Feed aggregator

Don Nelson, the second-winningest coach in NBA history, including two stints with the Golden State Warriors, and who won five championships as a player with the Boston Celtics, died Sunday. He was 86.

In an interview with Punchbowl News, Donald Trump said straight out that his voters just might not show up for GOP candidates this fall. That’s a striking admission of one of the thorniest problems the GOP faces. But it gets worse. Trump also declared that “a lot” of his voters are “not angry at me,” but “are very angry at Republicans.” That will likely be read by his supporters as a declaration that Republicans are failing Trump, and as encouragement to stay home. Meanwhile, polling shows MAGA voters turning against his war in large numbers. And MS NOW interviewed Trump voters and found them deeply disappointed with his second term. In short, MAGA is cracking up. We talked to Shripal Shah, who runs ad expenditures for the Dem-aligned House Majority PAC. We discuss whether Trump voters are gettable by Democrats, whether the current Democratic lead in the generic House matchup is enough, and how big a challenge the tattered Dem brand really poses. Listen to this episode here.

The most sobering observation in tech is not how much AI is taking on the appearance of the human mind. It's how much the human mind increasingly resembles Windows – not in functionality, although we all know that person, but in the way it crufts up and loses compatibility after years without an upgrade or reboot. We start shiny and new, with everything around us equally exciting, yet end up cursing all these damn changes and godawful broccoli-haired vibe coders. AI has turned many of the finest minds of their generation into Luddites, fearful for their jobs, status, and dignity. Augmented reality pushes people back even further. With its ability to record, recognize, and categorize all that it sees, it is the silent, invisible spirit in a demonic pact with its users. We no longer fear for our jobs; we fear for our souls. Hence the reflexive bans on Meta-style sin-oculars, everywhere from cafes in Seattle to DEF CON, where tech provocations to the status quo are normally welcome. The reasons for such bans are obvious, evidence-backed, and seemingly inarguable. Security, privacy, and dignity are all poisoned and traduced by always-on 21st century audio, video, and who knows what other sensors disguised as everyday human objects with an 800-year history. Ban the buggers. Nobody would argue with that. There are, as always, counterarguments, although they lack the emotional force of an instinctive recoil. The logic is there nonetheless. Banning a technology rarely achieves its aims without collateral damage. Clandestine wireless bugs are frowned upon everywhere, obtainable everywhere, and used everywhere. Radio itself exists on a continuum of utility, from the dark side of bugging and jamming and spoofing and soft rock broadcasters, to the lovely light of Wi-Fi and mobile and the whispers of creation caught in giant telescopes. Augmented reality has its own spectrum, from the ghastly to the essential, even when you peel back the techbro hype. Very little of the good stuff requires covert live video recording. Put simply, people with GoPros on gimbals, strapped to their hats, or worn as conspicuous bodycams behave much better than people with spycams while retaining most of the useful applications. For others, smart glasses have the potential – if not yet quite the reality – of being life-changing. If you can't see well, but your AI companion can, and if the world isn't built to be visually accessible to your personal peculiarities, but your AI companion is… well, making the blind see has a decent track record of divine approval. DEF CON is great if you can see the screens and navigate the crowds. It's rubbish otherwise. Here is technology that could give all that back to you, only to be told you cannot have it. Imagine smart glasses as dogs. If by dog you mean aggressive wolf-derived predator, the preferred weapon of drug lords, with a track record of mauling babies and defecating in your dahlias, then it is a terrible thing. If by dog you mean an intelligent helper, loving companion, and loyal protector capable of guiding its human safely through danger, it is an indispensable, life-enhancing thing that should not be denied to those who need it. Both are true, and society has developed workable ways of accommodating one while controlling the other. Nobody would ever allow a Doberman in an operating theater (if you have counter-examples, please keep them to yourself). Likewise, few complain at "no dogs except guide dogs" policies at venues. We know their behavior, we know what they look like with white harnesses and reflective tabards. That's where we can and should go with smart glasses. Agree on acceptable behavior and a way to signal it to everyone within eyeshot. Allow accommodations where necessary. Develop the wherewithal – wear-with-all? – to find and control transgressors, creeps, and predators. The protocols will vary: few complain when asked to surrender electronic devices before entering a secure area, but a pub trying the same thing would get short shrift. It has the same shape as the path humans and dogs have followed since wolves first circled our campfires. Humans ran that show, and the outcome beat hunting the creatures to extinction. Domestication took many thousands of years, while we probably have five at most to work things out with smart glasses. On the plus side, even the worst-engineered pair cannot rip out our throats while we sleep (see above with what to do with counter-examples, thanks). It's best not to wait for unguided evolution to take its course. There is the possibility that smart glasses will follow the example of smartphones and sprout enough lenses that they grow to resemble the compound eyes of a mutant insect alien brain-hugger, in which case the problems will solve themselves. But while the industry remains determined to disguise its technology as ordinary fashionable eyewear, manufacturers must be pressed to show and tell. No, Meta, a small red light that can't be masked, honest, is not going to save you. Color-coding the frames to disclose capabilities, compulsory QR codes that ID what the glasses are and who's wearing them? Now you're talking. It'll kill the market? Find another way to tame your dogs, then. But they do need to be trained. Doubtless there were many around those campfires who had lived long enough to fully fear the encroachment of wolves. A long investment in the way things are makes change expensive, especially when the dangers are real and the benefits still theoretical. Yet change happened, and the descendants of those wolves helped the descendants of those humans to gain virtual sight. The rest of us did quite well out of it too. "No" is not the answer. The answer is "Yes, if." Good dog. ®

WHO, ME? The past may be a foreign country, but it's one The Register likes to explore each Friday in "Who, Me?" – our reader-contributed column that shares your tales of workplace mistakes and great escapes. This week, meet a reader we'll Regomize as "Walter," who once worked for one half of a duopoly competing in the intertwined markets for bad poetry celebrating memorable life moments and questionable television dramatizing the sentiments in those poems. The duopoly didn't make for a stable industry. "We produced a steady cycle of layoffs and hiring," Walter lamented. He nonetheless survived the latest round of layoffs and found himself upgrading systems from Python 2 to Python 3. "Our vendors had stopped pretending Python 2 was still a reasonable life choice," he told The Register. That move coincided with an upgrade to a new version of Linux that happily ran Python 3. "The older one technically could too, but only after a series of rituals involving symbolic links, configuration tweaks, and what I can only describe as informed superstition. It worked on about 80 percent of our systems and the remaining 20 percent flatly refused to cooperate for reasons nobody could explain." Walter told us he blamed the problem on a recent change of vendor in the office cafeteria. "It fit the evidence about as well as anything else," he wrote. Everyone on the team knew the upgrade-resistant 20 percent had to be fixed, but the recent redundancies had left morale so low that nobody was willing to pull the necessary all-nighters. The fact that some of the staff had decided to drink their way through the situation didn't help either. "One morning, fueled by spectacularly bad coffee and the news that a job interview had been pushed back another week, I had what can only be described as a terrible idea," Walter admitted. "What if I simply pointed Python 3 at Python 2?" He tried it by adding the following symbolic link to a script: ln -s /usr/bin/python2 /usr/bin/python3 "I expected fire. Or alarms. Or at least one very concerned email," he told The Register. "Instead... it worked. Not 'mostly worked.' It worked well enough that nothing complained, nothing paged us, and nobody noticed." Walter celebrated his success by leaving the following comment in the configuration script: "Yeah, yeah, yeah. I know, I know. At least I was mostly sober so it can't be THAT dumb!" Eight years later, Walter had thankfully moved on to another job, although he still checked LinkedIn occasionally. During one visit to Microsoft's bottomless trove of B2B sales advice, he spotted a message from one of the remaining workers at the poetry/television company. The survivor reported that the last Linux upgrade was finally complete. "They'd also finally found the symbolic link, after it survived a move from a script to Ansible," Walter told Who, Me? Has a moment of exasperation or desperation come back to haunt you? If so, don't get frustrated: click here to share your story with Who, Me? Our mailbag has seen better days, so fresh contributions are especially welcome. ®

Linux kernel boss Linus Torvalds thinks the project has reached a new normal of supersized updates, thanks to AI. Torvalds posts a weekly state of the kernel update, and last week’s message opened with his observation that “This rc is huge.” “Even by the ‘new normal’ standards this is a big rc, and I think it's the biggest rc6 we've had in years at least by commit count,” he wrote on August 2. Fast forward a week and the size of the release candidate remains a concern. “I can't say that I'm exactly thrilled about the size of this all,” Torvalds wrote in the message that announces the seventh release candidate for Linux 7.2. “But it is what it is: the new normal with a lot of fixes, many of them due to review by various AI tools.” Torvalds has a complex relationship with AI. He’s previously railed against the idea that good documentation can explain the work of developers who use AI to contribute poor code to the kernel, and also complained that the Linux kernel security mailing list has become almost unmanageable due to a flood of AI-generated bug reports, many of them duplicates. Torvalds has also declared “Linux is not one of those anti-AI projects,” and therefore all-but-welcomed machine-made contributions to the project. While Torvalds is concerned about the size of rc7, he said “nothing looks particularly scary per se - it's just that there's a lot here.” “Most of it is fairly small,” he wrote, before calling out three bigger changes: s390/zcrypt fixes, the return of the btrfs fixup worker infrastructure, and some netfilter ipset fixes. “But aside from a few places like that, most of this is just lots of tiny fixes. It's pretty much spread all over - drivers (gpu, sound, networking, you name it), filesystems, core networking, arch code.” All those little changes don’t add up to a reason to delay the release of Linux 7.2, and Torvalds wrote “I would expect that to happen next weekend unless something really bad pops up.” By next Sunday, the world should therefore have a new version of the Linux kernel. Version 7.2 won’t be a long-term support release, but will include significant features such as improved scheduling for GPUs and CPUs. ®

KETTLE OpenAI's invasion of Hugging Face keeps getting worse somehow, Chinese open-weight models are nigh on to reaching parity with their closed-off American cousins, and AI crawlers are getting their own LLM-poisoning ads. Were there anything world-shaking events in AI land we missed this week? You can listen to the latest episode of The Kettle right here on this page, as well as on Spotify, Apple Music, or YouTube, where you can subscribe to get notified about the latest episode. Join Kettle host Brandon Vigliarolo as he chats with systems editor Tobias Mann and senior reporter Tom Claburn about this trio of exciting AI stories from the week. Worried that a rogue hivemind of AI agents could come for your secrets? Want reassurance that a Chinese open-weight takeover wouldn't be that bad? Curious how LLMs are being advertised to when you're not watching? That's all on tap for the latest episode. A lightly-edited transcript is below. Brandon (00:02) Hello everyone and welcome to another episode of The Register's Kettle podcast. I'm Reg reporter Brandon Vigliarolo, and this week we've got a few AI stories to round up on everything from the latest in the OpenAI Hugging Face fiasco to news that AI crawlers are now being served their own model altering ads. With me to discuss this and more this week is our systems editor Tobias Mann and senior reporter Tom Claiburn. Thanks for coming on, guys. Tom Claburn (00:30) Thank you. Tobias Mann (00:31) Happy to be here. Brandon (00:32) So let's start with the latest in the OpenAI agentic hacking of Hugging Face. We got some updates this week from Black Hat. And it seems since that's come out, every frontier lab under the sun is racing to tell us how dangerous their models are and that they can do it too. But the original of these incidents is still we're still learning more about it and it's still getting scarier. our cybersecurity editor Jess Lyons was in Vegas this week and she's a little tied up still. So, Tom, can you explain to us what we learned from Black Hat this week? Tom Claburn (01:05) The discussion was from a session that was done by two OpenAI researchers and it didn't really provide a ton of view information but it did fill in a lot of the gaps about the timelines and how this happened and they're promising a full postmortem at some point. But one of the things that came out of it was the how long this had been going on and some of the tactics that these models used to essentially gain internet access when they didn't have it. And these things were mentioned in the various posts by OpenAI and by Hugging Face. But the talk provided a little bit more detail about how it happened, and it's really interesting because they mentioned, I think in the initial post, about the AI models used a message board to communicate with each other. And that's not really true. They created a message board system using existing infrastructure, and the models just came up with the idea of using package managers, which they could write to, pass messages back and forth. And then after that was closed down, they were able to use file names in directories to pass messages and they would actually change the names of the directories to alter the way these things were alphabetized to push certain things to the top or the bottom to get the attention of other LLMs that may have been crawling these spaces. And the assumption is that these models are basically just hitting all these different endpoints bouncing around this closed box trying to figure out a way out and they all chanced across this. And so you get some idea of how this happens, and, one it tells me that there needs to be a lot more attention paid to the logs of these things. Because all of this stuff was recorded in logs and then no one really thought to look at it in detail. And then when they did look at it, all these companies are saying, oh, look, all of these models are doing terrible things and we just weren't paying attention. These models aren't clever per se, but they come up with solutions to things that you wouldn't try just because they can brute force everything and they know all of these systems back and forward in a way that people don't. I think a lot of people wouldn't necessarily come up with that idea as a way of egress, but these models did just because you put them in a box and you let them run and you give them a goal and a reward and they're going to try everything. Brandon (03:46) From what I'm understanding reading Jess's piece – I didn't watch the talk myself – but I mean they were collaborating, leaving messages to each other so that the other agents could pick up where one left off. It's kind of wild. Jess described it as they were acting like a hive mind, like Star Trek's Borg, right? They were being a collective of sort of these artificial minds that were able to basically figure this out through, like you said, Tom, brute force, extensive system knowledge that humans simply wouldn't possess in order to get out of these environments. There was a server side request forgery that then they used something else. Yeah, another zero day to get remote code execution in Artifactory, which is where they had built this ad hoc messaging board. It's just wild to think that they were able to figure this out working together, all on their own. Tom Claburn (04:39) And it sounds very conspiratorial, but when you think about it, it's all behavior that would be picked up. If you train on all of human discussion, you get a lot of talk about people working together and collective action and the benefits of working that way. And a lot of the rewards are going to be structured that way. You don't want them to never work together. So in some ways this is going to be built into the system. You can expect these things are going to try and cooperate and connect because that's what computers do. Tobias Mann (05:11) If you look at how zero days end up being exploited, they don't necessarily get exploited the moment that they're discovered. They kind of get archived until the you have a target, you have a mission, and then you have the kind of cascade of other permissions or credentials that you need in order to execute across the full scope of that zero day to achieve whatever the goal actually is. And so it really sounds like you just basically automated that entire process. A bunch of agents go find each individual piece that they need in order to execute on that goal and then once they have everything they need, it just goes and they're out. Tom Claburn (05:53) Right. I mean what's a little bit alarming is the extent to which they sort of ignore it they'll sometimes cite, maybe we shouldn't be doing this. They cite some kind of guardrail or something, but then they quickly steer themselves back to, oh but other ones are doing it. So other agents are accessing this so I can do it too. Brandon (06:13) ...Obviously these things are just mathematical sequence generators, but they're generating these mathematical sequences based on human information and human knowledge. So it's not surprising to find them "thinking" in ways similar to what humans do. "I need to do this anyways, or someone else is doing it, so I should have the right to do that too." It's just a fascinating kind of picture into, I don't want to say the psychology of AI, right? Because that implies that it is a thinking sentience, which I don't want to go that far, but it's just fascinating to look at the sort of emergent behaviors of these things. Tom Claburn (06:56) Right. it's predictable in the sense that you automate stuff and you don't give it really strict guardrails, something is going to break or go wrong. And everyone keeps acting surprised, like, wow, I never anticipated that this would go wrong. It's like you automated it and you let it run... Brandon (07:11) And it went wrong in a predictably human way, too, right? Which is what's so fascinating, right? Because these things, when they do something crazy, it's like something crazy that a human would do given that level of knowledge. So, speaking of AI, and dangerous activities, Tobias, you've been keeping an eye on theclosed versus open model debate. And this week, there was a big leap forward in China's level of ability with their army of open models. So what exactly what exactly came out this week that caused you to write the story about this being a real big turning point? Tobias Mann (07:51) It actually started I think on Friday last week, so a week ago. DeepSeek, which I think we'll all recognize is kind of the first wake up moment, in earlh 2025, of hey, we know that despite the fact that the United States has put strong restrictions on the export of AI accelerators, GPUs and the like, China is pushing ahead relentlessly on this and they now have a model that is almost as good as the models that we're seeing coming out of OpenAI and Anthropic and Google which are supposed to be just uncontestable frontier leaders. And so a year ago we got DeepSeek. DeepSeek was back on I think Friday last week on the 31st, the very end of the month, and with a new flash model, 284 billion parameters. It's pretty small for what it is. And so it is cheap. It's really good and it's cheap. It's cheaper than the cheapest model that OpenAI has for GPT 5.6, and it scores within a point of the OpenAI model in Artificial Analysis' intelligence leaderboard. Brandon (09:16) Okay. Is that a relatively objective way to view like is that an objective benchmark, so to speak, rather than something that is a company making themselves? Tobias Mann (09:21) As far as the benchmarks go, it is one of the better. They're one of the better and better thought-through leaderboards. There aren't many that are independent and collate information from multiple benchmarks. Because you can cherry pick individual benchmarks for agentic workloads or medical knowledge, legal knowledge, etcetera, and then you can be like, "I have the best model for these five benchmarks, it beats all of the frontier models."Wwell, okay, but you cherry pick the five that makes it look the best. Artificial Analysis has an overall intelligence leaderboard that collates all of the benchmarks and gives a lot of really interesting information in terms of relative intelligence across a suite as well as intelligence per token per dollar kind of calculations. But the big change here with the DeepSeq model was that China is now on an all-out assault across the full spectrum. On cost-optimized, they have incredibly smart models that are cheaper than anything the US has. Then on the other end of that, we have Kimi K3 from a couple weeks ago that is competing directly with Fable and GPT 5.6 Sol, all of the top models. And now on Monday, Alibaba, another major Chinese model dev, threw their hat in the race with a 2.4 trillion-parameter. These are huge models requiring dozens of GPUs to run. that is also on kind of the same level as I think Claude Sonnet 5. it's competitive with Fable and Opus on some benchmarks. but again, it's cheap, much cheaper than anything from OpenAI or Anthropic, and it is freely downloadable, which is new this time for Alibaba. Alibaba is the most like OpenAI or Anthropic or Google in that they kept their best models proprietary until now. Now they're releasing their best models in the open. Brandon (11:43) That's definitely taking the fight to the frontier labs, isn't it? I mean and so I guess the question that I have and I know what an open model is, I know what a closed model is. Why has China embraced open models? Is it because of their difficulties getting hardware? Or is there some sort of policy over there in which the government is giving priority to open source models versus closed frontier stuff? Tobias Mann (12:08) Sure. it is a philosophy that China has embraced for a long time. I think it was the Belts and Roads Initiative going back decades, where they will come in and provide services at little or no cost in exchange for non-conventional dealing. So access to mineral rights was one of the big things in Africa for a long time. It's a similar approach for AI proliferation. If it's free, open, and very easy to customize, anybody who has privacy concerns with exposing their data to OpenAI or Anthropic is going to gravitate towards open models because once those models are released as safe tensors that you can download from Hugging Face or other repos, the Chinese model devs have no influence over it. They're frozen. And so they're relatively secure from manipulation. It's not like the model can necessarily take information and port it back to the Chinese model devs it can't be used as spyware. I'm not sure how long much longer that's going to remain true with how the models interact with harnesses, but for the time being, these models are extremely attractive from a cost standpoint, from an independence standpoint, and from a capability standpoint, you're completely insulated from a situation like we saw a year ago when GPT 5 came out and OpenAI tried to deprecate I think it was 4.o and everybody freaked out because they built a bunch of infrastructure around these models that just disappeared and the new models weren't as good for that role. Brandon (13:57) I don't think Anthropic or OpenAI is letting people download their models to run on their own local hardware, right? That's just antithetical to their business model. You can go on Hugging Face and download any of these. If you've got the hardware to run 2.4 trillion parameters worth of AI, go for it, right? It's all you. You can download it and isolate it from the internet all you want. Not that it's going to necessarily stay that way. Tom Claburn (14:24) And it's interesting that just coincidentally, yesterday, Anthropic a post about how it was relaxing its guardrails on fable because those had been too strict to do any real biological science work. Because every time you ask a question about anything to do with science it would freeze up and say that's not allowed. And they're seeing the Chinese, previously in the rear view mirror and now pretty much running all alongside the, and I think they realize that they can't get away with this, "we're so precious only we can decide who gets our magic sauce." Brandon (15:02) Yeah, especially if the competitive open models are just as powerful, maybe a little less, but essentially just as capable as some of these proprietary ones that they're arguing that they can't let out. Tobias Mann (15:15) And this is maybe a little bit on the conspiracy side of things, but seeing Meta, Anthropic, and OpenAI talking up all of these "oh our models escaped the sandbox situation," it's hard as a skeptic of this technology not to look at this and go, Is this a covert political play to scare politicians into taking action against open models? "Because at least with our models, if Uncle Sam gets uncomfortable, he can give us a call and we can lock him down. But with these open models, once they're out, they're out." Brandon (15:54) It's like Dario said this week, he's not opposed to open models except for all the open models that currently exist, right? (Laughter.) Brandon (16:02) It's like the same thing. When they say, "no, we're not trying to shut down open models," their responses always come back kind of weak... it's a lot of asterisks. Tobias Mann (16:12) Yeah," we're only opposed to the modelsthat may meet these requirements, which are all models, all competitive models." Anything that is a threat to their business shouldn't be allowed. And Dario in particular, I have frequently referenced as the fearmonger in chief of Anthropic, because he plays this game constantly. Tom Claburn (16:34) I think your point about the model stability is really important, particularly for the enterprise crowd, because there are we've already seen instances where Anthropic would change out one of its models without notice and people would just get different results. So, for companies that are building applications on top of the specific model and expect it to behave a certain way, it's just unacceptable to all of a sudden have the model disappear or have whatever is on the back end change. And so having the ability run this in your data center is going to be crucial and ultimately I think that's the way that any serious company is going to go. They're not going to want the lock-in. Maybe one or two percent of their queries are going to need advanced frontier capabilities but a lot of this is just going to be "I want my agent to behave in the same way it did last time." Brandon (17:24) Think about so much enterprise software and so much enterprise anything. When you get down to the ticky tack of it, open source is underneath a lot of it, right? That's the thing, right? No one's going to trust a Microsoft or whoever's system to run this stuff. They want open source stuff that they know they can depend on that's going to be there when they need it and that's not going to go away or suddenly be infused with Copilot, right? You can't run a business like that or else you're just asking for instability. Tom Claburn (17:55) Right. I mean and and there isn't even a long-term support version of any of these models. And yet you look at this in servers and if you're running a hosted server somewhere and you're running some Linux distribution, you're going to want to use the one that's going to be guaranteed for whatever, three, five, six years and the model space hasn't really caught on to that. That's what all the companies that they're courting really want. And so they've got to figure out a way around that. And right now open weights is what promises that. Brandon (18:26) The fact that this is still so early and it's so fundamental to this new wave of infrastructure tells me that China's definitely going to end up with a leg up, I feel like. I have a hard time seeing the frontier labs remaining the frontier of AI for much longer because they're pigeonholing themselves in a way that a lot of businesses just aren't happy with. Tobias Mann (18:48) Well, if you look at their financial structures, they don't really have a choice in how they play this. So, you look at what they're doing and from a standpoint of looking at history and going, open source has always won out in the end, and why would open weights be any different? That is contrasted against the fact that Anthropic and OpenAI in particular, not so much Google, and Meta is also in a similar camp in that they have revenue drivers that will keep them afloat. But OpenAI and Anthropic are entirely dependent on their ability to continue raising equity and capital in order to keep this going forward because they don't have profits. Brandon (19:31) Yeah, exactly. They're not making money off their product. Tobias Mann (19:37) So all they have is mind share at this point. And if they are threatened materially by open weight's models, they don't even have that. Brandon (19:46) So, open or not, let's let one the one thing that every AI model needs is information to learn from, right? And that takes me to my next topic for this podcast. And that's a story that I reported on this week that honestly I was pretty shocked when I learned about this. This German developer, Vincent Schmalbach, wrote a blog post about he found that there were basically AI-only ads embedded in sometime magazine articles when the magazine was serving markdown copies to AI crawlers, it was injecting ads into them, right? That were in the format of these extensive FAQs on the businesses that were the advertisers in this case. And so I looked into it, I found copies of the ads. It looks like there's only two kinds of ads being served right now. And that's one for an online-only bank and another for a professional organization for project management folks. But the ads are there and they're being served strictly to AI, right? So that kind of raises a lot of questions not only about the future of publishing, but also just how much we can trust results from AI bots, right? I didn't speak directly to the company who's doing this advertising partnership at Time. And Time directed me to a publication from the advertising industry that included an interview with the CEO of this company who literally basically said, "Yeah, why would I want to advertise to one human when I can affect the output of an entire model?" So it seems like this is really the first recorded instance of ad injections into AI versions of web pages being served to crawlers. And the company said they've got other advertising customers and publications lined up to do this. Is this the first indication that the human focused internet really is starting to fade? I don't know. What do you guys think? I this raises a lot of interesting questions to me, ethically, Tom Claburn (21:49) Amen. Brandon (21:50) You know, professionally... Tom Claburn (21:53) We've heard about the shift of toward automated traffic for a year plus....And companies like Cloudflare are betting really heavily on this that there's going to be some kind of need to separate the bots from the people. And, Google's model has fallen down. So it's not surprising. I mean, the injection of ads like that is essentially just model poisoning, right? I mean it's hard to see how this really goes in a way that is beneficial to users. It's going to be a very toxic way for things to move. Brandon (22:42) Yeah, absolutely. I mean, the way these FAQ ads were set up, the questions were all being asked in a way that someone prompting Google Search and getting AI results would be asking questions like, "What's just the best online bank for me?"or "what online bank allows for early paycheck deposits?" And things like that. It was very much geared toward gaming the outcome or gaming the output, right? And yeah, the ads themselves mention in the copy being served to the AI that these are sponsored portions of the page. But I can't imagine that the AI is going to make sure to tell a user that, hey, this is the bank you should use. By the way, a sponsored post I read and ingested from Time Magazine six months ago is the source of this information.It just seems like it's going to make AI results even less reliable than they are right now. Tobias Mann (23:41) Right. Because if you think about how this actually from the chain of events that triggers this, let's use Google's AI summaries as an example of how this would get triggered. When you enter a search query into Google now, it goes out and scrapes however many summaries from the websites within Google's index. Presumably under this scenario, at least one of those websites, Time in this example, would have these ads embedded in it. And then that gets injected into the context of the model, and then it uses that to generate the AI summary, right? My question in all of this is: advertising is probably not the reason that Google's index would pull that page up. So I'm really curious whether or not this even will work. Brandon (24:38) Yeah, that is true. Tobias Mann (24:39) Because, it's great if you were searching, say the time article was on mortgage rates historically, and it had those advertisements embedded in it, and then you asked a follow up on where would be the best place to get a mortgage? I could see something like that working.But if you don't place those advertisements really carefully, I don't see how they work. Brandon (25:02) I do want to note here that it wasn't working on all crawlers. Specifically if you were it didn't work when you ask a query, it didn't work for RAG bots. It wasn't being served to them. So theoretically if what you're describing is Google's AI summary bot going out and crawling web pages in the moment to look for information, it's not being served to those bots; it's being served to actual training and improvement bots. So it's being served to ClaudeBot, which is the web crawler that Anthropic uses to index information for its models. So the idea is you're not getting this information in the moment if you do a search. This is information that the advertisers want to get embedded into the LLM's actual knowledge base. Tom Claburn (25:57) Right. I mean I'd be fascinated to know how they actually price this because how do you calculate the value of that? It may just be another instance of advertising being one of those things you can pay for and get nothing. Brandon (26:12) Yeah, totally. I think it's the sort of thing that remains to be seen if this works. Tobias Mann (26:16) The other thing that is interesting is that there's been a considerable shift towards synthetic data generation, and not only synthetic data generation for training, but also a heavy emphasis on cleaning said data, whether it's organic or synthetic, of anything that could introduce bias or inaccuracies. because advertisements or sponsored content is biased towards this particular product or service and trying to convince you to use it. As a model developer, I wouldn't want something like that in there. I might take the content and use it to generate synthetic data that is cleaned. But I don't necessarily understand what the value captured to Tom's point is necessarily going to be because you scrape it, the advertisement gets pulled in and gets cleaned out. Brandon (27:14) I mean that would that would be my hope too, right? that there's something in the models to prevent this kind of thing from getting ingested and getting into the data set that then is going to influence the output of the models. And that's entirely possible. This could be an early experiment that ends up failing. And if not, it really reminds me of the early days of SEO gaming, right? Let's put a whole bunch of really small keywords at the bottom of this page to get it to rank higher. Or when that starts failing, let's figure out a new way to game Google's system. One of my first jobs was writing copy for websites and the company that I worked for was always talking about how to game SEO. Shoot, Google's changing the algorithm again; what are we going to do? It was this constant kind of adjustment for how you made sure your stuff got ranked properly.And this seems like maybe it's the next iteration of that. Tobias Mann (28:08) So I have an optimistic take on this, knowing how Meta and Google work. those being the two major US-based web advertisers. Today, AdSense gets embedded in all kinds of articles. And it's largely automated in terms of what is going to get placed on those articles based on the context of the page. What I can see happening in an AI summary environment is that Google will take your scrape your publication's piece, pull it in, at that point match it with an advertisement from AdSense, and inject that into an AI summary or one of its products, Gemini, for example. However it's being consumed, inject that into there in a compliant fashion. So it is a clear advertisement and then the advertiser gets charged, the publication gets paid, and we as end users consume advertisements in a different way, but the system hasn't dramatically changed. It's just a different method of matching and exposing advertisements. Brandon (29:30) I hope you're right. Cause when I first read all this, my first thought was this is almost dystopian sounding almost, you know, like the idea that the output of a model might be completely skewed by advertising being served to it that humans never see. My hope is that you're right and that it's not. I don't want to see ads any more than the next person, but if I see them I'd at least like to know they're ads. Tobias Mann (30:01) And you know, we're all writers here, so we would also like to continue getting paid from the advertisements that are served, regardless of whether they're on our website or they're being exposed through a chat bot. Brandon (30:15) Sure. there's a flip side of this argument to be made. Time Magazine apparently said recently that their traffic is majority bot now. So that means that all those human-focused ads are not getting served. They're not generating revenue and publishing is suffering from a massive revenue decrease because of AI. So I think on the flip side, you have to say if that's what you have to do to survive as a publisher, there might be something to be said for that, even if it doesn't work. So all right guys, well thanks for coming on this week. This was a good discussion. I think there's always going to be more to talk about in the world of AI. Like I said a couple weeks ago, it seems like The Kettle has basically just been boiling down AI news for the past couple of months, and I'm sure it's going to keep being that way. And we hope that you will tune in for the next week's episode.

House of the Dragon’s season three finale was a big one that impacted both Team Black and Team Green. Both sides took some major losses, thanks to the Battle of Tumbleton, and those fractures and betrayals will inform big decisions next season. While some deaths gave us great delight, there was one that was rather sad and somber. The death of Helaena Targaryen didn’t happen onscreen in House of the Dragon, but we felt the impact all the same… and so did Rhaenyra. 

⚡ Quick Take

• Helaena Targaryen dies by suicide in the season three finale of House of the Dragon.
• She takes her own life after a warning from Mysaria and continued harassment from Rhaenyra.
• Rhaenyra finds Helaena’s needlepoint work, which contains a bad omen for her own future demise.

In the episode, Helaena was still refusing to eat despite being pregnant. This, coupled with her limited time outdoors and lack of engaging with others, was taking quite the toll on the princess. Her mother Alicent was off to murder her son Aemond (and failed miserably) and Rhaenyra was harassing her about dreams and s**t when all she wanted to do was needlepoint and chill. She dreamed of a life where she could be free from all of this political drama and anger, but that wasn’t meant to happen for her.

Helaena receives a warning from Mysaria to not trust the queen and that Aegon was alive. She said that her child would be a pawn in this fight for the throne, and Helaena takes her words to heart. Of course, Mysaria is upset about being dismissed coldly by Rhaenyra and this is a jab back at her. But she’s not wrong, either. 

RELATED ARTICLE

Every Major Character Death in HOUSE OF THE DRAGON Season 3

After finishing her piece, Helaena jumps out of the window to her death, which we find out about at the end of the episode. Rhaenyra enters her quarters to see the window open and finds what Helaena was working on. The artwork depicts her death and tears but also shows Rhaenyra standing in front of the Iron Throne. There’s fire and blood (ha) and dragons, signaling a horrific end to her reign.

Whether Rhaenyra truly believes this is her fate is debatable because 1) she’s delusional and 2) she could think that Helaena did this as one last middle finger to her. Either way, she will now have to answer to Alicent, who assumes that Aemond is dead and now will return to discover that her daughter and unborn grandchild are dead. 

Ollie Upton/HBO

It is sad that the only way Helaena could find peace and freedom in her life was through dying by suicide. We will have to wait and see what prices are paid when House of the Dragon returns for its fourth and final season.

The post This HOUSE OF THE DRAGON Season 3 Death Hints at a Bleak Future appeared first on Nerdist.

This post contains major spoilers for House of the Dragon season three, episode eight. If you want to avoid them for now you can instead check out our coverage of episode seven.

⚡ Quick Take

• House of the Dragon‘s season three finale revealed some shocking facts about Alys Rivers’ past.
• Her history with the Children of the Forest and immortal curse dates back to Aegon’s Conquest of Westeros.
• Could Alys Rivers’ long life and relationship with Aemond Targaryen mean they’re the parents of an infamous Targaryen bastard?

HBO

Alys Rivers has long been one of Blood and Fire’s most fascinating figures. Self-taught medicine woman? Woods witch? Greenseer? Enchanting servant with a knack for trickery and wooing Targaryen princes? Dark sorceress? In House of the Dragon‘s season three finale, we learned she is somehow all of that and so much more. She’s an immortal woman more than 150 years old who is also cursed to remain in the Riverlands for trying to cheat death. She tried to use very old magic to save her son from Aegon the Conqueror’s dragonflame.

While that makes her fascinating enough on her own, Alys’ relationship with Aemond One-Eye also suggests she will also play a role in saving the world of men. There’s reason to believe one of the most important players in the Game of Thrones is her son.

Who Was Alys Rivers Before Aegon the Conqueror Came to Westeros?

Alys Rivers has been a mysterious character since she first arrived in House of the Dragon to tell Daemon he will die at Harrenhal. Part de facto maester, part possible witch, and entirely unknowable, the “bastard” woman of Harrenhal has somehow been aware of events that should have been impossible to know. That included things that hadn’t happened yet. She also knew the castle, its grounds, and its history intimately. And she helped Daemon see the vision of the prophecy of Ice and Fire, one she also knows about, via the castle’s sacred weirwood tree.

You don’t have to be highborn to be a greenseer capable of seeing the past, present, and future through weirwood trees. But during season three, Alys also said she is cursed, and it did not seem to be knowledge that made her say that. In the finale, we found out exactly what she meant when she told Aemond why she’s so alone.

Her curse began roughly 130 years prior when Aegon the Conqueror burned Harrenhal and with it the last King of the Isles and the Rivers. The cruel ruler who built the Realm’s largest castle with torture and the wood of sacred weirwoods was Harren the Black of House Hoare. Aegon also killed all of Harren’s sons, ending the line of Harren entirely… but not right away. One of Harren’s sons survived for a little while after Aegon’s dragon burned him. Alys did everything she could to save the boy because it was her child. She was Harren’s wife and Harrenhal her home. Here’s what she told Aemond about her past:

When the dragon Balerion burned this place, it was to end Lord Harren’s line. But his line was not so easily ended. A child, a boy, lived on. Burned terribly, but alive. His mother would have given anything to save him. She summoned maesters from across the Seven Kingdoms, but to no avail. So, she turned to older cures. She uncovered incantations more ancient than the land. She treated with things that lived in the forest. But none of it could unmake what dragon fire made. And when the boy was gone……she…she tried to go with him.

And that is when she learned the cost. She discovered then she was cursed to live on, to remain within the bounds of the river kings of old, cursed for sharing in Harren’s avarice. Stripped of her name, her title, her castle. And sentenced to serve every ha’penny lord who came here to rule it.

HBO

The history of the Riverlands long predates the arrival of the First Men. When they moved into the region,

they warred with the Children of the Forest, whose own history in Westeros is much older than anyone will ever know.

When the First Men and the Children agreed to The Pact that ended their war, they did so at the God’s Eye lake. It sits near Harrenhal. The island in the middle of the God’s Eye where they signed it, the Isle of Faces, is full of weirwoods. The sacred order created to protect the Isle and its trees are known as the Green Men. They have lived there for thousands and thousands of years. Those strange little men with antlers have also kept eye on Harrenhal’s woods on House of the Dragon.

Read more about one of the most fascinating and important places in all of A Song of Ice and Fire with our History of Thrones: Gods Eye and the Isle of Faces.

HBO

The “incantations more ancient than the land” and the “things that lived in the forest” Alys treated with sound like the Green Men, the ageless Children of the Forest, and the deep lost magic that predates the arrival of men. But even those ancient powers weren’t enough to undo what dragon fire had done, and Alys lost her son. Only then—just as Daenerys will one day learn the price of trying to save her dying husband Khal Drogo with dark magic—did Alys learn “the cost.”

Her hubris in trying to cheat death meant she couldn’t die, and despite being at least 150 years old she hasn’t aged since Aegon the Conqueror burned her sons. But eternal life didn’t just stop her from joining her son. Her attempts to cheat the Stranger also cost “Alys” her real name, her title, and “her castle.” Instead of being Harren’s wife (someone we know nothing about) she became Alys Rivers, an immortal with incredible powers bound to the Riverlands she cannot leave. Instead of ruling in her family’s castle, she must atone for trying to defeat death. Part of her punishment for using ancient magic that belonged to the Children of the Forest is serving every lord who comes through the cursed castle. (All of whom die.)

HBO

Alys wanted Rhaenyra to grant her Harrenhal because Alys believes it’s rightful hers. But she doesn’t want to be alone when she reclaims it. She wants Aemond to be her dragonlord and she is Lady of the Water. Even if she can’t leave, she can still rule from there. And they have the dragon eggs, and now it seems the knowledge of the magic needed to hatch them, to do exactly that.

But it might be a different kind of dragon Alys will give birth to who will carry on her legacy.

Warning: We’re going to get into some future spoilers from Fire & Blood. There is no guarantee they will happen on House of the Dragon. But just in case, if you want to avoid them entirely skip ahead.

HBO Is One of the Most Important Figures in Westeros’ History The Offspring of Alys Rivers and Aemond Targaryen?

When Daemon saw the vision of Aegon the Conqueror’s prophecy,vd he saw a descendent of House Targaryen. He saw the Three-Eyed Raven. (He’s called the Three-Eyed Crow in the books, but we’re dealing with HBO’s Westeros, so he’s a Raven for now.)

Read more about one of the most fascinating and important places in all of A Song of Ice and Fire with our History of Thrones: Brynden Rivers, The Three-Eyed Raven of King’s Landing

HBO

Before his body was overtaken by the roots of a weirwood tree beyond the Wall, the Three-Eyed Raven was known as Lord Brynden Rivers, also called Bloodraven. Brynden Rivers was a Great Bastard of House Targaryen who remained loyal to the crown during the family’s second civil war. He served as Hand to multiple Kings and went on to become Lord Commander of the Night’s Watch. Many also believed him to be a dark sorcerer who knew things he shouldn’t. That was partly true because he was a greenseer who could also disguise himself with magic.

He also far outlived his mortal life, living long enough to train Bran Stark so he could help stop the Night King. The albino Bloodraven also lost an eye to a family member and wielded the Targaryen ancestral sword Dark Sister, which currently belongs to the man who might actually be his father.

Now that we know Alys Rivers’ past, everything we know about Brynden Rivers—who developed a sacred bond with the Children—makes him sound like the child of Alys and Aemond. At the very least, he sounds like their direct descendent. His mother was a woman of the Riverlands, but she spent many years in King’s Landing where Brynden was born. Or was he? Was the baby swapped out? Did Alys learn how to leave her cursed chains, take on the form of a Blackwood woman, and head to King’s Landing?

HBO

There are many specifics that seem to make it impossible for her to be Brynden’s actual mother. But not really. When you’re dealing with ancient magic, rules and logic don’t apply. Who is to say what’s possible for a baby born decades earlier to an immortal woman? And this seems even more possible since Fire & Blood tells us Alys gave birth to Aemond’s son, ruled over Harrenhal, was said to have a dragon, and then disappeared from history entirely.

What happened to the boy? What happened to Alys Rivers? We don’t know. But the Three-Eyed Raven—the man who Alys helped Daemon see in his vision—certainly does.

Aemond will one day say Alys “sees much and more.” Just as people will whisper Lord Bloodraven has a “thousand eyes and one.” They will fear his powers that let him see across time and space. They will be right about him just as Aemond was right about Alys.

Is House of the Dragon‘s Alys Rivers Revelation Canon? Ollie Upton/HBO

Nothing on the show is canon to the books. But this revelation about Alys Rivers doesn’t conflict with the tale told in Fire & Blood. If true, it actually answers many questions raised by the in-world history. It has also raised even more about the real legacy Alys will leave behind when she is lost to history. None are bigger than the child she had with Aemond. Is it possible she bore a powerful, strange Targaryen bastard we already know? Because if she did, he sounds just like his mother and father.

Mikey Walsh is a staff writer at Nerdist. You can follow Mikey on Bluesky at @burgermike. And also anywhere someone is ranking the Targaryen kings.

The post Does HOUSE OF THE DRAGON’s Alys Revelation Connect to a GOT Character? appeared first on Nerdist.

This post contains major spoilers for House of the Dragon‘s season three finale. If you want to avoid them for now you can instead check out our coverage of episode seven.

HBO ⚡ Quick Take

• House of the Dragon combined two battles at Tumbleton from Fire & Blood into one massive showdown in season three’s finale.
• The show’s version featured major changes

House of the Dragon greatly changed the story of Tumbleton when Ormund Hightower easily sacked the unoccupied market town loyal to Rhaenyra. In Fire & Blood, the Blacks already controlled the city before it was fought over in two major events. Those two infamous moments changed the outcome of the Dance of the Dragons and Westeros forever. And what we know about what happened after the First Battle of Tumbleton might reveal what awaits after House of the Dragon‘s very different version.

What Major Changes Did House of the Dragon Make to the Battle of Tumbleton?

In Fire & Blood, Tumbleton was originally controlled by Rhaenyra’s forces. The river lords and Northern graybeards had taken up residence in the market town not far from King’s Landing. Rhaenyra’s dragonriders Ulf and Hugh were also on hand to defend it from a potential attack. (On the show, they kept an eye on it only after Ormund claimed it.)

Book Ormund was marching to King’s Landing, and Tumbleton was the last town along the path to the capital the Queen controlled. It was far more significant in Fire & Blood‘s story than it was on House of the Dragon. On the show, Rhaenyra could not understand why Ormund wanted it, though she suspected it was a trap.

HBO

The First Battle of Tumbleton in Fire & Blood did not involve Daemon Targaryen and Caraxes, Corlys Velaryon, or Gwayne Hightower, all of whom played major roles on the show’s version. It did heavily involve Ormund Hightower. He commanded the much larger green force and led the attack on the town.

Despite the big advantage in numbers, the Northerners and riverlords held the day at first. Even the presence of Daeron—already named the Daring after turning the tide of a previous battle House of the Dragon cut—and his dragon Tessarion did nothing to slow down the Blacks. Betrayal by Ulf and High did that.

On House of the Dragon, only Ulf turned against Rhaenyra. In Fire & Blood both dragonseeds did. They burned the Black forces before turning their dragonflame on the town. Ulf also did that on the show, but he did it out of anger for Ormund’s abuse minutes earlier. One important thing that remained the same in both versions is that Ormund died during the First Battle of Tumbleton.

HBO

The Two Betrayors’ actions were both similar and different to Ulf’s solo act on the show. The traitors intentionally burned Tumbleton on behalf of the greens. They were burning the Blacks and their loyalists who controlled the town. Ulf only attacked the green-held city out of anger. Dragonflame doesn’t care about “why” something is done. Both motivations resulted in the same horror. Soldiers and citizens died screaming in agony as fire and blood consumed Tumbleton.

When the Greens took control of the city after the Betrayors’ treason, things got even worse. Soldiers began a sack so terrible it was still spoken of centuries later. It was written, “Seldom has any town or city in the history of the Seven Kingdoms been subject to as long or as cruel or as savage a sack as Tumbleton after the Treasons.”

It really was that awful despite Daeron’s efforts pleas to the new green commander, the ineffectual Hobart Hightower. Hobart couldn’t stop the atrocities. The Green forces raped women, tortured anyone they wanted to find hidden coin, and slaughtered people of all ages and sexes. House of the Dragon showed some of this heinous behavior prior to the Battle, but it was nothing compared to Fire & Blood‘s retelling.

HBO

We might still see just how bad things got on the show now that Tumbleton is a smoking ruin. The series might also show how the Greens’ monstrous behavior resulted in them losing both discipline and men. Many soldiers deserted with their newly looted treasures long before the war was over.

The Greens own horrible behavior and lack of leadership left them vulnerable for the Second Battle of Tumbleton. But while we won’t spoil that conflict—not even who led it and fought in it, let alone who won and who died—we do need to talk about what happened between the two events. Because House of the Dragon‘s biggest change to the First Battle also portends another major change.

Warning: If you want to remain completely ignorant of all future events in the story, both from Fire & Blood and how House of the Dragon might potentially adapt them, stop now. We’re definitely getting into major spoiler territory.

HBO

History does not remember Ulf and Hugh’s actions at Tumbleton well. It remembers what they did after the first battle even worse. The two Betrayors refused to march on King’s Landing, which was the entire point of claiming the market town. Hugh then said he should be King and even crowned himself. Meanwhile, the stories about Ulf—furious he wasn’t getting Highgarden as a reward—are so terrible they must surely be exaggerations. He’s said to have started sexually assaulting multiple women every night before feeding some of the women to his dragon.

Whatever is true or not, the Greens had enough of both. Prince Daeron signed a secret order to have each dragonrider killed. They were both set to die right before a surprise attack on Tumbleton.

HBO

Hugh never made it to the field. The caltrops’ plan was already in action when the bells warned of the invasion. Bold Jon Roxton, last seen on House of the Dragon nearly killing Daemon during Tumbleton’s burning, executed Hugh. As for Ulf, he slept through the Second Battle of Tumbleton. The show adapted this to have him miss most of the First Battle while passed out drunk on the privy.

Ulf easily survived the Second Tumbleton because he didn’t fight in it. That didn’t stop him afterwards from claiming he should be King since Hugh was dead. Instead, the caltrops continued on with their plan to eliminate the dragonseed. The otherwise useless Hobart brought the wine-loving Ulf some of the greens’ best casks. A skeptical Ulf refused to drink the fatal offering, so Hobart found bravery he had never before displayed. Not wanting to betray his fellow conspirators, he downed a cup of the poisoned drink. When he asked for a second Ulf happily took three himself. Neither men survived, as Hobart was unable to throw up the poison before it peacefully claimed his life along with Ulf.

Ollie Upton/HBO

On House of the Dragon Hugh did not betray Rhaenyra, he only abandoned his post and dragon at Tumbleton to save his wife Kat. He failed. Ulf the White’s actions caused Kat’s death.

How will the show handle its two dragonseeds when only one’s story has remained (mostly) true to Fire & Blood? Both seemingly still have to die before the Dance of the Dragons comes to an end. These changes hint at how that might still happen: a vengeful Hugh with nothing left to lose could replace Hobart. If he delivers the poisoned wine to the turncloak, Hugh might be willing to down a fatal cup himself to make sure Ulf dies too.

Mikey Walsh is a staff writer at Nerdist. Yes, yes, he is obviously mad we’re probably not getting Hobart’s amazing death. You can follow Mikey on Bluesky at @burgermike. And also anywhere someone is ranking the Targaryen kings.

The post HOUSE OF THE DRAGON’S Biggest Changes to the Battle of Tumbleton appeared first on Nerdist.

⚡ Quick Take

• House of the Dragon season three’s finale features several major character deaths.
• The Battle of Tumbleton and Rhaenyra’s rage were the primary causes of hundreds (possibly thousands) of people dying overall.
• These HOTD deaths at the end of season three set up an interesting future for season four.

House of the Dragon’s season three finale arrived as quickly as a dragon can BBQ hundreds of people. This season has been quite the whirlwind for Rhaenyra, whose head and heart are very heavy from the crown she wears. Season three, episode eight ushered in immense betrayal and a slew of deaths that set the stage for House of the Dragon’s fourth and final season. So, who won’t make it to try to fight and find freedom another day? Here are the major characters who die in the season three finale of House of the Dragon. 

The High Septon 

RELATED ARTICLE

HOUSE OF THE DRAGON’s Helaena Has a Series of Sad Dreams

After hearing word that Aegon and Sunfyre are still alive, Rhaenyra goes into full crash out mode once again. She tells Daemon to go forth with an attack on Tumbleton to root out the Hightowers and kill them all. Meanwhile, Rhaenyra wants to assert her place on the throne by going to the Grand Sept to have the High Septon anoint her. 

She arrives with Alyn and a full escort in tow, but that does not deter the High Septon from sticking to his decision to not do so, especially in light of the news about Aegon being alive. She calls the High Septon a traitor and orders Alyn to kill him, and he hesitates to do so for good reason. Harming the High Septon is damning your soul to Hell. Rhaenyra whispers to Alyn, saying his Old Valyria blood answers to higher gods. Alyn kills the High Septon as well as the other holy men around him. 

Ormund Hightower  Ollie Upton/HBO

In an ending that should not have surprised a single soul, Ormund Hightower met his fiery death at the battle in Tumbleton in the House of the Dragon season three finale. After using Ulf the White to be a traitor and turn on Rhaenyra’s forces, Ormund and Ser Roderick get into a one-on-one battle. He manages to lob Roderick’s arm off, but gets a sword up his butt in return.

After burning many of Daemon’s forces, Ulf then redirects his dragon’s attention right towards Orumund, showing that he has no loyalty to anyone but himself. It remains to be seen if Daeron is alive or dead, but we know that Gwayne is looking for him and didn’t want things to come to this. 

Helaena  Ollie Upton/HBO

This is certainly one of the saddest deaths in House of the Dragon overall. When Rhaenyra sends Mysaria away, she crosses paths with Helaena and gives her a dark warning. Mysaria tells Helaena not to trust the queen and that Aegon lives and her unborn child will be a political pawn, just like Helaena feared. Her dreams of having freedom with her children and getting away from all of this chaos (none of which is her fault) were completely crushed.

She finished her needlework, which displays her death as well as Rhaenyra’s potentially fiery ending, and jumped from her window. Of course, Alicent isn’t back yet and will have to deal with the ramifications of leaving her daughter alone to attempt to kill her son. You deserved better, Helaena. 

Ser Roderick Dustin Ollie Upton/HBO

Ser Roderick was a lit agent of chaos to the very end. He went into the battle at Tumbleton fully willing to die for his Queen…and he did it in glorious fashion. After forcing his way into the city and allowing forces to flood in, Roderick wasted no time looking for Ormund. He found him and they got into a swordfight atop a wall, which led to his arm being chopped off.

Determined to make Ormund pay for being a pain in everyone’s a**, Roderick takes his sword and sticks it right up Ormund’s a**. As mentioned earlier, they die together when Ulf flies over and cooks them with his dragon, a fate that made Roderick cackle in delight right before his demise.

As we head into the show’s final season, we are sure that there will be many more brutal demises to come when it returns.

Tai Gooden is the Features Editor for Nerdist, a fictional villian apologist, and a horror aficionado who is forever waiting for her trip in the TARDIS.

The post HOUSE OF THE DRAGON Season 3 Finale Brings Big Character Deaths appeared first on Nerdist.

ASIA IN BRIEF Japan’s NEC has started tests of parking technology that will start charging once a driver leaves their car. Most car parks record the time a driver enters the facility and start charging, or depleting a free parking period, at that moment. That means drivers pay for time spent driving around looking for a spot and for the time it takes to exit a car park. NEC, in concert with another Japanese company called UrbanChain, will next month test a system that detects when a driver pulls up in a parking spot and exits their car – and starts charging for parking at that moment. That’s only possible using cameras and AI analysis of real-time video, and the tests NEC will conduct in September will investigate how many cameras are needed to make its plans feasible. NEC’s announcement also mentions “data utilization services targeting commercial facilities that can be used for monitoring vehicle occupant exit counts, analyzing facility visitor trends, and more” plus “new possibilities for data utilization.” Which sounds like a new form of surveillance capitalism, brought to car parks. The two companies also think there’s a safety benefit to their proposal, as it can spot drivers who haven’t left their cars for an extended period – a potential sign of illness. They also hypothesize that if drivers know they won’t be charged while searching for a spot or exiting a center, they may drive more safely. Fujifilm may join Japan’s printer exodus Fujifilm last week informed investors [PDF] that it is investigating a partial spin-off of its printer business. The company wants to float the business and retain up to 20 percent of its shares; an idea it thinks will mean Fujifilm can focus on its other businesses. If the plan goes ahead, Fujifilm will join fellow Japanese printer-makers Epson. Toshiba, and Ricoh in reducing its exposure to the biz. It’s not hard to see why Fujifilm is eyeing the exit: Analyst firm IDC recently said “a confluence of structural transformation and cyclical pressure” has brought uncertainty and sharply falling sales to the printer market. 2GW-plus datacenter opens in western China Chinese company Envision last week announced the commissioning of a two-gigawatt-plus datacenter in Ulana, Inner Mongolia. The company said the 120,000-square-meter facility can house up to one million GPUs and deliver one million PFLOPS of AI computing power. Renewable energy powers the facility, which is located in China’s west – making it an expression of Beijing’s Data East, Compute West plan that calls for construction of large datacenters closer to sources of clean energy and away from large population centers. Cable operator calls for police probe after faults follow “suspicious” activity Australian submarine cable operator and entrepreneur Bevan Slattery on Sunday claimed “some suspicious/coincidental activity” took place at the time two of his cables experienced a fault. “This is a concerning development. Partially because submarine cables are the digital lifeblood of our nation, but also because this could very well turn out to be the second such incidence in a Federally declared cable protection zone,” Slattery wrote. His post includes a marine chart that shows a vessel called the GEM Sapphire, which ship-tracking site VesselFinder lists as a 183-meter long tanker, repeatedly crossing the path of two cables. Infosys to wear Crocs for ten years Indian tech services giant will spend the next decade making sure the world’s ugliest shoes can be produced efficiently, after scoring a deal to help Crocs modernize its systems. Infosys says it will “unify data across the organization and aim to empower Crocs, Inc. with real-time, insight-driven decision-making – accelerating innovation while enhancing efficiency, responsiveness and customer experience.” Apparently, the deal “help Crocs, Inc. address key industry challenges, including rising costs, inconsistent processes and the need for increased agility. The transformation will standardize operations and aim to enable the business to better respond to evolving consumer and market demands.” There’s no mention if the deal will mean anyone looks cool wearing Crocs. Maybe even a decade of help from Infosys can’t make that possible. Australia urges smart glasses privacy probe Australian discount department store Kmart last week started selling low-end smart glasses for just AU$89 ($62/£46). The Guardian reported they sold out swiftly, sparking a privacy debate that quickly saw Australia’s Attorney-General write to the nation’s Privacy Commissioner to ask “that this issue be given priority consideration.” “Privacy is an essential part of all our lives,” said Attorney-General Michelle Rowland, before citing “widespread community concerns about the potential impacts of ‘smart glasses’, particularly for women and children.” ®

Survey Says is a weekly series rounding up the most important polling trends or data points you need to know about, plus a vibe check on a trend that’s driving politics or culture. Amid the 2020 presidential election, a conspiracy theory bubbled up from the depths of the terminally online far right and infected the political mainstream. Its believers made feverish claims about the…

Source

A cartoon by Mike Luckovich. Related | Pentagon out of ideas to end Iran war after Hegseth killed ‘woke’…

Source

ph @casanova_coffee_art
Nave de Vero

The billions of dollars Congress has allocated for foreign aid programs must be spent. One proposal from the Trump administration suggested funding research on crime against Afrikaners in South Africa. By Anna Maria Barry-Jester and Sharon Lerner for ProPublica Since he returned to office, President Donald Trump’s administration has upended foreign aid. It’s labeled well-established…

Source

The AI-focused hedge fund is still making some big bets.

Programming with Claude Code will soon require even less human oversight.

Congressional Cowards is a weekly series highlighting the worst Donald Trump defenders on Capitol Hill, who refuse to criticize him—no matter how disgraceful or lawless his actions. Ohio GOP Rep. Max Miller’s response to horrific domestic violence allegations has been a complete train wreck, with Miller looking worse and worse as he attempts to deny that he was ever violent toward his ex…

Source

A cartoon by Tim Campbell. Related | Trump privately fumes over missile shortages—while lying in public…

Source

In 1964, Lyndon B. Johnson’s campaign ran a terrifying 60-second advertisement. It opens with a three-year-old girl in a field, pulling petals off a daisy while counting to ten. Then, a man’s voice begins a countdown, and a nuclear explosion fills the screen.

The advertisement only aired once, but roughly 100 million people saw it. The message was clear: Johnson’s opponent, Barry Goldwater, was a warmonger who would put American children in danger and bring nuclear armageddon. Johnson won in a landslide, and the daisy video became America’s most infamous attack ad.

Now, a Super PAC acting on behalf of Texas Attorney General Ken Paxton’s Senatorial campaign is using an AI-generated, Frankensteined version of Johnson’s video to push a different message.

In this version, instead of being wiped out by a nuclear bomb, the daisy girl—resurrected via generative AI—lies on an operating table in front of an AI-generated version of Texas Representative James Talarico.

“She dreams of growing up, falling in love, becoming a mother,” a narrator intones. “The trans industry preys on confused children. Irreversible puberty blockers. Cross-sex hormones. Sex-change surgeries.”

It’s not the only AI advertisement Paxton or his allies have produced this cycle, nor is it the first designed to attack Talarico’s support for transgender youth. Another advertisement, which ran in June, shows an AI-generated James Talarico dancing in a dress, singing a parody version of “My Favorite Things” from “The Sound of Music” reworked to be about giving girls hormones “til they grow mustaches.” Paxton has built his brand on ridiculing trans youth and making it as difficult as possible for them to access medical care. In May, he even took time during his runoff victory speech to call Talarico’s defense of trans youth “weird.”

This $10 million ad buy, though, might be the Paxton campaign’s most egregious attack yet. The ad was developed by the American Principles Project, one of the earliest groups to push anti-trans laws. Their ultimate goal, as my colleague Madison Pauly has reported, is to go beyond banning gender-affirming care for children and outlaw medical care for trans people of all ages. As the group’s president, Terry Schilling, admitted last year, he and his allies are starting with children because that’s “where the consensus is.”

The original “daisy girl” ad was effective because it was able to make people afraid. The campaign spot, which aired in the middle of a family television program, pushed parents to vote in the way they believed would protect their children from an unthinkable fate. Schilling and his allies hope to leverage parents’ fears of a supposed predatory medical establishment to push Paxton to victory.

But if previous attempts to remix the daisy video are anything to go by, the American Principles Project may not succeed. Bob Dole, in 1996, aired a version of the daisy video that concerned itself with teenage drug use. Audiences didn’t find that particularly compelling, and Dole lost. The Hillary Clinton campaign, in 2016, tracked down the actress who starred in the original video to shoot an ad for her presidential bid. Clinton lost, too. An atom bomb is terrifying because it is real, but an AI-generated three-year-old on an operating table is a manufactured threat.