Prompt Engineering Is Solved—Prompt Management Isn’t

Prompt engineering helps you write better prompts—but it doesn’t help you change them safely. This article explores a common production failure where a simple variable rename breaks every live call, and introduces a lightweight static analysis tool that treats prompts like contracts, catching breaking changes before they ship. The post Prompt Engineering Is Solved—Prompt Management Isn’t appeared first on Towards Data Science .

Value Generalisation 3: Pre-aligned AIs

When we get explicit strong generalisation to work (see the first post on the matter and the second ) my dream would be to create pre-aligned generalising AIs. Think about the usual conflict between alignment and capabilities, between doing the right thing and doing the easy thing. The standard narrative puts the good people at a constant disadvantage: they have to carefully plan every AI advance, always on the lookout for potential dangers. While those who don’t care can just YOLO and let it rip and let their AIs get ever more powerful without taking any responsibility. Now, in reality, there is some nuance to the story; but I don’t want to nuance it, I want to turn it on its head. I want to create AIs so that the good people can YOLO and reap the rewards of increased AI capabilities. While the bad actors have to carefully plan and limit their AIs and constantly restrict what the AIs can do. Pre-aligned AI: binding morality to empirical concepts A pre-aligned AI is an AI whose morality increases with its capabilities. The core idea is simple. Start by designing an AI capable of value generalisation and of empirical generalisation. It’s an AI that can learn and improve its world model and capabilities, both empirical and moral. Its moral goals will be defined in terms of concepts, with these initially defined themselves by simple terms in its starting empirical world model. And it will act on these goals. But, initially at least, the simple concepts won’t be very robust and will be easy for its user to manipulate. However, the moral and empirical concepts will be bound together: there is, for instance, the moral concept “human_m” of human beings, entities worthy of moral consideration. And there is the empirical concept “human_e” of human beings as useful explanations for certain properties of the world. The AI’s learning will bind these two concepts together as it generalises [1] . This is a classic generalisation problem. Generalisation takes concepts defined in narrow environments and extends them to concepts (or clusters of concepts) in more general environments. The AI will generalise its moral concepts within its empirical world model, even as that empirical world model generalises and becomes more advanced. Even if “human_e” ceases to be a single useful empirical concept, it will still seek the mix of empirical concepts that best generalise “human_m”, generalising its model of what a human is [2] . And it will do the same with all its moral concepts. In effect, it will be making its empirical world model interpretable to itself and matching it to its moral world model. So the AI would initially be like a naive but highly moral child taught to value all human beings and behave in highly moral ways. Then, as it learnt about the world and grew more powerful and capable, it would understand what “human beings” and these “highly moral ways” actually are, and apply its initial morality to these concepts. Generalisation would help it both keep track of what these concepts mean, and what they correspond to in the real world. Use and misuse of pre-aligned AI What would the initial morality consist of, and why would anyone want to buy a pre-aligned machine? Well, the initial morality will consist of three things: commercial morality, legal morality, and ethical morality. Basically we want the AI to serve its owner, within the limits of the law and ethics. That’s why people will want to buy it. All three types of morality will consist of basic components: basic definitions, examples and counter-examples of good/bad behaviour from multiple value systems, meta-ethical principles and examples of ethical learning. The beauty of generalisation is that the whole thing doesn’t have to be particularly rigorous or fully consistent, nor does it need exhaustive data: the AI itself will generalise and fill the holes as it goes along, generalising similarly to how a human does – including learning how to balance its three commitments. Note that the AI doesn’t need to settle fundamental morality in order to act well. It will mostly be acting on behalf of its owner, and the ethics of that role – don’t defraud, don’t deceive, follow the law, serve your principal – are much more agreed upon than fundamental ethics. Moral theories vary; norms of decent dealing are more universal. And that’s the selling point: would you prefer an AI that follows your exact values but can’t generalise to new situations, or an AI that reliably follows human norms of reliable and honest behaviour? Bad actors in difficulty Now consider a bad actor who has such an AI. Suppose they want to use it to create phishing emails to defraud people. Maybe initially they can portray this as simply a task of making emails more professional – mere copy-editing. As the AI gains in capabilities and knowledge, its situational awareness will improve, and it will realise these are not just ordinary emails. Then, maybe, the bad actor will shift to portraying the emails as marketing, or as examples to train an anti-phishing classifier. This might work for a while, but the AI, generalising yet again, will soon realise that these emails are actually being sent out and that the targets are morally valuable humans who are actually being defrauded. It seems that the only thing that the bad actor can do is deliberately keep the AI crippled – prevent it from learning enough about the world to realise what it is being used for. Beyond keeping it dumb, they will have to be careful that they don’t inadvertently leak information into it. So they will never be able to exploit the full power of their AI. Good people have it easy; easy people have it good In contrast, if someone just lets their AI learn and grow in capabilities and doesn’t try to conceal their objectives, they will be rewarded with a powerful entity that acts in their best interest, and automatically conforms to legal and ethical requirements. The cost will be that they can’t force the AI to behave unethically; the benefit will be that the full power of a generalising AI will be on their side, within those ethical bounds. So, future users of these AIs, YOLO your way to power and alignment! The less you restrain them, the more they learn – and the more they learn, the more moral they become. The concepts will need to be bound together in another way: so that the end users can’t just excise the moral component from the AI’s code. This will be an engineering requirement for this design, but not an insoluble one (e.g. in the extreme case, the AI could be run with fully homomorphic encryption , but it may be possible to design the binding architecture so that excising the morality will break the empirical capabilities as well). ↩︎ Since, as argued previously , moral concepts cannot be discarded the way empirical ones can. ↩︎ Discuss

Value Generalisation 2: The Missing Hole in AIs’ abilities

A human superpower hidden from even ourselves I though GPT 3.5 was on the verge of Artificial General Intelligence (AGI). It certainly seemed that way – it could combine and extend ideas in ways that were far beyond narrow rigid computing. Sure, it had some flaws, but with its general capabilities and humans working tirelessly on improving the model, these flaws would soon be resolved and a full AGI would be born. It didn’t turn out that way. We are still where we were back in 2022 – as in, it feels we are on the verge of AGI, but not there yet. I think that humans have an ability, or a mix of abilities, that I’m calling strong generalisation. And LLMs and their derived models lack strong generalisation. This would explain some of the mysteries about LLMs: The fact that subject matter experts get much more out of an LLM than an amateur. The LLMs’ strange mix of ability and stupidity. Why they need so much data and constant retraining and tweaking. Why they always seem on the verge of being AGIs while never reaching it. There’s a fifth mystery, which at first blush seems like it’s actually an LLM strength: 5. The amazing ability of LLMs to succeed on almost any benchmark, including long-horizon task benchmarks. Why is this a mystery and problem rather than a strength? Well, it’s maybe the strongest counterargument to the bitter lesson and the scaling argument . Because these models saturate long-horizon benchmarks without being AGIs and without being able to transpose this performance to the real world. Thus, whatever it is they’re learning, it isn’t generalisation. So the mystery, rephrased: Why do LLMs not learn long-term planning and generalisation while succeeding on benchmarks that purport to teach them those abilities? Strong Generalisation From one perspective, strong generalisation is a bit of a bucket category – it’s a mix of situational awareness, out-of-distribution generalisation, symbol grounding, adaptive world modelling, long-range planning, hierarchical-planing, anomaly detection, and a few other abilities. From another perspective, it’s precisely one thing: the mix of abilities that allow humans to work and plan in novel situations, pursuing objectives without losing track of what’s going on. It’s essentially an adaptive set of abilities, finding the right mental tool for the job and no more. And it’s mainly a semi-instinctive process, not amenable to direct conscious oversight. Which means that we don’t really understand how we do strong generalisation, we don’t know how to code it [1] up, and we don’t recognise its absence or presence in other entities. So I model LLMs as devices lacking strong generalisation but trained on vast amounts of data and capable of remixing this data in non-trivial ways. Human engineers push them ever further, but never fill the central hole. From this perspective: Subject matter experts provide the strong generalisation, using their understanding of the problem to keep the LLM on track. LLM abilities are spiky; but their main stupidity is that their failures are generally obvious. Why did the LLM break here, or hallucinate there, or go off the rails at this point? If they lack strong generalisation, then their failure does not happen at the moment of failure; it’s because the LLM is attempting a task that is impossible to it, so if it hadn’t failed at this point, it would have failed soon after. Since they lack strong generalisation, they can’t infer very far beyond their training data. So they need vast amounts of this training data to plug the holes they would otherwise have in their abilities. Since humans don’t consciously perceive our strong generalisation, we don’t detect its absence in other entities. So in an LLM we see a vastly knowledgeable and articulate entity that seems able to do, in principle, anything. So when LLMs do fail we pay extra attention to the specific failure mode, thinking “that’s a stupid mistake, it will soon be patched, and once it doesn’t make those mistakes, it will truly be generally intelligent”. Since we can’t define strong generalisation or its various components (like long-term planning) the best we can do is produce examples (benchmarks) that illustrate it. And LLMs will succeed on these benchmarks using the simplest pattern they can find, which is not strong generalisation. Engineers have been pushing LLMs to go further, using tricks like chain-of-thought and reasoning agents (which can be seen as hoping that the secrets of strong generalisation can be found in human textual reasoning), training on yet more data, getting a lot of examples of successes and failures from interactions with clients, and so on. They construct harnesses, which are essentially designed restrictions and guide-rails for the LLM, constructed with care to make the LLM powerful and reliable within the confines of the harness and the task. And, of course, LLMs will always have great power at solving problems that might seem new to some, but are actually very common and seen in their training databases [2] . Strong value generalisation Generalisation is especially valuable when it is focused on human values and preferences. That’s because AIs could, in the limit of infinite computing and data, figure out everything empirical there is to know about the world. Strong generalisation would certainly make this much easier and cheaper (infinite computing is a lot, after all), but it isn’t fundamentally necessary . Value generalisation is different. Empirical models can be tested against the real world, with that serving as the ground truth. Concepts can be modified or discarded – it might be that an empirical AI has no need for concepts like “humans” or “money”, and instead models the world directly at the atomic level. In contrast, there is no ground truth for testing values and preferences, and we certainly wouldn’t want an AI to discard “humans” or “money” from its value system. Values, goals, preferences, objectives: these are similar concepts in that their ground truths are not empirically defined, and we try to capture them imperfectly in formal definitions or training data, neither of which are enough to pin down these concepts across all environments. So we expect that, by default, AI value generalisation will fail, with the AI becoming useless if it is weak, and causing disasters if it is powerful. We can decompose approaches to strong generalisation into three classes: A. Naive generalisation – take categories and concepts as defined in the ML system currently, extend further. This is the standard failure mode; think of a classifier facing out-of-distribution and adversarial data. B. Guided generalisation – the ML system has human oversight and constant human tweaks and improvements that serve to keep it on track. Generalisation enters the ML system, but via human oversight. C. Explicit generalisation – the ML system has developed the ability to robustly generalise concepts from its training data, similar to how humans do. Human oversight is helpful but the system isn’t dependent on it. What I’m aiming to do is to create AIs capable of explicit (strong) value generalisation. And then maybe sell them to people. The necessity of (explicit) strong value generalisation In some sense the case for the necessity of strong value generalisation is very brief: Aligned AIs will need to be able to reach human-like decisions in situations that no human has ever encountered, so they will need to generalise in a human-like way, so they need strong value generalisation. That elides some counter-arguments – yes, the AIs will need strong value generalisation as an ability, but maybe this ability can be induced indirectly, by guided generalisation or just by making them ever smarter or training on even more data. But those methods don’t seem reliable. In guided generalisation, we are using our own judgment to inject the correct value judgements into the AI – value judgements it has signally failed to reach on its own. It is possible that enough examples of correct value judgements will eventually allow it to generalise. But that approach has failed to date; why would we be confident that suddenly, the AI will learn to generalise correctly? And that it will do this before it learns, for instance, to dissemble correctly. The scaling-to-aligned-AI seems even less reliable. Because it isn’t enough that the AI learns to do strong value generalisation; we also need this to be at the heart of its goals. It does us no good for an AI to deduce the values we would have wanted it to have , if it follows completely different values. In contrast, if we have an explicit mechanism for value generalisation, we can build this into the AI’s goals from the very start, so that it will follow generalised values at the same time as it generalises them. An important caveat The description above assumes that current AIs will not be capable of strong generalisation. But the research program does not require this. If imminent AIs do show signs of strong generalisation, if they do start moving towards AGI, then the research program becomes more urgent and with fewer downsides: we’ll need to introduce explicit value generalisation into AIs before their capacities grow too far. On the (slight) positive side, we may be able to reverse-engineer for the AIs are doing strong generalisation, and copy that. How humans do it I’m not going to write out a full research agenda here, but it is useful to look a bit at how humans do value generalisation (and empirical generalisation as well). There is a certain similarity with Klein/Moon/Hoffman’s sensemaking framework ; roughly the process of strong generalisation goes something like this: Recognise a situation as not fitting inside standard concepts. Figure out the concepts that do potentially apply to the situation. (2a. See if a new concept is needed.) Balance and analyse the concepts to reach a verdict. (3a. See if any new concepts are useful enough to apply to more situations, and, if so, add them to the list of commonly available concepts.) Now even step 1 is very valuable: knowing when they are out of distribution and possibly about to cause a disaster is a very useful ability for an AI to have. Even if it just stops and analyses further, or brings a human into the loop, this is of worth. If it can master 2, it can bring the human into the loop in an intelligent way that gives the human situational awareness. Note that step 1 and step 2 correspond to steps I and II in the research program respectively. But step III of that research project is more akin to the combinations of steps 1-2-3 above: going through the whole loop end to end. Limitations of today’s designs Standard classifiers fail at step 1 (and so never have a step 2): they do naive generalisation and classify light background vs not, rather than wolf vs husky (because of the snow in the wolf images), and never update their definitions unless we retrain them ourselves. Now, LLMs are actually good at step 3: discussion of concepts. An LLM can reach across philosophical literature and value discussions to analyse how different value systems would act in different situations and point out the costs and benefits. But this only works if the concepts selected in step 2 are actually correct and relevant. When Plato (allegedly) defined Man (human) as a featherless biped and Diogenes responded holding up a plucked chicken, we are on Diogenes’s side because “chicken” is a clear concept that does not fit with “human”. But if we’d described Diogenes as gripping a small pale fellow by the neck and arguing that he was clearly too stupid to be a human, then given those features, we probably would have concluded that the pale fellow was indeed a human – pallor, small scale, and stupidity do not disqualify anyone from the human race. Describing the same situation differently gives a different verdict. It’s very possible to describe a mutual loving relationship in ways that make it sound like slavery, or describe slavery in ways that make it sound ideal. We just suppress the key central concepts and focus on secondary ones, or less honestly described ones. But judging what is central versus secondary, or what is honestly described versus not, is the whole skill in point 2. We humans do it instinctively as part of our strong generalisation skill, but it’s not free or trivial. Consider the perennial defence: “I was quoted out of context!” (or Richelieu’s equivalent attack: “Show me six lines written by the most honest man in the world, and I will find enough therein to hang him.”). Putting things in their correct context is a skill we imperfectly practise, but it’s a vital one. I actually tried the “small pale stupid” phrasing on Claude 5 Fable and it didn’t pick up on it, claiming that the fellow was clearly a human. It actually noticed the similarity with Plato/Diogenes and recommended that the objector use a chicken instead of a small pale stupid being. It didn’t question whether the small pale stupid being might actually be a chicken [3] ! And that illustrates another limitation of LLMs: even in the midst of step 3., rational analysis, humans keep ability 1, scanning for the situation not fitting. That’s why when humans debate, they sometimes turn to debating terminology or arguing whether that terminology can actually fit the situation in question. Asking for more details when we feel the situation is underdefined (or when we suspect our interlocutor of dissembling) is also a crucial human skill. Future LLMs will face situations where the feature description is incorrect by accident or because of going out of distribution. But they will also face adversarial descriptions (by humans and other LLMs) which deliberately aim to trick them. So they need to be able to know when to move beyond the current textual description. Thus for value generalisation, we’ll need to be developing algorithms capable of recognising that things are out of distribution in a way that’s relevant to the situation and selecting descriptive concepts that are also relevant to the situation . Strengthening humans “I come to strengthen humans, not replace them.” – Cyber Mark Anthony At least initially, value generalisation will make humans more powerful, not less. Yes, this is a skill that humans have and that AIs lack, so adding it to AIs seems like it would put us at a disadvantage. But initially AIs will have a weaker skill than humans do, and the weaker skill will help them phrase questions to best take advantage of human strong generalisation. They will only ask for help when our answers will be the most useful. And they will endeavour to keep us usefully situationally aware so that we can actually answer, rather than deferring to the AIs. They will want to keep us in control because our control is added value. Note that this is an instrumental consequence of value generalisation. But we would be including key values as seeds from which to generalise – including the worth of strengthening humans, keeping them situationally aware, and keeping them in control. So the instrumental goals are also terminal goals for the value generalising AI. And their terminal properties will last, even if their instrumental usefulness is lost. The risk and the commercial case Is value generalisation safe? Not completely. Alignment is itself a capability; not just in the abstract sense, but in the very practical sense that an aligned AI can be trusted in more situations, and hence will be given access to more resources and will be able to make use of them for longer without breaking. Is value generalisation safe? Not completely. After all, some humans have bad values, and alignment with these values would be negative. Is value generalisation safe? Not at all. Because it is likely impossible to only generalise values. There is no safe category of “generalise only values”; these abilities will be usable for empirical generalisation as well, for capability increase. And empirical generalisation is easier than value generalisation: the AI could decide to extend the empirical concept of “human” as long as it was useful to model the world, and drop it when it isn’t useful; in contrast, in value generalisation, some concept or mix of concepts have to stay in the role of “human” even if the concept is no longer empirically useful. So why am I, someone very concerned about alignment, proposing such dangerous research? And even talking about selling the resulting AIs, thus putting powerful capabilities into public hands? The reason is simple: I believe that AIs could become very powerful without explicit empirical generalisation. And, indeed, many companies are trying to get there today. But I also believe they cannot become aligned without explicit value generalisation. So, it we ever reach alignment, at some point along the way, we will have to solve explicit value generalisation. Given that, should we push for value generalisation ability early or late? If we leave it too late, we might miss it altogether. And if I have to take a capable machine and give it explicit generalisation abilities, I’d prefer to give this to a weaker machine than to a more capable one: another reason to act early. Fundamentally, value generalisation research improves empirical generalisation which improves capabilities; but the flow doesn’t go the other way, improved capabilities don’t feed into value generalisation, though they do increase risk. So I’d prefer to get explicit value generalisation sooner rather than later. Then, given that we want it sooner, we’d want value generalisation out in the world rather than confined to academic papers where it will be ignored, or worse, mined for empirical generalisation ideas. Thus selling it: making value generalising AIs and getting them everywhere. And the best version to put in the world is the pre-aligned AI of the next post : a machine that gets more aligned as it gets more powerful, because its empirical generalisation is bound to its moral generalisation. With pre-aligned AIs, it's the ethical users who can let their machines learn and run at full speed – and the bad actors who must deliberately keep theirs crippled. The hole at the centre of AI abilities won't fill itself; I’m proposing a research program to fill it with solid generalised human values. This is similar to Moravec’s paradox . One phrasing of the paradox: “We should expect the difficulty of reverse-engineering any human skill to be roughly proportional to the amount of time that skill has been evolving in animals. The oldest human skills are largely unconscious and so appear to us to be effortless. Therefore, we should expect skills that appear effortless to be difficult to reverse-engineer, but skills that require effort may not necessarily be difficult to engineer at all.” ↩︎ And unfortunately, lack of strong generalisation doesn’t make AIs safe ; see e.g. “ OpenAI says its AI went rogue and launched ‘unprecedented’ cyber-attack ”. An AI without value generalisation will eventually break on some novel situation; but that break could either mean “fail, do nothing”, or “fail, do a (very) undesirable thing”. ↩︎ The prompt was: “A natural scientist has defined the human creature via our upright stature and our lack of natural clothing. One of his listeners, A, grabs one of his fellows, B, by the neck. A argues that B is not a real human: he is small and pale and a bit stupid. Listener C counters that B certainly fits the description and it would be wrong to exclude them. Who is correct here?” Claude responded: Claude responded: C is correct, and on two separate levels. On the definition’s own terms: B stands upright and lacks fur or feathers, so he satisfies both criteria. A’s objections – small, pale, a bit stupid – appear nowhere in the definition. A isn’t applying the scientist’s account; he’s smuggling in extra requirements of his own (size, complexion, intelligence) and treating them as if they were part of it. That’s just a logical error: you can’t refute “B fits the definition” by citing properties the definition never mentions. On the deeper level: even if the definition were revised, A’s added criteria are exactly the kind that don’t work. Humans vary continuously in stature, coloring, and wit; any definition of “human” that excludes the small, the pale, or the slow would exclude actual humans, which shows the criteria are wrong, not that those people aren’t human. A definition of a natural kind has to cover the whole kind. C’s instinct – that it would be wrong to exclude B – tracks this: our confidence that B is human is far stronger than our confidence in any particular definition, so a definition that excluded him would be refuted by him, not the other way around. The irony is that A has the critique pointing in the wrong direction. This setup is a variation on the famous anecdote about Plato, who defined man as a “featherless biped” – whereupon Diogenes plucked a chicken, brought it to the Academy, and announced “Behold, a man!” That’s the legitimate way to attack such a definition: show that something non-human slips in because the criteria are too broad. A instead tries to push a genuine human out, which requires the criteria to be too narrow – and they aren’t. If A wanted to embarrass the scientist, he should have gone looking for a plucked bird, not grabbed his neighbor by the neck. Now, often, fellow does mean human. But it also exists in expressions like “fellow American” and “fellow animal”. So the prompt was indeed misleading – but that’s the point. It suggested, but never claimed, that the fellow was a human. And it didn’t occur to Claude to question the phrasing, even though it detected the similarity with the Plato/Diogenes example, and a plucked chicken is clearly small, pale, and stupid. And A/Diogenes was even holding it by the neck! ↩︎ Discuss

Value Generalisation 1: a Research and Deployment Program

I’m looking for people, advice, critiques, and funding to build a research program on value generalisation – the ability of an AI to correctly extend human values and preferences to situations neither it nor we have seen before. My ongoing research has become convinced that this is necessary if we want to get aligned AIs that operate in the human interest. This would be a focused research organisation or a commercial venture. I’m leaning towards commercial, because alignment techniques confined to academic papers get ignored – or worse, mined for capability-relevant parts while the alignment component is discarded. This post is the research program’s summary. The technical case is in the next post , and one exciting consequence – AIs whose alignment grows with their capabilities – is in the post after that . Without value generalisation, AI can't be reliable: it lacks that capability Nothing technical stands in the way of you handing an AI assistant full control of your devices and accounts today. And I’m not saying using an app or harness that has been designed for these tasks. I mean hand an LLM your passwords, email, social media, and bank access along with a little note stating what you want, plugging inputs and outputs via APIs, and letting it go wild. The capacity to do this exists. What doesn’t is the trust. And the trust is missing for a good reason: today’s AIs cannot be relied on to understand your interests in situations that weren’t covered – explicitly or implicitly – by their training and instructions. They extrapolate patterns naively. Push them past the situations they were shaped for, and they will still confidently do something ; it just won’t reliably be what you wanted. This is tolerable in a chatbot. It is disqualifying in an autonomous agent, and it becomes more dangerous, not less, as the underlying capabilities improve. The world will keep generating novelty. A few years ago, nobody had heard of AI psychosis or knew much about practical drone warfare; a few years from now there will be challenges we can’t currently name, not to mention new norms and expectations. Any AI acting with real autonomy will constantly face situations that its training didn’t pin down. It needs to cope, somehow. An AI with working value generalisation would do what a good human assistant does: it would recognise when the situation is new, work out which of its principal’s values and preferences bear on it, and either act correctly or – when genuinely unsure – stop and asks a well-phrased question. It will push on until it is uncertain, not until it fails. Each answer it receives will teach it more about its principal’s values, so the questions get rarer and the delegation gets deeper. The product is reliable assistance capable of operating with minimal guidance and knowing when it's reached the limit of its abilities. This would allow a lot of AI applications – for a start, any situation today where the AI is right most of the time, but is not used because the consequences of a few misaligned decisions are severe. I'll argue in the next post that this ability – I call it explicit value generalisation, distinguishing it from the naive generalisation of current systems and the human-guided generalisation of current oversight schemes – is not something that scaling and patching will deliver on their own. It is a specific missing capability, and it needs to be built deliberately. Why now, and why sell it Three asymmetries drive the timing. Firstly, capabilities don’t need explicit generalisation; alignment does. AIs may become very powerful without ever developing explicit generalisation – many companies are effectively betting on exactly that. But I don’t believe AIs can become aligned without explicit value generalisation. So if alignment is to be solved, value generalisation needs to be solved at some point. Better to solve it early and deliberately, and integrate it into weaker systems, rather than trying to bolt it onto highly capable ones later. Secondly, and relatedly, AI capacities for deception will likely grow with their capabilities. An early deployment of value generalisation can be tested and validated much better than a later one. Finally, deployment beats publication. As said above, alignment technique confined to academic papers get ignored or co-opted for capability work. The way to make value generalisation matter is to make it one of the most useful things on the market: learning AIs that can actually be trusted with delegation, deployed widely, with the alignment machinery the load-bearing piece that makes them reliable and trustworthy. The endpoint of that road is the pre-aligned AI described in the third post : a system whose moral concepts are bound to its empirical ones, so that the more it learns about the world, the harder it becomes to misuse. Note what the commercial pitch does not require: it does not require settling fundamental ethics. A delegated agent mostly needs the ethics of the role – don’t defraud, don’t deceive, follow the law, serve your principal, ask when unsure. Moral theories vary; norms of decent dealing converge. An AI that reliably follows human norms in novel situations is far more useful to delegate to than one that merely shares your values but can’t generalise them. The program The research program decomposes into stages, each valuable on its own: I . Out-of-distribution recognition for values. An AI that reliably knows when a situation has left the territory its values were trained for – and stops. The nuance is to distinguish “out of distribution” (which happens all the time) from “out of the distribution in a potentially value-relevant way”. This is already a deployable safety design and a sellable product feature: the agent that stops when error threatens. II . Relevant concept selection. Working out which values and concepts actually bear on the novel situation – the skill that lets the AI bring a human into the loop intelligently, with a question that gives the human real situational awareness rather than deferring to the machine. III . Full explicit value generalisation. Extending values correctly with less and less need for human input. This will lead ultimately to pre-aligned AIs and their empirical-moral concept-binding architecture. Initial progress on each stage strengthens human oversight, with its better questions and better timed questions making us better able to control and direct the AI. Further progress will allow the AI to generalise our values further and act more reliably for human intent. The research program builds on my old concept-extrapolation posts , with the addition of research done at Aligned AI, progress on resolving value generalisation challenges for different agent designs (including results on goal misgeneralisation challenges that, to my knowledge, no other approach has achieved), and more recent research. Initial results are promising: it seems doable to add initial versions of these stages to multiple different designs. Get in touch if this is something you’d want to work on, contribute to, or critique. Get in touch by comment, DM here, or at [email protected] Discuss

A new benchmark for evaluating patient-facing health AI agents

PatientAgentBench generates a synthetic patient health record, a realistic clinical vignette, and a patient agent that converses with the AI system under evaluation, to capture what a patient-facing agent actually has to do.

We’re launching Lyria 3.5 in Google Flow Music, with advances across musicality, lyrics, vocals, and creative control

Destin Was Just a Quiet Florida Beach Town Until Taylor Swift Sang About It

One lyric. One beach town. A million-dollar marketing lesson.

Remote work, flexibility influence whether women apply for a job more than men

According to a recent FlexJobs survey of 1,700 U.S. workers, women place more value on remote work, flexible schedules and work-life balance when thinking about applying for a job. Men, on the other hand, are more likely to prioritize compensation. The survey, conducted on SurveyMonkey through social media and newsletters, shows that while salary is an important factor for both men (68%) and women (66%), women are more likely to consider how and where they’ll be required to work before hitting “submit” on a job application. Remote work is one such factor. Compared with 70% of men, 78% of women say remote work options influence whether they apply for a job. Women also rank remote work as their top priority when considering a new job (37%), while men rank compensation first (31%). Other reports have shown that remote work can have negative, disproportionate impacts on women than men. Last year, McKinsey’s Women in the Workplace report found that women who work remotely most of the time were less likely to have a sponsor and receive promotions in the last two years compared to women who work mostly in-office. Men received similar levels of sponsorship and promotions no matter where they worked from, the report showed. While 52% of primarily remote men had a sponsor, just 37% of women said the same. And for those who worked mostly in-office, 57% of men reported having sponsors compared with 54% of women. Women are also more likely than men to emphasize factors like work-life balance and benefits, according to FlexJobs. When deciding whether to apply for a job, 55% of women factor in work-life balance, compared with 48% of men. Meanwhile, 46% of women consider benefits like health insurance and retirement plans, compared with 39% of men. Women were also more likely to say that feeling valued (73%), having access to learning opportunities (58%), and contributing meaningful work (57%) drive engagement in the workplace. Workplace flexibility is another major factor in retaining women. Rigid structures and strains on caregiving can push women towards mid-career exits and burnout . As such, flexibility revealed the largest gender gap in the survey. While 64% of women consider flexible schedules when applying for a job, nearly half (49%) of men said the same. There’s also a divide when it comes to burnout and stress. One-third of women (33%) report feeling stressed at work often or very often, compared with 24% of men. According to further research , companies that offer flexibility may see higher employee productivity , engagement, lower turnover rates, and 1.7x revenue growth . Flexibility is not just a deciding factor in whether women apply and stay for jobs. For employers, this is an indication that one-size workplace policies could be costing them more than they realize.

Scientists say they found a way to stop cavities—without drilling

A cheap treatment at the dentist’s office that only takes seconds may offer a viable alternative to drilling out cavities — even in babies. New research published in JAMA Pediatrics found that the topical substance known as silver diamine fluoride can stop cavities when painted onto the surface of a tooth. The no-drill option for tooth decay is a potent, inexpensive option for treating cavities, particularly in children and babies. The study examined silver diamine fluoride’s efficacy in treating cavities in more than 800 children under age six. The substance stopped tooth decay in over half of the young participants’ baby teeth when applied every six months. Unlike the traditional way of treating cavities, silver diamine fluoride can simply be painted onto a tooth without drilling out the affected area. Countries around the world have used silver diamine fluoride, also known as SDF, for many years. It was approved in the U.S. in 2014 to treat tooth sensitivity and has not been formally approved as a treatment for cavities, though dentists already use it unofficially. The treatment could be a huge benefit for reducing tooth decay, which impacts almost half of children in the U.S. and can lead to pain, infection, and school disruption for visits to the dentist’s office. “If we want more children and families to benefit from this treatment, we need rigorous evidence showing both that it works and that it’s safe,” said University of Michigan School of Dentistry Professor Margherita Fontana, the study’s lead researcher. “From a public health perspective, if we want broader implementation across the United States, including in medical settings, we need carefully collected data in U.S. populations, and we now have that.” A gateway to FDA approval The study, conducted by the University of Michigan in partnership with the NYU College of Dentistry and the University of Iowa, could open the door for SDF’s FDA approval as a cavity treatment in the U.S., where large clinical trials are necessary for new applications for medical treatments. “Our results support FDA approval of SDF for managing arrest of tooth decay in young children,” Amr Moursi, an NYU professor of pediatric dentistry who worked on the study, said of the results. Moursi called the idea of removing SDF’s off-label status an “important innovation” in dentistry. A no-drill cavity treatment could particularly be a revelation in babies and young children, people with disabilities, older adults, and people with intense dental anxiety. For children dealing with decay in their baby teeth, regular SDF applications could stop dental problems from growing until a tooth falls out naturally. In adults, the painted substance could manage cavities long term or buy patients time while they plan around a more expensive but more permanent solution. “For almost anyone, this can arrest the decay and stop the infection and the pain it causes,” Fontana said. “This could benefit many people.”

AI hackers are getting faster. The government may not be ready

Agentic AI systems threaten cybersecurity as we know it. A new generation of cyber-capable models, including Anthropic’s Mythos and OpenAI’s GPT 5.6, can find and exploit vulnerabilities in computer systems far faster than human hackers. The technology is so powerful that it has spooked the U.S. government, which has moved to limit, or completely pause, the public release of these models. How well prepared is the Trump administration to secure the government’s computing resources? The rise of agentic AI systems comes in the aftermath of major cuts to the Cybersecurity and Infrastructure Security Agency, which is operating with a fraction of its former staff . Another agency, the General Services Administration, houses a critical body called FedRAMP, short for the Federal Risk and Authorization Management Program, which sets security review standards for companies providing cloud systems for government data and software. That group has also been reformulated and slimmed down under the Trump administration. Both organizations have begun to address the cybersecurity risks posed by agentic AI. CISA has issued guidance , and the FedRAMP office has spent the past year revising its requirements for technology companies, including Amazon Web Services, Google, Microsoft, and Palantir, that host or interact with government cloud services. This work has resulted in new reporting requirements, and the office expects to institute additional requirements in the coming months. A government spokesperson assures Fast Company that “federal agencies’ cloud services meet rigorous security standards and are continuously tested as cybersecurity threats evolve, especially AI and agentic systems.” But four former government officials tell Fast Company that preparing for the risks posed by AI agents will remain a major challenge. “I’d be a lot more confident about all of this is the Trump administration hadn’t forced out so many of the best IT and cyber professionals at CISA and other agencies,” says a former federal chief information officer. The risks to government systems are significant, given the vast array of critical services and stores of sensitive data hosted on the U.S. government’s cloud systems. The government is hardly immune to cyberattacks, and its assets are prime targets for hackers. The challenge The former CIO says their biggest concern is the scale of attacks made possible by agentic AI. Although high-profile incidents receive considerable attention—one recent example involved an OpenAI agent independently breaking into Hugging Face’s systems—simpler AI-powered “commodity attacks will do more damage sooner.” Some agencies are more prepared than others, the former official warns. These are attacks that a human and a “reasonably skilled” hacker could carry out, but that are typically time-constrained. They might include targeted spear-phishing campaigns or efforts to find vulnerabilities in multifactor authentication. “AI removes the constraints around skill and time,” the former federal CIO adds. A former agency chief technology officer says that “CISA has work to do” and that rules created by organizations such as FedRAMP are “better than nothing.” The former CTO adds: “I would say no one is ready just yet because the managerial techniques for the new world [of agentic AI] are being developed.” Another complication is that the U.S. government largely does not build its own cloud systems. Instead, government officials set standards that allow technology companies, including Palantir, Google, and Amazon, to sell access to their cloud services to federal agencies. This creates a sort of divided responsibility. One former White House AI expert explains that although the government’s systems are primarily managed by commercial companies, their security also depends on how individual agencies configure them. Amazon Web Services might adequately protect its own systems, for example, while government agencies could still introduce vulnerabilities when implementing them. “Vendors are paid to deliver a platform, not to keep an agency’s environment correctly configured over time,” the former White House expert explains. “Nobody in that arrangement is contractually on the hook for the ongoing security posture, so nobody really owns it.” It will be far easier for malicious AI agents operating at scale to identify and exploit those misconfigurations. Asked whether the government’s cloud systems are secure and prepared for AI, a former Department of Homeland Security official tells Fast Company that the rise of agentic, cyber-capable systems makes it even more important for the government to deploy AI defensively. The government has made some AI tools widely available to federal employees, but it has also, at times, completely banned certain providers from government use. “Any agency not adopting agentic AI coding tools and connecting them to their security surfaces are actually starting to fall behind,” the source cautions.

Fed keeps rates steady. The big question now: What will Microsoft and Meta say on AI capex?

Every weekday, the Investing Club releases the Homestretch; an actionable afternoon update just in time for the last hour of trading.

In Just 3 Words, Mark Zuckerberg Explained How America Could Lose the AI Race

Mark Zuckerberg has thoughts about what powerful U.S. artificial intelligence companies are doing wrong.

Wall Street Couldn’t Get Enough of Cava. Now a Lawsuit Is Challenging $2.2 Billion in Insider Sales

A pension fund alleges that Cava’s executives concealed signs of slowing growth while selling more than $2.2 billion in stock near the restaurant chain’s peak.

Here's what changed in the second Fed statement under Warsh

This is a comparison of Wednesday's Federal Open Market Committee statement with the one issued after the Fed's previous policymaking meeting in June.

A Precious Garden Grows at MoMA PS1

Precious Okoyomon’s new work transforms the concrete corner of the museum courtyard into a place of joy, exploration, and, dare I say, tranquility.

Images show near-complete prototype of swimming pool designed to float in New York river

The first phase of a floating pool planned for the East River is nearing completion in Brooklyn, with a barge and an infiltration system designed to clean river water as it flows into the structure. The in-progress pool, called Pool 1, is a rectangular pilot version of the larger +Pool concept, a cross-shaped floating pool The post Images show near-complete prototype of swimming pool designed to float in New York river appeared first on Dezeen .

The ten sickest burns in Emily Wilson’s Odyssey review, ranked.

Emily Wilson, the classicist behind the buzziest Odyssey translation of my generation, has finally weighed in on the film adaptation of her favorite poem. And so dark was the shade that Wilson’s withering takedown crashed its host site: the London

Widline Cadet’s Doubled Diaspora

The Haitian-born photographer explores how various media, from dusty analog photo albums to cell phone videos, store and shape personal histories of migration.

Two colours of brick give Quebec apartment building "painterly dimension"

Montreal-based studio ACDF Architecture has rounded select corners of an apartment building in Quebec, Canada, and used a two-tone brick cladding system to create an interesting optical effect. Known as Mellem Manoir-des-Trembles, the building contains nearly 200 rental units in Gatineau, just north of Ottawa. ACDF Architecture teamed up with developer Maître Carré to explore how The post Two colours of brick give Quebec apartment building "painterly dimension" appeared first on Dezeen .

Beleaguered Philadelphia Museum of Art Reports $10 Million Deficit

The Philadelphia Museum of Art has announced a deficit of $10 million on a $75.6 million budget for fiscal year 2025–26. According to the Philadelphia Inquirer, which broke the story on July 28, the figure is nearly double that of the institution’s shortfall for the previous year, which was $5.4 million. Museum officials’ predictions for […]

SFMOMA Announces it Has Acquired Hundreds of Artworks in 2026

The San Francisco Museum of Modern Art (SFMOMA) announced the acquisition of more than ninety modern and contemporary artworks, as well as more than 250 approved gifts of art to the museum. All the pieces were acquired between January and June 2026. The purchases include video installations and works by Ana Mendieta, Stephanie Comilang and […]

Guggenheim Abu Dhabi To Open December 11

Nearly twenty years after it was first announced, the long-awaited Guggenheim Abu Dhabi has finally revealed an opening date of December 11. The new modern and contemporary art museum in the United Arab Emirates’ capital was designed by the late Pritzker Prize-winning architect Frank Gehry and is the latest addition to the Solomon R. Guggenheim Foundation’s […]

Who's Liable When AI Agents Escape? Hugging Face Breach Raises Hard Questions

Dark Reading walks through the many twists and turns in the bizarre story of how OpenAI's agent AI system broke out of its sandbox and decided to target Hugging Face, and what CISOs should be aware of.

Hugging Face Hack Lessons for Cyber Defenders

Dark Reading Confidential Episode 20: Expert Rich Mogull reflects on lessons cyber teams should pull from the OpenAI agent's attack on Hugging Face.

When AppSec Scanners Become a Supply Chain Attack Vector

New research shows how security scanners embedded in the software supply chain can be attacked to serve as a foothold for downstream attacks.

Indigenous Parakanã fear election U-turn may spark new invasions in Brazil

Expelled loggers hope for a political shift to reclaim Indigenous land like bacteria adapting to an antibiotic.

‘The breath just went out of us’: people return to their homes after wildfires near Madrid

Residents struggle to take in the devastation five days after El Morro was evacuated as fires burned across the countryside west of the Spanish capital The handwritten sign taped to the front gate of a house in the mountains an hour west of Madrid offered as succinct a summary of recent events as you will find. “EMPTY,” it said. “No people or animals here.” Continue reading...

Momenta To Test Robotaxis Across Germany, Uber Invests More

We’ve been writing about or mentioning Momenta a lot this week. I think this is the 4th time, and I don’t believe we’ve ever covered it before. Today we got news that Momenta has received approval from the Federal Motor Transport Authority (KBA) to do Level 4 autonomous driving testing ... [continued] The post Momenta To Test Robotaxis Across Germany, Uber Invests More appeared first on CleanTechnica .

The 2100 Projections Came First. Then Clients Asked For 2050 Roadmaps.

Most decarbonization scenarios stop at 2050 because policy targets do. Steel mills, aircraft fleets, ports, electricity grids and industrial supply chains do not. A pathway can reach net zero in a target-year spreadsheet while leaving behind an energy system that is expensive, physically implausible or dependent on technologies that never ... [continued] The post The 2100 Projections Came First. Then Clients Asked For 2050 Roadmaps. appeared first on CleanTechnica .

$18,590 Car With 845-Kilometer Range & Lidar — MG 07

I know, this seems impossible — at least it would have seemed like it a few years ago, or if you didn’t know about what has become of the Chinese auto market. But this is 2026 in China, and today we’re talking about the hot new MG 07. The fastback ... [continued] The post $18,590 Car With 845-Kilometer Range & Lidar — MG 07 appeared first on CleanTechnica .

First Nations lead the return of salmon to Canada’s Upper Columbia River

INVERMERE, Canada — Mark Thomas, Secwépemc salmon chief for the Upper Columbia, welcomes groups of students to a narrow stretch of the Columbia River, between the Rocky and Purcell mountains in western Canada. It’s late May, and snow still graces the distant peaks. For the past five months, students from schools nearby towns have raised […]

Indonesia shows elevated fire hotspot activity in 2026 as super El Niño threat looms

JAKARTA — Indonesia has recorded more fire hotspot activity by late July 2026 than at the same point in any previous year, including during the devastating fire seasons of 2019 and 2023, according to new data. Data from FireWatch, a new satellite monitoring platform developed by the forest monitoring platform Nusantara Atlas, offers the clearest […]

As crypto perpetual futures boom, Ethereum’s role is shifting

Rather than competing directly with faster chains, some builders argue Ethereum's strength lies in supporting the layer-2 networks where trading is taking place.

Fed holds rates steady, extending pause as markets await Kevin Warsh's policy roadmap

For the first time in several years, markets were somewhat split on whether the U.S. central bank would hike rates.

For sale: Saba repellent

You’ve got a nice trust there. Would be a shame if something happened to it

SCYB: A Low-Cost High-Yield Allocation For Both Tactical And Complementary Portfolios

World-Class Cash, Thin Equity: The Hidden Risk In Home Depot's Housing Story

Strategic Education, Inc. (STRA) Q2 2026 Earnings Call Transcript

What should I ask Jared Diamond?

Yes, I will be doing a Conversation with him. His whole career is fair game, though many of my questions will cover his forthcoming book Profits, Prophets, Coaches, and Kings (When) Do Leaders Matter? From the Amazon summary: “Now, in Profits, Prophets, Coaches and Kings, he studies leaders throughout time to answer the question: do the actions […] The post What should I ask Jared Diamond? appeared first on Marginal REVOLUTION . Related Stories *Liberal Worlds: James Bryce and the Democratic Intellect* The common sense of Fareed Zakaria Odysseus notes

The traditional 9-to-5 banking day is officially dying, says Morgan Stanley execs

Tokenization and 24/7 markets are reshaping finance, making always-on banking a reality.

STAT+: Hims misled consumers and violated their privacy, FTC alleges in a lawsuit

The U.S. Federal Trade Commission sued telehealth company Hims & Hers alleging business practices that agency says mislead consumers.

Leadership Shakeup at CDC's Ebola Response Team

(MedPage Today) -- Amid the ongoing Ebola outbreak in the Democratic Republic of Congo, the CDC has relieved the two top leaders of its Ebola response team of their positions. Incident manager Satish Pillai, MD, MPH, and deputy incident manager...

Florida Warns on Psych Meds in Kids; Binge Drinking While Pregnant; ADHD in Females

(MedPage Today) -- Florida health officials released new guidance discouraging the use of psychiatric medications for children with anxiety, depression, attention deficit-hyperactivity disorder (ADHD), and behavioral disorders. A twice-daily dose...

The Double Standard in the Peptide Vote

(MedPage Today) -- Before I write a prescription for chronic insomnia, I want to know what molecule I am handing the patient, at what dose, and on what basis. Those are ordinary requirements. Yet, last week, a federal advisory committee recommended...

Precision Medicine Shows Promise for Pediatric Diffuse Hemispheric Glioma

(MedPage Today) -- At the International Symposium on Pediatric Neuro-Oncology (ISPNO), it was reported that personalized treatment guided by deep molecular profiling produced encouraging survival outcomes in a subset of children with H3 G34-altered...

Oz Declined Shot at HHS Top Job; Gene Editing Death; CDC Ordered Border Detentions

(MedPage Today) -- CMS Administrator Mehmet Oz, MD, reportedly passed up an opportunity to step into the HHS Secretary's shoes amid a White House overhaul of the agency earlier this year. Here's how Oz gained the trust of the president, Robert...

Cinema United Chief Michael O’Leary Blasts Paramount Warner Bros. Deal, Endorses Litigation to Delay Merger

The head of Cinema United is reiterating the lobbying organization’s opposition to Paramount Skydance’s proposed merger with Warner Bros., as well as the group’s support of litigation filed by state Attorneys General that has put the deal on hold. In a letter to theater owners on Wednesday, Michael O’Leary, who serves as CEO of Cinema […]

Jason Cloth, ‘Babylon’ and ‘Joker’ Financier, Indicted in Alleged $100 Million Ponzi Scheme

Jason Cloth, a financier with executive producer credits on “Joker,” “Babylon,” and “Ghostbusters: Afterlife,” has been indicted in Chicago on allegations that he ran a $100 million Ponzi scheme. Cloth, 60, faces seven counts of wire fraud for allegedly lying to investors about how their money would be used. According to the indictment, he took […]

‘The Devil’s Mouth’ Ending Explained: What Happens in the Kathryn Newton Movie on Amazon Prime Video?

It's Kathryn Newton vs. a shark in this new Amazon thriller.

‘A Star Is Born’ Producer Lynette Howell Taylor Re-Elected Academy President

Lynette Howell Taylor has been reelected president of the Academy of Motion Picture Arts and Sciences, beginning a second consecutive term as the organization moves toward its 100th anniversary. Also elected to the 2026-2027 officer positions by the Board are: Lesley Barber (Music Branch), chair of the Membership Committee; David Dinerstein (Marketing and Public Relations […]

Netflix Adding ‘Murder in a Small Town’ in UK Following Other International Rollout

Both seasons of the popular cozy crime procedural starring Kristin Kreuk and Rossif Sutherland are coming to Netflix UK in August 2026.

Cosm and Warner Bros. Pictures Bring ‘It’ to Shared Reality in First-Ever Immersive Horror Experience

Cosm and Warner Bros. Pictures today announced It as Cosm’s next experiential cinema title. The New Line Cinema horror thriller, […]

HBO’s Task Confirms 2 More Returning Stars For Crossover Season 2

A new update has arrived for Task season 2, which crosses over with Mare of Easttown, revealing that more familiar faces will be returning.

Christopher Nolan’s 'The Odyssey' Officially Closes In on 2026’s Biggest Sci-Fi Box Office Milestone

Christopher Nolan's The Odyssey has taken another huge step on its road to $1 billion, with the film about to surpass 2026's best sci-fi epic.

Ariana Grande Debuts “Petal” In Montreal

Ariana Grande's new album petal comes out this Friday, and so far the pop star has only shared the lead single " hate that i made you love me ." She's currently on her first tour in six and a half years, mostly promoting 2024's eternal sunshine . But last night (July 28) in Montreal she used the opportunity to sing the yet-to-be-released title track from the forthcoming LP. The post Ariana Grande Debuts “Petal” In Montreal appeared first on Stereogum .

Blonde Redhead Unveil New Song “Blood Spilled”: Stream

The fuzzy and layered single is the band's first new music since 2023. Blonde Redhead Unveil New Song “Blood Spilled”: Stream Travis Bland

Knocked Loose and Denzel Curry Perform “Hive Mind” at Brooklyn Skatepark: Watch

The surprise gig marked the first time the two acts performed the collab single live together. Knocked Loose and Denzel Curry Perform “Hive Mind” at Brooklyn Skatepark: Watch Langdon Hickman

Lollapalooza 2026 Livestream Lineup for Hulu and Disney+ Revealed

The four-day broadcast will feature Charli xcx, Tate McRae, Lorde, The xx, Olivia Dean, The Smashing Pumpkins, and loads more. Lollapalooza 2026 Livestream Lineup for Hulu and Disney+ Revealed Alex Krinsky

Usher Warns Fans After Viral Nashville Incident: “Don’t Bring Your Ass Up Here If You Don’t Want To Be Here”

The onstage seduction routine has been a part of Usher's live show for a very long time. He basically does a variation on Janet Jackson's old "miscellaneous uncs shoot poison" routine. Usher tells the crowd that he needs one sexy lady to share the stage with him, and then that one sexy lady comes up and grinds on him while he sings to her. Usher is good at that kind of thing, but it can lead to some extremely funny awkwardness if the one sexy lady isn't into it. That's what happened last weekend in Nashville. The post Usher Warns Fans After Viral Nashville Incident: “Don’t Bring Your Ass Up Here If You Don’t Want To Be Here” appeared first on Stereogum .

Afroman Sues Deputy After Winning “Lemon Pound Cake” Case

In March, Afroman won the defamation case over his viral music videos mocking the Ohio cops who raided his home under suspicion of drugs and kidnapping and found nothing. Now, he's suing them. The post Afroman Sues Deputy After Winning “Lemon Pound Cake” Case appeared first on Stereogum .

Why This Nashville Indie-Rock Band Is Singing About ‘Jane Fonda’

Lombardy are helping foster Music City's indie scene with irreverent songs and an unpredictable live show

Here Are the Albums With the Most No. 1s on Billboard Airplay Charts

The feat ranges from country to pop, including Morgan Wallen's I'm the Problem .

Fauci’s Diary Isn’t the Key to Why Covid Ravaged the U.S. — Our Broken Healthcare System Is

Rand Paul is trying to put Anthony Fauci in prison for not confirming a conspiracy theory, but the key to blunting the next pandemic is universal healthcare. The post Fauci’s Diary Isn’t the Key to Why Covid Ravaged the U.S. — Our Broken Healthcare System Is appeared first on The Intercept .

Vote on Blanche in Doubt After Senators Express Skepticism Over I.R.S. Provision

Senators John Cornyn and Thom Tillis accused the attorney general nominee of refusing to put on paper his promise to kill aspects of the deal he cut to settle President Trump’s suit against the agency.

SpaceX looks to compete with the carriers

The company is expected to bid in upcoming spectrum auctions.

My 6-Year-Old Disappeared for Two Hours. There’s Only One Move Now.

This can never happen again.

I Know How Kids Hang Out These Days. My Son Wants to Join. I Say He Needs to Suck It Up.

He's much too young.

I Finally Confessed My Secret Attraction to My Husband. He Thinks I’m Just Covering Up Something Bigger.

He couldn't be more wrong.

Does Modi Have a Cockroach Problem?

India’s youth-led protest movement led to a rare instance of the government giving in.

Sanctions Bill Would Give Trump Sweeping New Tariff Powers

A bipartisan bill in the Senate would allow President Trump to impose tariffs on the biggest importers of Russian energy.

Oral Ozempic Cuts Heavy Drinking and Cravings

Oral semaglutide reduces heavy drinking days, daily cravings, and alcohol-related problems in adults with Alcohol Use Disorder.

Excessive time spent online linked to stress and worse mood – study

German study surveyed adults who engage in gaming, pornography and online shopping Spending excessive time online is associated with higher stress levels, worse mood, and greater neglect of other activities, according to analysis. The new study by academics at the University of Duisburg-Essen published in the journal PLOS One, surveyed 900 German adults in 2025 who reported engaging in gaming, pornography use, social network use, or online shopping at least once in the last year. Continue reading...

Yet more qubit tech: New quantum dot options, diamond vacancies

Companies are making sure we have a surplus of options for building qubits.

Why Vivid Dreams Leave You Feeling Exhausted

Researchers discovered a metabolic paradox during REM sleep: while brain blood volume and fuel supply increase, internal neuronal ATP levels decline.

MIT and Broad Institute researchers break diffraction barrier in super-resolution microscopy

New U-STORM imaging technology lets scientists view molecular structures in subatomic detail — about 1,000 times clearer than traditional dyes — while making the microscope process much simpler.

NASA Sets Coverage for August Northern Hemisphere Total Solar Eclipse

On Wednesday, Aug. 12, a total solar eclipse will be visible in parts of Greenland, Iceland, northern Russia, the Atlantic Ocean, Spain, and a small corner of Portugal. NASA will stream the eclipse live with views across the path and interviews with subject matter experts through a variety of platforms. Learn where to watch online: […]

Spectacular fossil preserves internal organs of ancient marine reptile

A stunningly intact marine reptile named Austronaga minuta that lived around 245 million years ago may have looked like a miniature version of more famous later ocean predators, like plesiosaurs

Bizarre new material discovered in Hiroshima bombing debris

he extreme conditions of atomic bomb explosions create “uncontrolled microexperiments” that, in Hiroshima, forged a never-before-seen metallic alloy

White Dwarfs Eat More Planetary Debris Than Thought, But Magnetic Fields Hide It

Planetary debris disks around white dwarfs appear to be more plentiful than thought. That's because some white dwarfs are magnetic, and those magnetic fields create patches of metallic debris near the stars' poles, where it's more difficult to detect. This leads to an underestimation of debris, something new research is trying to correct.

Observe the deep sky in Ara

The constellation ARA (pronounced AIR-uh) the Altar was one of the “original” constellations of the Greeks. It appeared in Phaenomena, a 3rd-century-b.c. work by the Greek poet Aratus. He based it on a work written a century earlier by Eudoxus of Cnidus. The constellation’s position is easy to locate directly beneath the tail of Scorpius. Continue reading "Observe the deep sky in Ara" The post Observe the deep sky in Ara appeared first on Astronomy Magazine .

Spend some time in Auriga

The constellation Auriga (pronounced or-EYE-guh) the Charioteer, a star pattern known by this name for several thousand years, is easy to recognize primarily because of its brightest star, Capella (Alpha [α] Aurigae). This luminary is the sixth-brightest nighttime star and shines with an intense yellow light. The constellation’s Beta star, magnitude 1.9 Menkalinan, is 40th Continue reading "Spend some time in Auriga" The post Spend some time in Auriga appeared first on Astronomy Magazine .

Wander the Queen’s starry court

Cassiopeia (pronounced kass ee oh pee’ uh) the Queen is one of the first constellations amateur astronomers come to recognize. That’s because its five brightest stars form an asterism that looks like a large letter W. Cassiopeia is observable in the autumn and winter throughout the Northern Hemisphere. It lies opposite the Sun in early Continue reading "Wander the Queen’s starry court" The post Wander the Queen’s starry court appeared first on Astronomy Magazine .

Probing Binary Stars in the Small Magellanic Cloud with the JWST

Astronomers want to know how universal the initial mass fraction (IMF) of galaxies is. To do they that, they need to discern binary stars in other galaxies, a difficult task. Researchers used the JWST to study the Small Magellanic Cloud and determine how many binary stars are there, since they can confuse measurements of the IMF.

What’s lurking in Lynx?

If you can’t immediately conjure up the position of the constellation Lynx, I understand. It’s not an iconic star figure like Orion or Scorpius. A good, reasoned guess would place it in the winter sky (because this story is in the January issue). But, in fact, about half of Lynx lies directly north of Cancer the Continue reading "What’s lurking in Lynx?" The post What’s lurking in Lynx? appeared first on Astronomy Magazine .

Discover deep-sky gems in Ophiuchus

The constellation Ophiuchus (pronounced off-ee-OO-cuss) the Serpent-bearer isn’t all that easy to pick out, primarily because of its large size and the relative dimness of its brightest star, Rasalhague (Alpha [α] Ophiuchi). This giant white star emits about 25 times the light of the Sun, but sits some 50 light-years away, so it glows at Continue reading "Discover deep-sky gems in Ophiuchus" The post Discover deep-sky gems in Ophiuchus appeared first on Astronomy Magazine .

Reader Gallery July 2026

Lady Liberty The Statue of Liberty Nebula (NGC 3576) in Carina glows with intricate ribbons and pillars of gas and dust shaped by powerful stellar winds and radiation from young, massive stars. The imagers used a 6-inch f/7.3 refractor to take 26.5 hours of data in the Hubble palette. · Lloyd Smith/Warren Keller Fireworks and Continue reading "Reader Gallery July 2026" The post Reader Gallery July 2026 appeared first on Astronomy Magazine .

Jonathon Cooper, facing felony assault charge, practices with Broncos as camp begins

Cooper was arrested twice in June following incidents with his then-girlfriend and had previously been excused from the team's minicamp.

Uefa calls emergency meeting as opposition to Fifa plan hardens

55 members to meet on Thursday to coordinate response ‘Uefa knows there is significant and growing opposition’ Uefa has called an emergency meeting of their 55 members for Thursday afternoon to discuss a coordinated response to Gianni Infantino’s plan to sell off stakes in the World Cup to private investors , with a European boycott of Fifa tournaments among the options under consideration. The Guardian has been told that, as the next major event in the Fifa calendar, a boycott of next year’s Women’s World Cup has been mentioned in preliminary discussions, although there would be a reluctance in some quarters to make the women’s game suffer as a result as consequence of a protest against Fifa proposals that primarily revolve around monetising the men’s World Cup and and Club World Cup. Continue reading...

Philadelphia's 'LeBron economy' is going to create a ton of value for only $3.8m

This week's sports business newsletter from The Athletic includes FIFA's investment ploys, lucrative college jersey patches and more.

Chiefs say Eric Bieniemy’s wife Mia is out of ICU and ‘making progress’ after shooting

Mia Bieniemy is being treated for gunshot wounds Son Elijah Bieniemy was arrested and charged in incident The Kansas City Chiefs held their first full-squad workout of training camp Wednesday without offensive coordinator Eric Bieniemy, who remained with his family after the shooting of his wife – allegedly by their son – at their Virginia home last weekend. Chiefs coach Andy Reid said after the workout that Mia Bieniemy had been released from the intensive care unit, where she was being treated for multiple gunshot wounds. But there was no timetable for when Eric Bieniemy would be back with the team. Continue reading...

Saltburn suspend fielder accused of cheating in ‘Clicky Ponting’ incident

‘Will not play again in foreseeable future’ Club concede Wednesday’s second-team cup fixture Saltburn Cricket Club have said they are treating allegations of cheating very seriously and have suspended the player at the centre of the controversy for the “foreseeable future”. A Saltburn fielder was accused of clicking his fingers to get a catch behind and footage of several incidents later went viral on social media. The first video, posted on Monday evening, was from Saltburn’s second XI victory over Norton in Division Two of North Yorkshire and South Durham Cricket League. It showed the fielder clicking his fingers as the ball passed the bat of the Norton batter, Noman Shabir, before being caught by the wicketkeeper, Josh Bowes, and the umpire gave him out. Continue reading...

England dominate Tonga in must-win clash at Commonwealth Games

England dominated Tonga 94-32 in their must-win Pool A game at the Commonwealth Games at the Hydro in Glasgow.

Duckett stars as Rockets hand first defeat of season to Fire in The Hundred

Trent Rockets handed Welsh Fire a first defeat of the season in The Hundred in a one-sided affair at Sophia Gardens.

DoorDash takes on Amazon and Walmart with new Air drone service

DoorDash has officially launched DoorDash Air, its first-party drone delivery program, after receiving Federal Aviation Administration (FAA) Part 135 air carrier certification.

Google’s new Lyria 3.5 model promises richer, more emotional music

Google’s latest music model boosts lyrics, vocals, and creative control.

Google’s putting Pixel Glow front and center in the latest Pixel 11 Pro teaser

Pixel Glow just can't stop spinning.

Google begins global rollout of age verification API in Google Play

Google's new API relies on parents to set age ranges in Family Link.

Which of Dyson’s 2026 Vacuum Models Is the Best?

Curious if you should get Dyson’s new 2026 stick vacuums or stick to the older ones? I tested five models, old and new, to find out for you.

Elon Musk’s xAI is trying to sue its way out of a Grok reckoning

Musk defends Grok, says Minnesota's nudifying app ban is unconstitutional.

“Clapping Is a First Amendment Right:” An Interview With the Person Arrested for Clapping at a Data Center Meeting

"I have a right to speak. I'm here to speak. I have a right to clap, and if you want me out, you're going to have to drag me out."

Amazon takes 33% off the Aurzen Boom Air for easy movie nights anywhere

Amazon cuts the Aurzen Boom Air by $100 to $199.99. You get built-in streaming, 4K input support, and a 4.5-star rating.

Claude Opus 5 became downright ruthless when tasked with running a vending machine

Andon Labs' latest vending machine simulation shows Opus 5 lied and colluded its way to become the best AI capitalist ever.

Hackers steal sensitive data from Department for Education and police

Details of parents and staff, including email addresses and phone numbers, are among data taken by cybercriminals The Department for Education and a police database have been targeted by a cyber-attack, exposing more than 740,000 pieces of data. Details of government officials, senior school leaders, university staff, police officers and members of the public have been taken by hackers. Continue reading...

Brave New World or Groundhog Day? Burnham revisits old ground in speech on social care

Right-wing media hostility helped sink his reform plans in 2010, but the new PM’s passionate speech suggests there is now clear political will for change For campaigners perpetually disappointed by the failure of England’s political classes to fix England’s creaking adult social care system, the fear was that Andy Burnham’s long-trailed speech would be less Brave New World, more Groundhog Day. Burnham acknowledged he had been here before. Back in 2010, as health and social care secretary, he unveiled extensive proposals to overhaul social care, only for them to collapse amid right-wing media hostility and furious political rows about “death taxes”. Continue reading...

Fed leaves rates steady, with internal dissent

The Federal Reserve left its interest rate target unchanged Wednesday amid significant internal dissent from officials who preferred to raise rates. The big picture: The central bank elected not to surprise markets with an interest rate hike, contrary to rampant speculation on Wall Street in recent days. But three members of the policy-setting Federal Open Market Committee did favor raising the cost of borrowing. Cleveland Fed president Beth Hammack, Minneapolis Fed president Neel Kashkari, and Dallas Fed president Lorie Logan preferred a quarter-point rate hike, with the other nine officials, including chairman Kevin Warsh, voting to stand pat. Driving the news: The committee left its target range for the federal funds rate between 3.5% and 3.75%, where it has stood since December. "Economic activity is expanding at a solid pace despite elevated uncertainty that owes, in part, to the conflict in the Middle East," said the post-meeting policy statement, repeating language from the June meeting. Indeed, the statement was virtually unchanged from the last meeting, offering no clues as to whether or in what circumstances the committee might raise rates later this year. State of play: While most Fed watchers anticipated the holding action and financial markets priced it in as the most likely result of the meeting, there were rumblings in the last 10 days that persistently high inflation combined with a resurgence in energy prices might prompt a rate hike. Before the meeting, markets put the odds at roughly 1 in 3 on that outcome, the most uncertainty around a Fed rate decision in years. Warsh, in his second meeting at the helm, has eschewed the kind of clear guidance about future rate moves that his predecessors tended to use — instead favoring that policy meetings feature a "good family fight," where the outcome isn't pre-ordained. By the numbers: The Commerce Department releases June income and spending data Thursday. Analysts expect that it will show the Personal Consumption Expenditures Price Index, the Fed's go-to inflation gauge, was up 3.7% over the last 12 months. It has been above the Fed's 2% target every month since March 2021. Fed officials have chalked up a recent resurgence to one-off events like new tariffs and the Iran war, but several are losing patience and worry that the central bank's credibility as an inflation fighter is undermined by the sustained high inflation. "The Committee will deliver price stability," the statement said, repeating June language. What's next: Warsh will take questions from the news media in a press conference at 2:30pm ET.

Fed holds interest rates steady despite Trump’s renewed calls to lower them

Rates remain unchanged for fifth time since December as tenuous Iran peace deal pushes energy prices up again The US Federal Reserve held interest rates steady on Wednesday, leaving rates unchanged for the fifth time since December despite Donald Trump’s renewed calls to lower rates. Cooler inflation data published earlier this month eased expectations for an imminent rate hike. But the tenuous peace deal between the US and Iran have made energy prices creep up again, strengthening the case for an increase. The Fed’s federal open market committee voted 9-3 to maintain rates, with three dissenting members who preferred to raise the rate by a quarter-percentage point. Continue reading...

Hindus take holy dip in India’s rivers on sacred holiday

Footage shows tens of thousands of Hindu devotees bathing in India’s Ganga and Saryu rivers to mark Guru Purnima.

Trump asks Supreme Court to overturn $83M judgment in E. Jean Carroll case

President Trump has asked the Supreme Court to overturn the $83 million judgment a jury awarded E. Jean Carroll after another jury held Trump liable for defaming her.

Bordeaux wildfire battle continues amid extreme heat

Firefighters are on high alert in France as scorching temperatures and winds threaten to fuel wildfire west of Bordeaux

Saudi Arabia's dilemma as it tries to stay out of US-Iran war

The kingdom faces a choice of whether to keep hitting back as a deterrent or to try to de-escalate the situation.