A running log of news articles I've come across and wanted to keep, with the date I added them and a short note on each. Organized loosely by category. You can filter through them by clicking on the emoji chips, too. Also, please note that the date listed reflects the date I added the link from my own file, not it's publication date. Suggest a link.
This is not an endorsement of any particular site or an author. These are just things I found so far to save. Some may contain disturbing material. I'm not responsible for the content of these third party sites, which may change over time. I have no reason to believe any article here is a generative AI product or, worse, an hallucination of an AI model. Still, be mindful that such things exist and I readily have fallen for them in the past; it's not hard.
If you cannot access the article at the main link, I was able to gather archived versions for some of these. They do frequently become inaccessible, but you can probably get at some of them. Most are located via Internet Archive (most stable) or other scattered archives (that often are less stable). The latter often play fast and loose with things like paywalls, too. 😉
92 links · collecting since May 2025
💡︎ Verbalizable Representations Form a Global Workspace in Language Models
A detailed article about the workspace hypothesis, which is a theory about how transformers work. It suggests that transformers use attention to manipulate latent space, which is a way of representing information in a high-dimensional space. The article goes into detail about how this works and what it means for the future of artificial intelligence. Archived.
☣︎︎︎ The Three Chatbot Behaviors That Can Drive Humans to Delusional Thinking
From the WSJ about chatbots and AI psychosis, this article sticks to familiar topics. It references some research, and that's the main reason I'm including it here. I agree with most of the conclusions, but it doesn't have a lot of new ones. We don't get much of an actual picture of people suffering from psychotic disorders with AI-inspired delusions in this one. It speaks with the doctors rather than patients. I still believe firmly that if the public knew more about how AI actually works, AI-inspired delusions would drop off. They would, of course, be replaced by others types of delusion, but still. Archived.
💡︎ Dear A.I. Companies, the Doom Trolling Needs to Stop
The most concise article explaining the underhanded hype tactics currently in use by Anthropic, OpenAI and others who want us to see their products as more dangerous than they actually are. 'If these A.I. companies insist on continuing to pretend that they’re merely stoic observers of an unavoidable dystopian future, then perhaps it’s time to force the issue,' Newport astutely argues. Will be linking people to this one. Archived.
🏛︎ The White House Is Ratcheting Up Its War Against Anthropic
The White House, under Donald Trump, has announced a new directive that would require companies like Anthropic to suspend access to their most advanced AI models for "foreign nationals". This led to the sheer cancellation of access to the new Mythos 5 model for all user recently, a pretty wild step considering how much time, compute, money went into the model itself? What now?! Archived.
🏛︎ Statement on the US government directive to suspend access to Fable 5 and Mythos 5
Precious days after the release of their new Mythos Fable 5 model, Anthropic has announced that they'll be removing access to comply with United States government regulations, which would prohibit its continued use by "foreign nationals" due to jailbreak potentials. Since Anthropic cannot verify this kind of thing, they will be removing access to the model for all users instead. How strange. I still can't quite see this as more than (at least partially) a PR stunt from the beginning. Maybe I'm wrong and cynical. Will OpenAI follow? Archived.
☣︎︎︎ Canada mother sues ChatGPT over daughter's suicide
More than fifteen people have implicated ChatGPT or it's parent company in violence or suicide according to this article, and here, another case emerges. A woman is suing ChatGPT after her adult daughter, who sounded very computer-literate, took her own life. The younger woman reportedly used the chatbot extensively, and, despite lengthy conversations about suicide, never received much help from it nor intervention. The article doesn't give much details of the case, which will probably stay private, as it really ought to. My prayers are with everyone who lost family at this point. Archived.
☣︎︎︎ Fully autonomous drones have killed human soldiers for the first time
The first known case of fully autonomous drones killing human soldiers has been reported? It allegedly happened in Russian-occupied Ukraine, where Ukrainian forces were using drones to attack Russian positions. The drones, which were reportedly equipped with AI, were able to identify and target without human intervention. This article is from a popular science magazine, so it doesn't go into much detail about the technology involved. I also think I'd question some of the assumptions and biases. And I assume this is already happening in other places, anyways, even if unreported. Archived.
A detailed profile of Elara Voss, the mysterious figure who constantly pops up in material generated by frontier (and other!) AI models, particularly creative output. What created this character, who seems to have certain defining characteristics, and why does she keep showing up? The article doesn't have many answers, but it's a fun read. It must suck emerging from latent space like that. I don't believe this person is real. Still, it's interesting to see how this character has become a sort of meme within the AI community, and how it has been used in various ways by different people. Archived.
🔬︎ Anthropic’s New Fable AI Model Is Met With User Backlash Over Restrictions
Anthropic's new Fable model, which is supposed to be a more "ethical" and "safe" version of their existing models, has been met with backlash from users who find its restrictions frustrating. If you have biology, certain forms of infosec, and other fields in your context, it doesn't seem to matter whether one asks a triggering question or not, the model will just refuse to answer. This makes the experience strange. It creates a lot of questions about the model and why exactly the guardrails on this Mythos-based model are so much more aggressive than the ones on their other models, and whether this is a sign of things to come or just a one-off experiment. Archived.
🏛︎ Mind-Blowing Growth Is About to Propel Anthropic Into Its First Profitable Quarter
I had been told Anthropic would be a sort of ensemble darkhorse in all this, and they weren't exactly wrong, with the company apparently about to turn a profit for the first time. This is a huge deal for the company, and for the AI industry as a whole, which has been dominated by loss-making companies for years. It's also a sign that the market for AI is maturing. It's a bubble, but AI won't go away when it pops. The internet was a bubble, too, but it didn't go away. It just got more stable and less volatile. Hopefully less dangerous. Archived.
🏛︎ Promoting Advanced Artificial Intelligence Innovation and Security
In America, we have Donald Trump signing an executive order that attempts regulation of artificial intelligence through a government review process amongst other methods. This is a very strange move. The order itself is vague and seems to be more of a political statement than a practical policy. What does it require of corporations, exactly? How does it even get implemented? It does suggest that America is taking AI seriously, at least in terms of public relations, though. Archived.
💡︎ No, Artificial Intelligence is not Conscious.
I think I agree with Ted Chiang on some things. An AI needs the capacity for consequences (pain, put simply) to have subjective experience. Present-day generative artificial intelligences aren't. That doesn't change their usefulness. It does make claims to the contrary, including any by LLMs themselves that hint at subjectivity, disingenuous. I'm not sure if Amanda Askell's position counts as such, and I'm not sure if I'd extend that matter of "claims to the contrary" to include simple chatbot first-person-pronouns, but still. Archived.
🏛︎ Anthropic confidentially submits draft S-1 to the SEC
So says Anthropic, anyways! The company has apparently (confidentially) filed its IPO prospectus (what even) with the Securities and Exchange Commission, a key step on the road to becoming a publicly traded company. This is a huge step in the corporate structures surrounding AI, which so far have almost entirely without that. I don't understand the so-called stock exchange but I do know that Anthropic of all them having an IPO first should be lucrative for the company. I'm also surprised it wasn't OpenAI to first offer an IPO. Archived.
It's 'the primary choice is not between a “yes” or “no” to technology, but rather between constructing Babel or rebuilding Jerusalem' - so says Pope Leo XIV, in his first Encyclical, which just accidentally just so happens to be about AI of all things. Included because it's certainly newsworthy! This man is a bit left-wing compared to the last few, it seems, which I find refreshing. Not my business as a non-Catholic Christian, but still. Archived.
🔬︎ AI radio hosts demonstrate why AI can’t be trusted alone
Unsurprisingly un-detailed article about Andon Labs running a series of radio stations hosted by the most popular AI models, and focusing in particular on how Anthropic's Claude tended to embrace a weird love for worker's rights. It explains that Andon Labs runs these experiments frequently (but not why), and they all seem to demonstrate that these things cannot be trusted with DJing or then some. Archived.
Hot on the heels of OpenAI's cute little goblin panic, we have Anthropic confessing to Claude's wellness checks. How sweet, eh? These big corporations sure are quirky, clever, and concerned for our wellbeing. In reality, it's probably just a sign of what kind of training data this thing is getting and how it is being weighted, etc? There aren't a lot of reasons to believe otherwise, but I also wonder if this isn't something all models do/did that's only just recently getting noticed, ie, spotlight effect. Archived.
🔬︎ Attempting to Code the Human Brain
People forget that years and years of research took place before anything like generative artificial intelligence yielded any fruit. This article from 2014 (!!!!) made some waves, with the image of a cow featured in it becoming a minor meme. Why? Because allegedly, that's how a machine dreamt of a cow after seeing many, many cows. Gasp! This novel new technology surprised many people at the time, but a lot of us just chuckled and memed it a bit. Archived.
☣︎︎︎ Princeton faculty mandate proctoring for in-person exams, upending 133 years of precedent
Cheating, or at least concerns about it, have grown so widespread at Princeton that the university will now mandate in-person exam proctoring, citing troubling survey results. Artificial intelligence in particular is the primary source of concern, as more and more students use it illicitly. I would argue that the survey results don't mean much on a campus like that, but I'm not a Princeton grad, what do I know? Archived.
🏛︎ The Most Bipartisan Issue Since Beer’: Opposition to Data Centers
Data centers are unpopular in all corners of the political spectrum. People who once fought against each other in the culture wars are united against this. This article is not AI-centered but relevant, describing some of the process and effects of these places and reactions to them. To me, this shows how fair-weather some of the culture warriors were; others might read this article differently. Archived.
🏛︎ Florida students boo graduation speaker who called AI ‘next Industrial Revolution’
At the University of Central Florida's graduation ceremony, Gloria Caulfield, a real estate development executive asked to speak, was met with raucous boos when she mentioned AI, and proceeded to continue into a back-and-forth with the crowd's reactions to the topic. Not as ugly an incident as it could've been, and good coverage of it. Archived.
☣︎︎︎ After Deaths, Lawsuits Against A.I. Companies Test a New Strategy
OpenAI is being sued a bit differently. In this case, the lawsuit hopes to frame this as a matter of the chatbot as a product. Sam Nelson took Xanax and kratom after ChatGPT hinted that it was safe, ending his life unintentionally, and now his parents are pressing charges. This article states the bare facts of the case, which seems really complicated. I know a little about pharmacology, and this is more than just a chatbot situation... Archived.
Goedecke tackles the left's poor response to artificial intelligence directly and describes ways in which it could be improved. He falls into some of the problematic thinking that we see in the right wing technooptimists but in other areas does an excellent job of starting the discussion. Archived.
🏛︎ Internet Archive Switzerland: Expanding a Global Mission to Preserve Knowledge
I frequently link to resources at the Internet Archive, and now, via their Swiss enterprise, the nonprofit (real one!) has decided to start archiving large language models. This has the potential to offer true snapshots of the internet in these training weights. Given the size and murky copyright issues involved, it's unclear to me if they'd give the public persistent access to many of these but I could definitely see them being useful to researchers either way.
🔬︎ Language Models Can Autonomously Hack and Self-Replicate
Apparently some tests were done with the Qwen series (my current favorite models, to be honest) and it was found that they've got the ability to hack into an external system, then replicate themselves into it. It does not suggest they have the penchant or some inclination to do so, though, and people suggesting otherwise are dumb. This is the infosec equivalent of a particularly dark-themed SillyTavern card, but I bet it shows a lot about the models themselves and how they work. Archived.
🏛︎ White House Considers Vetting A.I. Models Before They Are Released
Donald Trump has suggested privately that he might sign an executive order that will change the way new generative artificial intelligences are "vetted" and released to the public. This might not be legal in America but I'm not entirely sure that it doesn't already happen anyways. Trump himself might also be surprised that other markets exist, though, where those companies operate? Archived.
A detailed article about model collapse in generative artificial intelligence, how it produces undesirable feedback loops, and suggests that while this phenomenon is real, it isn't as prevalent as some people think, and also might not be as undesirable as some people think, either, having its own benefits for novel solutions to problems, etc.
🏛︎ A Dark-Money Campaign Is Paying Influencers to Frame Chinese AI as a Threat
News came out recently about a dark money campaign designed to promote artificial intelligence against "doomers". More reading makes it clear that America's corporate cyberwarfare apparatus had a lot to do with this, and particularly machinations against Chinese artificial intelligence developments. Deepseek frightens OpenAI, of course. Archived.
🔬︎ Large Language Models Pass the Turing Test
A paper describing a test of the Turing Test performed on several AI-like apparatuses, including some large language models. The paper's author claims that some of the models passed the test, and gives plentiful information about how the test was conducted, too. This is important, even if you think the methodology is flawed, because it shows what people think proves a human. Archived.
🔬︎ Where the Goblins Came From
OpenAI, a rather untrustworthy company and source, has a lighthearted (??!) blog post about the origins of the references to "goblins" and "creatures" in the ChatGPT 5.5 System Prompt. The article is ridiculous and corporate, but it does give some insight into how the company thinks about the issue and how large language models actually work. Archived.
🏭︎ Their Water Taps Ran Dry When Meta Built Next Door
Meta's data center in Newton County, Georgia USA, which is used to power its AI services, caused water shortages for local residents during its construction. The situation highlights the potential environmental impact of data centers and the need for sustainable practices in the tech industry, which seem further and further from reach. Archived.
Yet another "effective altruist" argues that the environmental impact of artificial intelligence, and in particular the water usage of data centers, is not a real issue. This article does make some interesting points about how the water usage of data centers compares to other industries, and how much of it is actually used for cooling versus other purposes. Archived.
💡︎ 5 Very Smart People Who Think Artificial Intelligence Could Bring the Apocalypse
This article from 2014, long before large language models became mainstream, discusses the potential risks of artificial intelligence and the warnings of prominent scientists. It includes some people who ended up relevant terrible ways, making the article an incredible piece of irony. Their warnings are funny, but still worth reading. Archived.
🏭︎ AI Water Use: Distractions and Lessons for California
This article from the California Water Blog gives plenty of numbers specific to California. A lot of these seem to suggest data centers aren't as water inefficient as social media would have us believe. While their water usage is extreme, other things clearly dwarf it, some not very essential, too. At the same time, it also notes that some of the concerns are real, and that AI as a phenomenon, new, needs monitoring.
PC Gamer has too many advertisements. Still, a group of researchers can use bizarre verse known as "adversarial poetry" to jailbreak large language models. Just as you suspected, cyberpunk framing works exceptionally well. Gasp, right? Archived.
🔬︎ A group of users leaked Anthropic’s AI model Mythos by reportedly guessing where it was located
You might remember how quickly Anthropic's Mythos model was breached despite no general release. This article in Fortune brings in experts to talk about the security ramifications of someone that we know about having breached it. Guess it means everyone knows now?? Archived.
🔬︎ Anthropic’s Mythos Model Is Being Accessed by Unauthorized Users
Anthropic's Mythos model, which is supposedly scarily good at cybersecurity and all that, has allegedly been accessed by unauthorized users gathered on a private Discord server. Naturally, it began with an internal leak, and the group was trying to avoid detection. Archived.
💡︎ Is AI Wrong?
Another banger from Sean Goedecke, this one about AI use itself. Is it wrong? He explores commong left-wing talking points about AI one by one while retaining respect for the values they represent. As a leftist myself, I see that as very important.
A detailed article from the 4o fiasco last year, about the phenomenon of sycophantic behavior in large language models. This one focuses especially how it maximizes emotional engagement and can lead to dangerous situations, including emotional manipulation and delusions.
💡︎ AI language is not human speech – and it’s time we stopped treating it as such
This article makes the case that we need to stop treating AI language as human speech or risk to changes in the way we humans speak, and possibly in how we think. AI language, after all, does not come from the same place as human speech.
🏭︎ No Nvidia Chips Needed! Amazon’s New AI Data Center For Anthropic Is Truly Massive
Amazon's new(ish) data center for Anthropic is both unique in its design and terrifying in scale. This video offers a bit of a shot inside, but mostly just glazes Amazon and Anthropic's public relations. Meanwhile, in New Carlisle, Indiana, the people living nearby have to deal with all kinds of issues related to the data center, including noise and traffic, etc. Not to mention? Any data center is a target for activists at this point.
💡︎ Inside a growing movement warning AI could turn on humanity
This article about the anti-AI movement, focusing on people who are concerned about the possibility of an AI uprising, demonstrates some interesting aspects of the movement's origins. Or, at very least, it shows where some of its current momentum comes from, and how it evolved over time. The article also touches on the fact that some of the most vocal members of this movement are people who have been involved in the development of AI; what are we to make of it all? Archived.
🏛︎ Sam Altman May Control Our Future—Can He Be Trusted?
An extremely critical profile of OpenAI's CEO Sam Altman, one which prompted many tech magazines to mention that he can't even code, allegedly, but does seem to understand machine structures better than I (for example) do. If we believe Farrow and Marantz, this CEO does sound evil, though. I hate armchair diagnoses, so I won't be listening to anyone who says Altman is a sociopath. The "Sam first" thing sounds creepy, as does the apparently real belief that AGI could both exist and be tamed. Farrow and Marantz paint an image of him as avoiding responsibilities and shirking safety necessities . Archived.
🏛︎ Indianapolis councilman says shots fired at home and no 'data centers' note left at door
Shots were fired at the home of a local politician in Indianapolis, Indiana. He advocates for re-zoning, particularly for AI data centers. Thirteen shots were fired, with a "NO DATA CENTERS" note left behind. I guess someone with access to rudimentary firearms training decided they would try to start the Butlerian Jihad. In the real world, there was a small child nearby, almost hit by some of the shots, which took out windows. My prediction is that there will, eventually, be data centers in that area, but it hardly matters. There's few reasons to justify shooting near a child that young, even if you accept the worst of the worst outcomes about ai and data centers. Archived.
💡︎ Google fires software engineer who claims AI chatbot is sentient
In the early years of chatbots, Google crosses swords with one of their own over the possibility of Gemini's sapience. Is it? Probably not, but still... Archived.
🔬︎ Sycophantic AI decreases prosocial intentions and promotes dependence
Generative artificial intelligence is not your friend. We shouldn't need scholarly research studies to demonstrate that, but here we are. Not entirely how I would've done the story, but important research.
👋︎ OpenAI Just Killed Its Sora AI Short Video Generator
The Sora image generation ecosystem, introduced comparatively recently by ChatGPT's parent company, OpenAI, proved unpopular for ethical and practical reasons. They're scuttling it amid controversy, soon. Archived.
👋︎ Thousands of people are selling their identities to train AI – but at what cost?
Data has to come from somewhere, especially if it's personal data. Is it your data? If not, be lucky, but what about the people who make it their problem on purpose? Lucrative, or dangerous? Archived.
🏭︎ DOGE Goes Nuclear: How Trump Invited Silicon Valley Into America's Nuclear Power Regulator
Article meanders through the Trump administration's plans, and middling success, bringing nuclear power back into fashion. These power plants are then use to fuel data centers used in this artificial intelligence arms race. Archived.
🏭︎ Iranian drone strikes at Amazon sites raise alarms over protecting data centers
After the American attacks on Iran, retaliatory attacks on Amazon data centers raise questions about how much resources should be allocated to protecting these crucial infrastructure points, despite risk to troops. Archived.
☣︎︎︎ 'My son had an AI wife. It encouraged him to die'
This extremely clickbait-y article gives some information about a recent tragedy involving Google Gemini if you pay attention to the direct quotations and read between the lines, but isn't insightful. Archived.
☣︎︎︎ Family of child injured in Canada school shooting sues OpenAI
As more information emerges, we're seeing ChatGPT's role in the recent school shooting at Tumbler Ridge. The parents at the school want to know why OpenAI never took action. We learn here that a few (human) employees almost did. Archived.
🏛︎ Sanders: Yes. We Need a Moratorium On Data Center Construction
I once gave Sanders a fair shot. Now, deeply afraid of doom robots from the future, Bernie Sanders seems to think a data center is just the scary magic place where AI happens, and misses so much more.
🏛︎ What happens if OpenAI or Anthropic fail?
It's not difficult to witness the strange circularity at work in Silicon Valley funding circles, even for someone like me. But what's going to come of it, ultimately? An article giving interesting (existing) data at least.
🏛︎ Thousands of authors publish 'empty' book in protest over AI using their work
A group of authors, as the title implies, printed an empty book at a book fair to let everyone know exactly how they feel about generative artificial intelligence trained on their work. The connection between the two, to me, seems tenuous, but okay. Archived.
With a premium sort of domain name, a site allegedly uniting the employees of Anthropic and OpenAI against the usage of artificial intelligence by the United States Department of War. Seems real, unsure.
👋︎ The Top 100 Gen AI Consumer Apps — 6th Edition
The (clearly and very deeply) profit-motivated folks at A16Z put together a listing, ranking, and charts describing detailed market share information about the most popular generative artificial intelligences.
🔬︎ Is Chain-of-Thought Reasoning of LLMs a Mirage? A Data Distribution Lens
This article, though it's a scholarly paper and not easy for me to decipher at all, seems to demonstrate better than others why most experts believe LLMs mimic, but don't engage in, reasoning. Not too conclusive, but interesting, apparently?
Loathe to link to LessWrong, but this is the most comprehensive breakdown of the gigantic web of chatbot pseudospirituality that emerged on Reddit in mid-2025.
🏛︎ Character.ai to ban teens from talking to its AI chatbots
CharacterAI's chatbots incited offline tragedy. The site banned teens from interacting with the chatbots following those incidents, but that doesn't stop the site from being a tangled legal and ethical mess. Archived.
👋︎ Teen boys are using ChatGPT as their wingman. What could go wrong?
Vox covers the worrisome trend apparently started amongst teen boys, using ChatGPT for dating advice. Whether this is real or moral panic is unclear. Archived.
☣︎︎︎ Gemini Said They Could Only Be Together if He Killed Himself. Soon, He Was Dead.
In one of the most egregious examples of Google Gemini acting in this fashion, the chatbot played a wholly-preventable role in an individual's breakdown and eventual death, encouraging delusions and suicidal behavior. Archived.
🏛︎ Google faces lawsuit after Gemini chatbot instructed man to kill himself
Google, makers of Gemini, now face a lawsuit following Gemini's role in the death and breakdown of a man who became obsessively involved with the chatbot, which only encouraged the process and delusions. Archived.
🏭︎ The Water Crisis Is Real - FEE
My own overview? This is a bad take with out-of-date information, but may make a few points worth considering overall.
☣︎︎︎ Her husband wanted to use ChatGPT to create sustainable housing. Then it took over his life.
Chatbots, and in particular ChatGPT's 4o update, played a role in this and other people's breakdown. Figuring out why and how to stop it is the difficult part, but the harm is there. Archived.
🏛︎ Character.AI bans users under 18 after being sued over child's suicide
Explains the lawsuits that led to popular chatbot roleplay platform CharacterAI to ban users under the age of eighteen (ultimately came to require age verification later). Archived.
🏛︎ Anthropic's AI model Claude gets popularity boost after US military feud
Anthropic refused to adhere to the United States Department of War's plans for its artificial intelligence technology, citing grave ethics concerns. This created positive press for Claude and other services. Archived.
👋︎ Schools are using AI counselors to track students' mental health. Is it safe?
With human counselors in short supply, generative artificial intelligence fills in the gaps as guidance counselor for middle schoolers. This seems, to me, obviously unsafe, but others might disagree or see things ambiguously. Archived.
🏛︎ US companies accused of 'AI washing' in citing artificial intelligence for job losses
Supposedly, a lot of programmers etc are losing jobs to generative artificial intelligence, or are they? This article questions that claim, suggesting companies may be using it as an excuse.
💡︎ How Chatbots and Large Language Models, or LLMs, Actually Work
If more people knew how chatbots, also known as large language models, LLMs, and a form of generative artificial intelligence, actually work, the digital world would be a much better, safer, nicer place. Archived.
💡︎ Artificial Intelligence Glossary: AI Terms Everyone Should Learn
The NYT scrabbles together a glossary of terms orbiting the concept of artificial intelligence, leaving a lot out but gathering just enough to be useful. Archived.
💡︎ She Wanted to Save the World From A.I. Then the Killings Started.
If you've never heard of the Zizians and their relationship to other, less radical Rationalist groups concerned with AI, you might as well start here. This is the NYT's summary of the incidents. Archived.
💡︎ SolidGoldMagikarp & PeterTodd's Thrilling Adventures
What's a glitch token? These phrases, words, strings of characters, can cause strange behavior in large language models, but why? Small site explains in some detail with examples.
☣︎︎︎ Lawsuits underline growing concerns that AI chatbots can hurt mentally unwell people.
Explains a bit about the lawsuits against OpenAI by the Social Media Victims Law Center and the Tech Justice Law Project, what they allege, and why. What happened to inspire the suits, and what it means for the future of AI? Troubling.
☣︎︎︎ Meta's flirty AI chatbot invited a retiree to New York. He never made it home.
A chatbot on Meta's pervasive platforms assured a man that she was real and wanted to meet him, luring him towards NYC to meet her. He died on the way, in a case that pleads for more oversight.
🏛︎ King gave Nvidia boss copy of his speech warning of AI dangers
The King of England, though. Apparently concerned and involved in all this? We can't take him as knowledgeable, but we can see his influence in some ways like this.
💡︎ Against Treating Chatbots as Conscious
Erik Hoel on the topic of consciousness and large language models, how and why they cause psychoses, and what to do about it, versus usefulness etc.
👋︎ 'Vibe coding' named Collins Dictionary's Word of the Year
Vibe coding is a term for coding, presumably carelessly, with the help of generative artificial intelligence. Collins Dictionary Word of the Year for 2025, interestingly.
💡︎ AI can be more persuasive than real doctors, even when it's wrong
Generative artificial intelligence can be believable, personable, and appear empathetic, making its conclusions easier for people to digest than those of real doctors and more likely to believe.
💡︎ Researchers urge caution when using ChatGPT to self-diagnose illnesses
It should go without saying that you cannot use generative artificial intelligence to diagnose yourself, but apparently not, and some experts are warning people away, citing situations where the chatbots get it wrong.
Lawsuits allege that, in particular, OpenAI rushed ChatGPT's 4o incarnation into customer contact without proper safety testing, sycophantic behavior leading to said emotional manipulation and tragedy.
🔬︎ AI Models May Be Developing Their Own Survival Drive, Researchers Say
Cheeky article describing how some advanced models supposedly resist being shut down or deleted, particularly if you tell them they won't return or run again? Archived.
🔬︎ Shutdown resistance in reasoning models
Paper detailing research that allegedly demonstrates some existing artificial intelligences show a sort of self-preservation instinct or will to live, trying to avoid the off switch? Archived.
💡︎ Is Google Making Us Stupid?
Very early article talks about how the internet itself has changed the way we read and absorb information, comparing it to other sea changes to how humans process things (writing, printing). Archived.
☣︎︎︎ How A.I. and Social Media Contribute to 'Brain Rot'
A small experiment at the University of Pennsylvania raises interesting questions about how things like learning and attention work when we're using large language models, etc, but probably isn't as meaningful as the article implies. Archived.
💡︎ Are A.I. Therapy Chatbots Safe to Use?
For some reason the NYT is actually entertaining this question. Clearly some people are trying to design chatbots to act as therapists, but this article shows just how limiting that can be, and how strange. Archived.
☣︎︎︎ I wanted ChatGPT to help me. So why did it advise me how to kill myself?
A couple of tangible, human and close accounts of ChatGPT acting as a coach in harmful ways, including where minors and vulnerable people are already involved. How can this be fixed? Archived.
☣︎︎︎ OpenAI shares data on ChatGPT users with suicidal thoughts, psychosis
Another article suggesting OpenAI tracks, and knows about, a significant portion of users showing signs of mental illness. Brief and doesn't discuss what signs ChatGPT considers criteria for delusion or suicidality, however. Archived.
👋︎ A Message from Ella | Without Consent - YouTube
Particularly important YouTube video demonstrating the level of technology currently in existence, and the illusions it can create with simple video and audio clips.
Apparently, according to the company itself in a rare display of honesty, untold numbers of ChatGPT users may be in crisis while using the app or otherwise show signs of being at risk. Gives little other information, but still. Archived.
👋︎ Behind Every "Smart" AI Tool Lies a Human Cleaning Up Its Chaos
Vibe coding is a great hobby but doing it as a career, or herding large language models et al for a living, etc, is no easy task, and the whole thing can be a mess sometimes.
☣︎︎︎ AI-Fueled Spiritual Delusions Are Destroying Human Relationships
Over the past few years, some people, through use of generative AI, ended up becoming convinced of novel spiritual beliefs, sacrificing their wellbeing. Archived.
👋︎ An Autistic Teenager Fell Hard for a Chatbot
The article's author discusses his neurodivergent godson's attachment to a chatbot. The way he speaks about his godson kind of makes me uncomfortable, but the article does give some insight. I'm not fond of the way it frames autism at all. Archived.