy-handed office sitcom plotline:

You need to use the LLM tools, or else you'll get a bad performance review for not innovating at full velocity.

You cannot use too many tokens on the LLM tool, or else you'll get kicked off for the rest of the month and none of your workflow will be feasible anymore.

No one can tell you what a token is.

This is obviously not a long-term stable state of affairs. But at this point it's pretty clear that if there's any employee who's expensive enough to the company to justify paying the full price for LLM usage, it's only going to be the very highest-paid software engineers for whom speedier coding is legitimately worth thousands of dollars a month. (And executives! Tech executives are sufficiently coddled that they get access to every toy regardless of the cost. That's practically a defining attribute of a tech company.)

Are enough employees on the upper side of that wall, over the entire global economy, to make the current LLM-as-a-service industry long-run profitable? Absolutely not. Too many companies are competing to build the same thing, the capital costs are very high, and the utility is limited to a thin sliver of very specialized applications. Is there any scale at which this industry does make sense, providing specialized coding tools to a subset of engineers plus a misinformation toy for executives? Maybe, but it's impossible to see how that justifies orienting the entire tech industry and a big chunk of financial markets around this one service.

#the wonders of capitalism

Some protesters held signs condemning Israel saying "We would rather die as Jews than live as Zionists" and "We refuse to serve an army for the sake of the Zionist religion."

Yeah even the ultra-Orthodox living in a complete bubble can see what's going on better than centrist Israeli politicians. I realize that what's happened is that October 7 completely warped Israeli politics in exactly the same way that September 11 did US politics, and everything since then has followed the same playbook of enforced patriotism and revenge destruction of every city in Afghanistan and then a completely unrelated war in Iraq that killed even more civilians, but two and a half years in any anti-war political movements are still relegated to the activist fringe. At any rate, the next elections in Israel are unlikely to produce a functional government largely due to these exact Haredi protesters and their demands, which isn't going to make it any easier to stop the military killing machine.

jakke

The golden age of Microsoft's GitHub Copilot appears to be at an end.

So this article is thin gruel, and I'm sorry tech journalism is dead now, but basically TLDR these code generation LLM tools that used to be all-you-can-eat now cost money to use intensively and that turns out to be incredibly expensive for many use cases. And this was unavoidable, right? Just running these systems costs money, and spending hundreds of billions on infrastructure to build these systems is incurring massive capital costs, and unlike e.g. search or social media there's no good way to serve ads in a code generation tool - so the era of subsidy and experimentation is over, and the era of spending hundreds or thousands of dollars a month to keep using Copliot at the same velocity is upon us. This is the same trajectory that e.g. Uber or AirBNB went through, but the big difference is that those services had very low capital costs for the company serving them (they didn't have to pay to build the cars or the houses) but LLMs obviously have a higher capital cost than anything since the railroad boom.

So what does this mean for the future of LLMs? I think the actual serious adults who are setting out to make thoughtful profit-maximizing society-eroding globe-heating decisions have correctly seized on code generation as the one thing that LLMs can conceivably replace some amount of human labour at, because junior software engineers have a job that requires very little external context and all the information is available in text format and dozens of tools (linters, commit hooks, compilers) catch errors and many of them are incredibly low productivity and their pay is extraordinarily high - so it's reasonable to say that a company that hires a thousand of them a year could reasonably hire slightly fewer, and might even be willing to pay $50k or more per replaced low-quality entry-level engineer per year. But where does that leave the whole ~*~generative AI industry~*~ if that's the use case?

I think there's room for at least one or two reliably profitable companies somewhere in there, and those companies will become industry standards, and universities will scramble to reorient their coursework so that graduates have the correct proficiencies in LLM interaction for the labour market. All of this will consume only a tiny fraction of the planned data centre buildout, and most of that capacity will get cancelled and never built, and a bunch of stocks with massively overinflated values are going to crash, and that's probably going to trigger a stock market correction. Most of the losses of this adjustment from "AI will do everything and we will live as immortal trillionaire god-kings" to "LLMs are a useful tool for automating menial software engineering tasks" are ultimately going to be borne by giant tech companies though, and the tech giants could each endure a $100 billion writeoff and be basically fine, so I think the impact on the overall economy will be minimal.

jakke

@brostateexam The identity of this friend is going to remain a mystery, but anyone who owns stock or mutual funds or has a pension or anything is exposed to a highly inflated company right now. I think this next couple weeks is really going to be informative in terms of understanding the willingness to pay the higher price for these tokens. Ultimately it's like two dozen companies that are really hiring software engineers at scale and setting the industry standards, right?

So if their CEOs see this and say "oh okay yeah this looks reasonable" then the current status quo of high expectations could keep on going multiple additional quarters because there's still that really low probability of really massive labour reduction that makes it possible to rationalize these really high valuations. Conversely if those same CEOs see this and say "oh okay wow we can't justify spending $1000 per employee per month on a coding tool" and start rationing out licenses or exhorting employees to keep using LLMs but in less costly ways or ditching Copliot for other services that are still in that magical subsidy phase, I think that means the end is high for this bubble. Fortunately, CEOs in 2026 love showing up on video podcasts to talk about this kind of thing for eight hours straight, so we'll find out soon enough.

The golden age of Microsoft's GitHub Copilot appears to be at an end.

So this article is thin gruel, and I'm sorry tech journalism is dead now, but basically TLDR these code generation LLM tools that used to be all-you-can-eat now cost money to use intensively and that turns out to be incredibly expensive for many use cases. And this was unavoidable, right? Just running these systems costs money, and spending hundreds of billions on infrastructure to build these systems is incurring massive capital costs, and unlike e.g. search or social media there's no good way to serve ads in a code generation tool - so the era of subsidy and experimentation is over, and the era of spending hundreds or thousands of dollars a month to keep using Copliot at the same velocity is upon us. This is the same trajectory that e.g. Uber or AirBNB went through, but the big difference is that those services had very low capital costs for the company serving them (they didn't have to pay to build the cars or the houses) but LLMs obviously have a higher capital cost than anything since the railroad boom.

So what does this mean for the future of LLMs? I think the actual serious adults who are setting out to make thoughtful profit-maximizing society-eroding globe-heating decisions have correctly seized on code generation as the one thing that LLMs can conceivably replace some amount of human labour at, because junior software engineers have a job that requires very little external context and all the information is available in text format and dozens of tools (linters, commit hooks, compilers) catch errors and many of them are incredibly low productivity and their pay is extraordinarily high - so it's reasonable to say that a company that hires a thousand of them a year could reasonably hire slightly fewer, and might even be willing to pay $50k or more per replaced low-quality entry-level engineer per year. But where does that leave the whole ~*~generative AI industry~*~ if that's the use case?

I think there's room for at least one or two reliably profitable companies somewhere in there, and those companies will become industry standards, and universities will scramble to reorient their coursework so that graduates have the correct proficiencies in LLM interaction for the labour market. All of this will consume only a tiny fraction of the planned data centre buildout, and most of that capacity will get cancelled and never built, and a bunch of stocks with massively overinflated values are going to crash, and that's probably going to trigger a stock market correction. Most of the losses of this adjustment from "AI will do everything and we will live as immortal trillionaire god-kings" to "LLMs are a useful tool for automating menial software engineering tasks" are ultimately going to be borne by giant tech companies though, and the tech giants could each endure a $100 billion writeoff and be basically fine, so I think the impact on the overall economy will be minimal.

#the wonders of capitalism

Anti-establishment candidate Graham Platner seemingly came out of nowhere to become the presumptive Democratic nominee for the Senate seat i

So glad the Dems cleared the field for a rich white guy with a vaguely counterculture veneer and absolutely zero past scrutiny for any of his decisions. Maine has plenty of competent Democrats with more charisma and more interesting backstories and more previous time in the media spotlight (the state house Speaker is charismatic and good at talking about policy! the mayor of Lewiston has been really vocal about ICE in a convincing way! I have never been to Maine and know very little about it but there is probably another example who isn't quite as visible in the national media!) but the party brass made the decision they were going with the prep school graduate former military contractor who bought a seafood business to re-establish a presence in Maine, and they all played the game correctly and bowed out, and now it turns out that Nazi tattoo guy is also cringey extramarital sexting guy, and once again it looks pretty much impossible for the Democrats to take back the Senate. All of this could have come out in the primaries, but the party machine is so afraid of primaries that they want to carefully handpick successors, and it's so hard for people living in a wealthy well-connected DC bubble not to pick wealthy well-connected people who spent many years living in DC as the natural safe choice, and this guy is turning out to be such a shitty candidate he might as well have been grown in a lab to get Collins re-elected.

#abortions for some miniature American flags for others

jakke

queer-for-queer remains a very funny very Seattle identity

Like at its core the idea is just "I'm a cis person looking for a relationship with a cis person on the other side of the gender binary, and I'd like to get married and have children with them on a short timescale, but I don't want all the baggage that comes from heterosexuality, and I would like both of us to have a lightly counterculture and possibly also androgynous vibe". And I don't think there's anything wrong with any of that, and it makes sense to develop terminology and descriptors for people who feel this way, because there are a lot of them. But unfortunately none of our identities or goals or values are still embedded in a broader social context, and so you end up having to nod very understandingly through conversations about how the AFAB partner can't out-earn the AMAB partner in a relationship or else their parents won't approve lest you be accused of being a bad gatekeeping close-minded homosexual.

jakke

@threeninesfine So I think in 2026 ~*~identifying as queer~*~ doesn't really have much to do with sexuality or gender anymore to a lot of people, and is more about a specific aesthetic and political affiliation. And I realize that's kind of grating to those of us from a generation where the term did have a lot to do with sexuality and gender, but language and usage evolves, and now people identify things like tattoos and astrology and using terminology from drag culture as queer, because the kind of people we used to identify as "queer" (specifically, the most marginalized members of that former community) adopted these things, and then counterculture saw marginalized people adopting things and adopted them too so they could present themselves as more transgressive (which is a pretty standard cycle of appropriation since at least the Elvis era), and those activities and practices and consumption decisions retained the "queer" label even as they spread through broader mainstream culture, and now those activities and practices and consumption decisions signify "queer", rather than anything related to sexuality or gender. And so I don't begrudge anyone the "queer for queer" label, because they really are just using the word in the sense it's commonly used in these days, but it's still very funny when all of this gets wrapped up in very traditional patriarchal norms like who earns more money or who needs to be taller or who needs to drive when you're in a car together.

jakke

@brostateexam Yeah of course I get what you mean! Feels really discordant to have a term that encapsulated a specific set of communities become a much broader and vaguer term that abstracts away from any marginalization. But saying "no that word needs to keep meaning what it used to mean! not what it's now popularly understood to mean!" isn't realistic, you know?

(That being said, I still push back on the increasingly common descriptor "gay but not queer", because that usually gets thrown around to conflate "they are fussy and dowdy and unstylish" with "we are suspicious that they have bad politics", and that seems thoughtless and dangerous. Gays should be allowed to be just as frumpy and swagless as anyone else without falling into a suspect class.)

jakke

queer-for-queer remains a very funny very Seattle identity

Like at its core the idea is just "I'm a cis person looking for a relationship with a cis person on the other side of the gender binary, and I'd like to get married and have children with them on a short timescale, but I don't want all the baggage that comes from heterosexuality, and I would like both of us to have a lightly counterculture and possibly also androgynous vibe". And I don't think there's anything wrong with any of that, and it makes sense to develop terminology and descriptors for people who feel this way, because there are a lot of them. But unfortunately none of our identities or goals or values are still embedded in a broader social context, and so you end up having to nod very understandingly through conversations about how the AFAB partner can't out-earn the AMAB partner in a relationship or else their parents won't approve lest you be accused of being a bad gatekeeping close-minded homosexual.

jakke

@threeninesfine So I think in 2026 ~*~identifying as queer~*~ doesn't really have much to do with sexuality or gender anymore to a lot of people, and is more about a specific aesthetic and political affiliation. And I realize that's kind of grating to those of us from a generation where the term did have a lot to do with sexuality and gender, but language and usage evolves, and now people identify things like tattoos and astrology and using terminology from drag culture as queer, because the kind of people we used to identify as "queer" (specifically, the most marginalized members of that former community) adopted these things, and then counterculture saw marginalized people adopting things and adopted them too so they could present themselves as more transgressive (which is a pretty standard cycle of appropriation since at least the Elvis era), and those activities and practices and consumption decisions retained the "queer" label even as they spread through broader mainstream culture, and now those activities and practices and consumption decisions signify "queer", rather than anything related to sexuality or gender. And so I don't begrudge anyone the "queer for queer" label, because they really are just using the word in the sense it's commonly used in these days, but it's still very funny when all of this gets wrapped up in very traditional patriarchal norms like who earns more money or who needs to be taller or who needs to drive when you're in a car together.

This report is solid, although of course it's kind of dated given the data it depends on. There's a lot of writing about the K-shaped economy, but it tends to focus on how the consumption from the top tier keeps the whole system afloat, and not on the bottom half. This sentence in particular seems so bleak to me:

Nearly 38 million households could make ends meet if wages increased by $10 an hour, plus an additional 10 million households if costs decreased by $500 a month.

Like, even if wages increased by $10/hour, most of the households that currently can't make ends meet still wouldn't be able to make ends meet. And a $10/hour raise is already unrealistically large for a lot of people. We're absolutely not going to be able to market economy our way out of this one, you know?

The subsequent parts of the report are going to get more into the cost drivers and how much of this is housing or childcare or transportation or what. And I think that's a really helpful antidote to the dead-end liberal progressive fantasy of "we will just send everyone $1000 a month and the market will handle the rest" because that would drive up costs, it would not deliver assistance to the people who need it most, and it would unavoidably take away funding from public goods that really only the government can provide. Some of the solution needs to come from government subsidy or direct provision on the supply side, and it's worth understanding where this would be the most helpful.

#the wonders of capitalism

queer-for-queer remains a very funny very Seattle identity

Like at its core the idea is just "I'm a cis person looking for a relationship with a cis person on the other side of the gender binary, and I'd like to get married and have children with them on a short timescale, but I don't want all the baggage that comes from heterosexuality, and I would like both of us to have a lightly counterculture and possibly also androgynous vibe". And I don't think there's anything wrong with any of that, and it makes sense to develop terminology and descriptors for people who feel this way, because there are a lot of them. But unfortunately none of our identities or goals or values are still embedded in a broader social context, and so you end up having to nod very understandingly through conversations about how the AFAB partner can't out-earn the AMAB partner in a relationship or else their parents won't approve lest you be accused of being a bad gatekeeping close-minded homosexual.

Rather than overhaul the system immediately, the city will gradually update valuations – and the tax – according to the budget documents. Starting in the 2028-2029 tax year, the property values will be based on comparable sales.

The Mamdani administration is trying to levy an additional property tax on second homes worth $1 million or more. Buried in the middle of the story is this pair of sentences, which is a huge shocking departure from the ways that New York City currently deliberately undervalues all its property assessments, which has caused a massive unsustainable budget hole. And I can't find any information about this anywhere else!

If this update is happening for every property in New York, that's going to be great news for the city budget and a pretty sizeable new tax bill for a lot of very rich people whose properties have been undervalued by millions of dollars for many years now. If this update is happening just for these second homes, that's a very funny split with absolutely zero justification. I can't find any other news online about this and I can't spend more than five minutes digging through the state budget to try to find it today, but I'm really curious as to what's happening here.

#the wonders of capitalism

Mélenchon was also the most popular left-wing politician among left-wing voters, ahead of rivals such as former President François Hollande and MEP Raphaël Glucksmann. Some 49 percent of left-wing voters supported Mélenchon, compared to 43 percent for Hollande and 36 percent for Glucksmann, the poll said. Mélenchon was also the most popular left-wing politician among left-wing voters, ahead of rivals such as former President François Hollande and MEP Raphaël Glucksmann. Some 49 percent of left-wing voters supported Mélenchon, compared to 43 percent for Hollande and 36 percent for Glucksmann, the poll said.

France is cruising towards a far-right government and I guess in 2026 that shouldn't even feel surprising anymore but what's wild is that left-wing voters prefer Mélenchon over all the alternatives. He's been in elected office for four decades with no significant achievements, he's lost badly in three previous presidential runs, he loves Putin and hates Zelensky, he's indulged in wild conspiracy theories and old-school bigotry, and he can't seem to get through an interview or a news cycle without screaming at a journalist. None of these guys are particularly charismatic, but it's wild that Mélenchon is the voters' favourite.

Purchasing highly specific buttons for friends at the furry convention

heading to a furry con this weekend and people are actually excited that I'll be there

I may not be as attractive or charismatic or talented as any of the people I'm rooming with, but I will absolutely commit to the bit in a maximum-effort way, and furries love high-effort extended comedic bits more than anyone else.

Trending Blogs

Recently Viewed Blogs

jakke