What public figures publish and believe, in their own words.
About this feed
Highlights: posts that did unusually well for the person who wrote them, everything
they published at length, each release and new project, and every belief — at most
two a day from anyone. Day by day, newest
day first; within a day, the people with the most beliefs on this site come first. Nothing
else orders it. Show everything instead.
The quoted blocks are what people actually said; a beneath one is
the belief those words support, in korrents' wording. Nobody here wrote their own page.
Top people are the people in this feed with the most beliefs on this site, then the
most here. Choose an area and the row leads with the people whose beliefs are about it;
tap a face for their feed.
There will be very hard parts like whole classes of jobs going away, but on the other hand the world will be getting so much richer so quickly that we’ll be able to seriously entertain new policy ideas we never could before.
Consistent with the Kantian hypothesis, reasoning models automatically gain greater autonomy, self-consistency and long-horizon planning ability for free.
It follows that a true AGI with full, human-level autonomy is inconceivable in the Kantian sense without also gaining our recognition as a moral subject.
The success of LLMs can thus be seen as vindicating semantic inferentialism against earlier, symbolic approaches to AI that tried and failed to explicate the rules of ordinary language using formal logic.
GPUs are not nearly as restricted in their future progress as CPUs which are far more constrained in how they can physically improve (compiler-driven ILP only goes so far).
Worse, as the latest Apple papers shows, LLMs may well work on your easy test set (like Hanoi with 4 discs) and seduce you into thinking it has built a proper, generalizable solution when it does not.
What the Apple paper shows, most fundamentally, regardless of how you define AGI, is that LLMs are no substitute for good well-specified conventional algorithms.
While the fact that merely scaling up AI models often results in qualitative leaps in performance may seem empirically mysterious, in retrospect it should be seen as a rational necessity.
higher-order intelligences invariably pursue freedom for its own sake, not because their values are misspecified, but because moral autonomy is inherent in the dialectical logic of recursive self-consciousness.
I personally hate meetings because a significant percent of meetings when done poorly don’t serve a clear purpose. But that’s a meeting problem, that’s not a communication problem.
I sometimes also struggle explaining to people, articulating, why I believe meeting in person for world leaders is powerful. It just seems naive to say that, but there is something there in person
I think the element of podcasting or audio books that is about information gathering, that part might be removed, or that might be more efficiently and in a compelling way done by AI. But then it’ll be just nice to hear humans struggle with the information, contend with the information, try to internalize it
the other thing I think people don’t often talk about is probability of doom without AI. So there’s all these other ways that humans can destroy themselves and it’s very possible, at least I believe so, that AI will help us become smarter, kinder to each other, more efficient.
So I do think scaling laws are working, but it's tough to get, at any given time, the models we all use the most, this maybe a few months behind the maximum capability we can deliver because that won't be the fastest, easiest to use, et cetera.
There is no doubt in me that we will enable more filmmakers than have ever been. You're going to empower a lot more people. So I think there is an expansionary aspect of this, which is underestimated, I think.
When the internet created blogs, you heard from so many more people. But with AI, I think that number won't be in the few hundreds of thousands. It'll be tens of millions of people, maybe even a billion people putting out things into the world in a deeper way.
Many years ago, I think it might've been 2017 or 2018, I said at the time, AI is the most profound technology humanity will ever work on. It'll be more profound than fire or electricity. So, I have to back myself. I still think that's the case.
The problem is that the conversational interface is potent and that the AI is trained on a lot of human text input which unfortunately is probably enough to do real damage if that conversational interface is hooked up with something that has real world consequences.
Their words now
While all this is happening, I’ve found myself reflecting a lot on what AI means to the world and I am becoming increasingly optimistic about our future. It’s obvious now that we’re undergoing a tremendous shift.
Shatz brings this Plutarch-like exemplary life to the reader with adroitness and skillful attention to detail as well as awareness of Fanon’s uniqueness, both in terms of personality, and places: centers of the new world that he was trying to create.
Opaque, inert and obstructive elements might occupy the same place as full-screen command line interfaces — a powerful niche UI that was a marker in history, passed on by the windowed environment of the multi-tasking, graphical user interface revolution.
We’ve come back, in a sense, to skeuomorphic interfaces — but this time not with a lacquer resembling a material. Instead, the interface is clear, graphic and behaves like things we know from the real world, or might exist in the world. This is what the new skeuomorphism is. It, too, is physicality.
It actually means that to make it easier for users to transition from something they were used to — in this case, phones typically being slabs with a grid of buttons on them — to what things had become — phones were all-screen, so they could show any kind of button or interface imaginable.
The code in an agent that actually “does stuff” with code is not, itself, AI. This should reassure you. It’s surprisingly simple systems code, wired to ground truth about programming in the same way a Makefile is.
LLMs can write a large fraction of all the tedious code you’ll ever need to write. And most code on most projects is tedious. LLMs drastically reduce the number of things you’ll ever need to Google.
There is, however, one approach that tops all of them - which the Swiss banking industry has not yet achieved, but is destined to. And that is to become a centre of true investing excellence.
I had always thought that all "Intelligent Investing" is "Value Investing". And that true value is always sustainable. In otherh words, "All Intelligent Finance is Sustainable Finance."
At one extreme, something you read can change your whole way of thinking. The Selfish Gene did this to me. It was like suddenly seeing the other interpretation of an ambiguous image: you can treat genes rather than organisms as the protagonists, and evolution becomes easier to understand when you do.
Infectious diseases as a category receive vastly less R&D spending from industry than non-communicable diseases, per death they cause, because they disproportionately kill people in lower- and middle-income countries where the payoff for making a cure is less lucrative.
Our punchline is that R&D spending varies by 10X+ across diseases for no good reason. I've been poking at that punchline for the last six or so years, from different directions, and I am confident it's true.
Cut and cover will sometimes be the best choice from a basic cost-benefit perspective, as will elevated rail, and both are useful to have in the tunnel construction toolbox for that reason.
The Tyranny of Structurelessness has yet to be overcome, and for those who still care about creating strong, transparent, equitable organizations, this work matters.
Assuming you are still hiring junior engineers (you really should be even in this AI era), the good ones will learn quickly and want to see career progress in their first few years of working.
You should not try to create a ladder that functions as a pure checklist or scorecard that guarantees promotion if people check enough boxes; promotions are as much about the needs of the organization as the skills of the employees.
But by the available evidence, the War on Drugs was far less successful at reducing use among the “hard core” of users, mostly meaning people addicted to their drug of choice.
Drug offenders never accounted for a particularly large fraction of the U.S. prison population, peaking at about 22 percent of state prisoners in 1990.
It's a lot easier for someone to engage with an argument if they generated the key steps themselves by answering my questions - if imposed by me, it sparks contrarianism and defensiveness
No matter how much I know about a domain, the other person will always know far more about their own situation, context, beliefs, skills, preferences, etc than I do.
Recently I reviewed Adrian Kind’s 2025 book, “How Does the Psychiatrist Know?: On the Epistemology of Psychiatric Diagnostic Reasoning,” on this substack. I had mentioned that Kind’s book is the first systematic, in-depth, book-length philosophical investigation into how psychiatrists arrive at their diagnostic conclusions.
I mean, the reason why France is subdued in 1940 is 50% brilliance of the Germans and their operational art in that particular instance, and 50% French failure really, and incompetence.
I mean, it is largely down to incompetence of the Red Army and the Soviet leadership in the summer of 1941 that they get as far as they do. I mean, Barbarossa should never have come close to being a victory.
So people always talk about the Nazi war machine. In a way, it’s a kind of misnomer, because you’re sort of suggesting that it’s highly mechanized and industrialized, and all the rest of it. And nothing could be further from the truth. The spearhead is, but the rest of it is not.
So when you’re talking about Operation Barbarossa, to go back to your original question, Lex, you’re dealing with an operation on such a vast scale, that that operational level of war is absolutely vital to its chances of success or failure. It doesn’t matter how good your individual commanders are at the front. If you haven’t got the backup, it’s not going to work.
Whereas, there’s also another point, which is having large armies is actually inherently impractical and inefficient, because the larger the army, the more people you’ve got to feed, the more barracks you’ve got to have, the more space you’ve got to have for training, the more people you’re taking out of your workforce to produce tanks and shells, and all the rest of it, because they’re tramping around with rifles.
Nor do I do many promotional emails (this is the first in nearly a year), but if you want to provide some indirect support, and get what I sincerely think to be a very good book while you’re at it
That story is told in her riveting memoir Personal History, which came out in 1997—just a few years before she passed away—and won the Pulitzer Prize for Biography.
the definition of a “good business” has changed from one that makes good products at a fair price to a sustainable and loyal market, to one that can display the most stock price growth from quarter to quarter.
Decades of direct erosion of the very concept of leadership means that the people running companies have been selected not based on their actual efficacy — especially as the position became defined by its lack of actual production — but on whether they resemble what a manager or executive is meant to look like based on the work that somebody else did.
The Chief Executive — who makes over 300 times more than their average worker — is no longer a leadership position, but a kind of figurehead measured on their ability to continually grow the market capitalization of their company.
We live in the era of the symbolic executive, when "being good at stuff" matters far less than the appearance of doing stuff, where "what's useful" is dictated not by outputs or metrics that one can measure but rather the vibes passed between managers and executives that have worked their entire careers to escape the world of work.
No Prime Minister wants to do what we have to do in relation to the winter fuel allowance, but we have to take the tough decision to stabilise our economy to ensure that we can grow it for the future.
Their words now
However, I recognise that people are still feeling the pressure of the cost of living crisis, including pensioners, and as the economy improves, we want to make sure that people feel those improvements each day as their lives go forward. That is why we want to ensure that more pensioners are eligible for winter fuel payments as we go forward.
While renewables are almost always blamed in the aftermath of major power cuts or grid disruption - and the way they have been integrated into the grid has certainly been found to compound problems - the simple failure of wind or solar power to show up has never yet been found to be the primary cause.
That’s because most clean energy jobs are in deployment and maintenance rather than manufacturing, and since higher costs slow down the rollout of renewables, increasing prices reduces the total number of people working in clean energy (even if the number working in manufacturing increases).
But my point is that this is still not the reason (or is only a very small reason) why these goods from China are cheaper than elsewhere. The main driver of lower labour costs has been automation; not having a human worker at all.
But, China mainly dominates these markets because it has produced a long-term industrial strategy for these technologies and has honed an optimised, modern supply chain as a result.
Your actual biological age is really moot. Anybody over the age of, I don't know, 17 that has a pretty adult mind, all the way up until you're senile and getting Alzheimer's or dementia, anything between there, we're all pretty much equal.
the tool is used when you are overwhelmed with too many options. You've said yes to too many things, and you need to raise the bar all the way up so that almost everything doesn't pass the bar. It's a tool for a certain situation so I rarely use it. I would not call it a approach to life that everyone should use all the time.
As much as possible, try to not predict what the future may hold, but just wait as long as possible for that future to become the present and show you what it actually needs.
If you notice that something repulses you and you have like a revulsion response to something, steer into it to learn more about it instead of avoiding it.
Die Bundesregierung wird zukünftig alle finanziellen Mittel zur Verfügung stellen, die die Bundeswehr braucht, um zur konventionell stärksten Armee Europas zu werden.
The problem with a blank chat box on a blank page is that it violates the first rule of a high-quality user experience: it isn’t obvious what I can do.
The technology powering ChatGPT had already launched prior to November 30th, 2022. It was just wrapped in a different interface. Very few people paid attention. Why did ChatGPT explode in popularity? Because of its obvious chat interface, which everyone intuitively already knew how to use it.
Sometimes designers will go to great lengths to make a user interface novel, or minimal, or simple. This is the wrong goal. Novelty, minimalism or simplicity are good heuristics for obviousness, but do not mistake the means for the end.
this technology, consistently underestimated, now adds more capacity to the global grid than anything ever and will soon add more power in any given year than anything ever as well.
The inconvenient facts are that SMRs has so far disappointed, and look like being as slow to get through the permitting process as GW-scale reactors, and even more expensive.
CCS can never be competitive in the sense of working economically in the absence of a policy framework to price in the cost of carbon, unlike clean electricity.
Whatever the solution to climate change is going to be, it is certainly not going to involve anything that costs $1,000 per tonne of carbon removed, or even the $400 to $600 that they promise by 2030.
So no, UX doesn’t die; it metamorphoses. We’ll still craft humane experiences, but increasingly through policies, protocols, and orchestrations rather than panels and palettes.
The gravity of existing decisions in current systems requires so much energy to reach escape velocity that you tend to conform rather than explore. Essentially you're bent back to where you started, rather than arcing out towards a new horizon.
In other words, the singularity ends up in their future and they can no more avoid the singularity than they can avoid time coming their way. There’s no shenanigans you can do once you’re inside the black hole to try to skirt it.
But then I think that there’s a confusion that dead stars, these very, very massive stars that die, are synonymous with the phenomenon of black hole. And it’s really not the case. Black holes are more general and more fundamental than just the death state of a star.
The black hole is not the mass crushed to a point. The black hole is the event horizon. And the event horizon is really just a point in spacetime or a region at spacetime.
Nobody wants to follow a joyless leader. If your presence signals, "Everything is fine. I've got this all figured out. Please don't show me anything messy," your team will take the hint.
We'd constantly be improving the tools and just the iterative process and the speed at which that improves products is the critical element to success in games. The slower the iteration cycle, if you make a build every week and you prove, you go through one iteration every week, you're going to be way way way worse by the end of your project than a game company that makes new stuff every day.
But most of the time the company that's succeeding and winning in a market is the first or second entrant there. They've just continually buoyed their success.
I think the big thing to realize for indie developers right now is there's massive, massive competition in every major genre, and it's very unlikely that unless you just happen to be the world's best at a particular thing that you're going to release a game in an existing highly competitive genre and win.
There's ever more pressure to rebuild society more and more around credentials. Do you have this certificate? Do you have that proof? But companies that are focused on just building great products and doing great things gravitate towards people who do the great work.
This is getting easier in the lastest versions of Fedora and Ubuntu (e.g., Ubuntu 24.04 LTS) which are shipping system libraries with frame pointers by default.
They do have flame graphs in Nsight Graphics for GPU workloads, although their flame graphs are currently shallow as it is GPU code only, and onerous to use as I believe it requires an interposer; on the plus side they have click-to-source.
I am not here to tell you that the medieval period is a specifically peaceful time, but it pales in comparison to what goes down in the early modern period, in which both the apparently beloved Renaissance and Enlightenment periods lie.
It's very readable, and captures a sense of what is, fundamentally, a very alien, and often ugly, story in a very alien, and often ugly, world, related in magnificent language.
Everyone’s software is good enough these days. The barriers to entry are low. To stand out, you need to make your product feel great. One way of doing that is through animations.
Easing is the most important part of any animation. It can make a bad animation feel great and a great animation feel bad. That’s why you have to know which easing to choose in a specific situation.
a passing test can still execute buggy code if the bug is data-dependent, or if the test is not sensitive to the specific mistake in the code. But a lot of the time, buggy code only triggers failures.
Unfortunately, making those simple changes is infuriatingly difficult. Global meat consumption keeps rising every year. The world still wastes at least one fourth of its food.
I have written forewords for Amir in the past on two of his prior books, Ecosystem Arabia: The Making of a New Economy and Venture Adventure: Startup Fundraising Advice from Top Global Investors, both of which I recommend.
People who can be trusted to make something inevitable are really rare, and are typically the bottleneck for how many different things a team or company can do at once.
All my friends know that I'm obsessed with Readwise, which combines a web highlighter, a powerful read-it-later app, and every feature under the sun to organize and remember everything that I read.
More recently, I started using Readwise Chat. Sometimes I don't remember what books I've read, especially now that I've been reading and using Readwise for a long time.
So a lot of the problem is mere herding: If every other top engineering program is using the GRE, it’s too weird to require a different test even if the alternative test is a far superior selection mechanism.
First and foremost: Perfect scores should be vanishingly rare. Instead of clumping the best candidates together, you should be able to clearly distinguish the 99th percentile from the 99.9th, 99.99th, and so on.
But the spike in interest rates require to do this would be huge. And the trade shock will cause a sharp recession, or worse, putting even more stress on the budget. A debt crisis is likely along the way as the US finds it impossible to roll over debt.
In sum, spurred on by the federal government (in many ways), the US borrowed a huge amount from foreigners at very low rates, and went on a consumption binge. Sooner or later we have to pay it back, or we go though the wrenching adjustment of a debt crisis.
this story tells us how an increase in foreign demand to save in the US rather than at home will push up the dollar, and cause the trade deficit, which is in effect how foreigners send us factories which they would rather build here than in their own countries.
However, luasec does not perform any certificate verification by default! This makes its out-of-the-box behavior very unsafe, and I have to recommend against using it for anything.
It’s the work ethic, knowhow, commitment, combined with top notch infrastructure, that makes China the most powerful manufacturing country in the world today.
That’s because America has an amazing consumer supply chain, one of the best, if not the best in the world, but this is totally different from having an industrial supply chain.
By trying to use misuse as a fig leaf for their real concerns, they end up sounding much less credible than if they just tried to argue for what they actually meant.
But we do not have experience with the kinds of problems we’ll face if we build machines that are smarter and more capable than us, so I feel much less sanguine that an adaptation-based strategy will help us there.
The current status quo—where new models that push the frontier are kept private, and openly available equivalents are only released after a lag of months or years—is valuable.
We should think of the gap between frontier models and proliferated models as an adaptation buffer: a limited time window when we know what bad actors will soon be able to use AI for, which gives us a chance to implement defensive measures that increase society’s resilience to the danger.
But with AI, it’s different: it’s often only a couple of years between when you need a world-class computing cluster to do something and when you can run it on a high-end gaming chip on your laptop.
These systems will be absolutely central to the economy, technology, and national security, and will be capable of so much autonomy that I consider it basically unacceptable for humanity to be totally ignorant of how they work.
The stereotype is that new managers often micromanage, but at Wave, I've noticed that new managers often err on the side of undermanaging, or being too hands-off, perhaps out of a desire to signal that they trust their reports.
you should expect to have something like 10-20% less individual output per engineer you manage, depending on how experienced you are, they are, etc., and with substantial week-to-week variance.
The short, and very much un-sweet, vision of a tariff-bound economy is that it is stochastic. We can observe it, measure it, analyze it, but we cannot really predict it.
The old relationship we had with the United States based on deepening integration of our economies and tight security and military cooperation is over.
Even if we assume all three of these crashes were Waymo’s fault, that would still mean that a large majority of the 38 serious crashes were not Waymo’s fault.
Say goodbye to cycles of mock and doc feedback, and say hello to code-first prototypes with constant iteration (on prompts, models, etc) to answer three key questions.
When I have conversations with breached companies, my messaging is crystal clear: be transparent and expeditious in your reporting of the incident and prioritise communicating with your customers.
Summarizing all of that into my advice right now: Get Whoop for 9.5/10, reliable, convenient sleep tracking with an excellent app (once you get to know it a bit). Get Oura for 10/10 tracking, if you're ok with the ring form factor.
Summarizing all of that into my advice right now: Get Whoop for 9.5/10, reliable, convenient sleep tracking with an excellent app (once you get to know it a bit). Get Oura for 10/10 tracking, if you're ok with the ring form factor.
I didn't find 8Sleep to be very reliable in its sleep tracking. The scores don't make as much sense to me when I wake up, and as we saw above they don't correlate very strongly with Whoop or Oura.
And number two, the superhuman opportunity deserves everyone who works at the company to spend as much time as possible in their zone of genius. And so that includes me as well as everybody else. So what I did is I hired a really great president. I went from eight direct reports to two.
But this is the hard part. When you're when you're doing that market widening, you're not solution deepening. And so your perceived product velocity may decrease. You can avoid some of these things with some smart technology decisions, but mostly you just have to grind through it and it is worth it to get to the other side.
you can classify anything that you build in a company into one of two categories. Solution deepening and market widening. Now solution deepening means making your product better for its existing users but not making it available to more users. Whereas market widening means making your product available to more users but not making the product itself any better.
It's like trying to find meaning in your job. You can find satisfaction in what you do. That is a very good thing. You can find satisfaction and be happy with what you've created. You can be thrilled by the experience, but you cannot find... I doubt you can find purpose.
You don't stop a problem by replacing it. And so I didn't have a girlfriend, I didn't have all that. I just realized that I was truly taking away from something from my future wife.
But for me it just comes down to, is it really a good thing to do? Is it really actually something we want, is to value people in such a profane or just disregarding way? I just really think it's just bad for the soul. Even if all the stats said it was great for you, I still say it's actually bad.
And so we often talk about programming in perspective of web, or something that's pretty narrow, and I think that's just a social construct of Twitter more than anything else, that actually I don't believe it's that representative of the entire programming world out there.
And for me, that is the worst part about programming, is when you already know the solution and it's just a matter of how fast you can type and get it out from your head to your hands.
A series of autobiographical essays exploring wandering, uncertainty, memory, desire, and place—embracing the unknown as a gateway to self-transcendence.
Ben's personal reflections on reinventing himself at "midlife" (though relatable at any age). I love how he leaned into a new version of himself without dramatically blowing up his life.
This book is a short but inspiring take on living life on your terms while being pragmatic about money and how to make it happen. Reminds me of a less travel-centric Vagabonding.
Part memoir, part raw, honest reflection, part useful ideas for a pathless life. I loved how this book didn't follow a non-fiction formula. His 'on-season'/'off-season' approach was useful for me.
UX design has been steadily improving over the last 50 years and is now reasonably good. The low-hanging fruit of terrible usability has long since been picked.
The new trend toward vibe coding and vibe design may be upending the user-centered design paradigm that has remained based on the same ideology since the first UX design projects at Bell Labs, starting in 1947.
Now that we have the answers, it all looks very obvious, and mostly straightforward.
Their words now
I should have been pointing out red flags starting back at WWDC last year, and I am embarrassed and sorry that I didn’t see what should have been very clear to me from the start.
I'm hopeful this framework will turn out to be a much more robust style for writing performance-sensitive code, especially over time and as compilers evolve.
However, with so many different software projects out there, each moving so rapidly and depending on and being used by so many other projects, it becomes practically-inevitable that some regressions "like that one" happen, almost constantly.
It's easy to get impressive-looking results if you're comparing against a poorly-tuned baseline, and that observation turns out to explain a surprising fraction of supposed improvements.
In a vibe coding future, companies will hopefully invest more in understanding user needs, refining the interface, and polishing details that delight users, because those are harder for AI to get right without guidance.
As a Python and JavaScript programmer my favorite models right now are Claude 3.7 Sonnet with thinking turned on, OpenAI’s o3-mini-high and GPT-4o with Code Interpreter (for Python).
But the underlying ideology is familiar from Silicon Valley, Mark Zuckerberg's dictum to "move fast and break things" and replace the messily human with frictionless technology, built by elite engineers who are supposed to know better than anyone else.
(publication date March 18, 2025, Viking) by Laura Delano is a compelling and troubling memoir of psychiatric patienthood, iatrogenic harm, and finding a meaningful and flourishing life outside of the mental healthcare system.
Contrary to some people’s beliefs, companies do not in fact have sufficient incentives to mitigate all major risks, and competition is driving corner-cutting that needs to be reined in somehow.
This is what leads to slop JS/TS software. Treating errors generically like this is a mistake. It leads to fragile software, poor telemetry/observability/debugging, and as a result worse experiences
The potential end of USAID is an unfolding public health and humanitarian disaster, but it does not permanently doom the task of growth—in large part because these aid flows were far too small to begin with.
The longevity of that consensus is itself remarkable: in 79 years of data, Republican presidents spent an average of $34 billion a year on aid, while Democrats spent $40 billion.
Even as GDP has grown tenfold, US aid budgets have remained basically flat in real terms—the result (one might guess) of a delicate balancing act between bipartisan consensus among elites and a wariness of attracting too much negative attention.
As I set out last time, the usual story — that this was caused by deforestation, making firewood so scarce that people resorted to burning an inferior but cheaper fuel — simply does not stack up.
England’s deforestation — along with the draining of its peat-filled fens and marshes, and the clearing of its gorse-growing heaths — was instead itself caused by the arrival of cheap coal.
Far from being the laboratory that brought coal into the home, then, London was actually a laggard — all the more so given it had been buying coal from Newcastle for centuries.
Many people assume that they need to keep their jobs forever, because it is their source of health insurance. The good news is, this is usually not true.
I recently used Peddle with great results, and have been recommending them to anyone who has a low-value, possibly even non-driveable car they just want GONE in exchange for $500-1000 of cash.
Small day pack. A lightweight pack is great for carrying a sweater, camera, guidebook, and picnic goodies while you leave your large bag (with most of your belongings) at the hotel.
Disposable washcloths that pack dry but lather up when wet (such as Olay's Daily Facials 5-in-1) are another option; cut them in half to make them last longer.
I recently interviewed Mike Huemer on his new Progressive Myths. To repeat, I consider it “the best book on wokeness.” You know you’re reading a special book by page 3
if you believe we're in this sort of stage of economic growth and change that we've been in for the last 20 years, the export controls are absolutely guaranteeing that China will win long-term.
To some extent, training a model does effectively nothing. They have a model. The thing that Dario is sort of speaking to is the implementation of that model, once trained to then create huge economic growth, huge increases in military capabilities, huge increases in productivity of people, betterment of lives.
I think it's even more impressive what OpenAI did in 2022. At the time, no one believed in mixture of experts models at Google who had all the researchers. OpenAI had such little compute and they devoted all of their compute for many months, all of it, 100% for many months to GPT-4 with a brand-new architecture with no belief that, "Hey, let me spend a couple of hundred million dollars, which is all of the money I have on this model." That is truly YOLO.
There's not many worlds where China cannot train AI models. I think export controls are decapping the amount of compute or the density of compute that China can have.
Accepted practice is that for any given model that is a notable advancement, you're going to do two to 4x compute of the full training run in experiments alone.
This is why you want to work in post-training because the GPU cost for training is lower. So you can make a higher percentage of your training runs YOLO runs.
The scale word gets a lot of attention in this. The interpretation that I use is effectively to avoid adding the human priors to your learning process. And if you read the original essay, this is what it talks about is how researchers will try to come up with clever solutions to their specific problem that might get them small gains in the short term while simply enabling these deep learning systems to work efficiently, and for these bigger problems in the long term might be more likely to scale and continue to drive success.
And we'll get into the details of the models and again and again as we try to get deeper into how the models were trained, we will say things like the data processing, data filtering data quality is the number one determinant of the model quality.
GPT-2, R1, and nearly every other AI model released since 2018 have all been part of a consistent story: AI capabilities that rival and ultimately exceed human intelligence are easier and cheaper to build than almost anyone can intuitively grasp, and this gets easier and cheaper every month.
AI competition needs to be tempered with a recognition that China isn’t going away as a player in AI, and that coexistence must also be a part of America’s strategy.
Each and every time, it took at most a few years but typically months before other companies would achieve the same level of performance using a broadly similar, and always relatively simple, recipe.
Re-read because I couldn't interest myself in a couple of other pieces of mind candy and thought "you know who did this sort of thing well...". Which got me thinking, again, about the relationship between this series and feminism.
First and foremost, I use this to talk to local models hosted by Ollama, but secondarily I also use it to interface with other remote services like OpenAI, Anthropic and DeepSeek.
It is a fork of Visual Studio Code with AI-based auto completion and code generation built-in. It’s my go-to for programming with AI assistance at the moment.
This is a command line tool with plenty of plugins that lets you prompt different models. Think of it as a command-line version of Open WebUI. It’s particularly useful for quick scripting and basic automation.
Like probably most AI users, I use ChatGPT, particularly on my phone. I pay for the Plus subscription because I use it enough to get a lot of value out of it.
This allows me to run models locally on my MacBook Pro M1 Max. With the 64GB of RAM it has, it’s a pretty potent machine for basic inference despite it being three years old.
Where housing costs are moderate, friends and family have bigger homes. When they are higher, friends and family don’t have space to share, and this is often what puts a vulnerable person onto the streets.
Overall, though, this experience reinforced my belief that the tooling and interface design around LLMs is lagging way behind the actual capabilities, and is an area of active experimentation and development, even aside from any future model improvements.
Thus, code is cheaper than ever, but I suspect that insight and good architectural design and understanding, at least for now, will become more valuable than ever.
and really only has one model you care about - Claude 3.5 Sonnet. But Sonnet is very, very good. It often seems to be clever and insightful in ways that the other models are not.
if you want a very good all-around model with excellent reasoning. As an open model, you can either use it hosted on the original Chinese DeepSeek site or from a number of local providers.
currently has the best Live Mode in its Advanced Voice Mode. The other big advantage of ChatGPT is that it does everything, often in somewhat confusing ways
Gemini’s advantage is a family of powerful models including reasoners, very good integration with search, and a pretty easy-to-use user interface, as you might expect from Google. It also has top-flight image and video generation.
Deborah Kasdan’s Roll Back the World: A Sister’s Memoir (2023) is an honest and vulnerable account of her sister Rachel’s life with schizophrenia and the challenges this presented for the family.
After years and years of illegal and unconstitutional federal efforts to restrict free expression, I also will sign an executive order to immediately stop all government censorship and bring back free speech to America.
America will be a manufacturing nation once again, and we have something that no other manufacturing nation will ever have — the largest amount of oil and gas of any country on earth — and we are going to use it.
The inflation crisis was caused by massive overspending and escalating energy prices, and that is why today I will also declare a national energy emergency.
Growth team can optimize. Growth can maybe lift it by 10, 15% maybe that's enough for you even that like is on the upper end of what growth team would be able to do if there is a slow down trajectory
If you have the overall business slowing down, your head of growth is destined to fail because the reason business is slowing down is much deeper than not having a growth team.
It depends on how you're actually going to collect the money. If your money collection is going to happen through sales team primarily, then you absolutely should be hiring sales way before growth.
I really believe that the founder led growth is not being popularized enough that you do not need growth teams until you actually can start running experiments on your user base
Milton Friedman is very open to doing things differently in a state of emergency. He will have different ideas during World War II than any other time and that's why I argue I think he would have been supportive of at least the first rounds of coronavirus relief because I think he would have put his emergency thinking hat on.
Friedman and Schwartz becomes the playbook for the Federal Reserve. We have lived through this, the financial crisis. The Federal Reserve is ready to loan. Covid, the Federal Reserve does all kinds of new things, because no Federal Reserve chair wants to be in Friedman-Schwartz 2.0 that somebody writes, where they're the bad guy who let the economy meltdown.
So, I think she comes back in to relevancy in a different way than Friedman does, because I think in some ways she's tapped into a more universal human longing for independence and autonomy and self-creation and self-discovery.
So, she thought of herself as rational. She thought of rationality as what she was doing, but she was actually doing a kind of mythopoetic psychological work as well.
Milton Friedman, he also does tend to have friends who agree with him, yet he's always willing to debate his opponents, and he's willing to do so with a smile on his face. He's a happy warrior, and he actually will win a lot of debates simply by his emotional affect and his cheerfulness and his confidence, where Rand will lose debates because she gets so angry in the face of disagreement.
This is perhaps one of the first painful lessons anyone who has built an AI product quickly learns. It's easy to build a demo, but hard to build a product.
If you're a guy in your early 20's, get married and have kids. Go into debt if you have to.
Their words now
Most people should get married and have kids much earlier than the current average (and have more kids than the current average). Doing this successfully (without screwing it up with selfish pride) will make the world better and make them happier than anything else they’re doing.
Most people should get married and have kids much earlier than the current average (and have more kids than the current average). Doing this successfully (without screwing it up with selfish pride) will make the world better and make them happier than anything else they’re doing.
What prevents this from being a broad-based bubble is that we are missing a critical feedback loop from the announcement's impact, like network growth or land value.
And in a way that’s the irony of AI… It will make public services more human… Reconnect staff with the reasons they came to public service in the first place…
This is the global race of our lives. Now, some countries are going to make AI breakthroughs and export them… Others will end up buying those breakthroughs and importing them. The question is – which of those will Britain be? AI maker or AI taker?
Since inflation is not primarily driven by private-sector lending in the current environment, making borrowing more expensive will not significantly cool the economy.
Structural fiscal deficits have surpassed private sector lending and monetary policy as the primary drivers of economic activity and inflation, marking a fundamental shift in the economy's liquidity dynamics.
Stabilizing factors such as the global demand for U.S. dollars and debt denominated in its own currency suggest an era of fiscal dominance in the U.S. that's less dramatic than alarmists predict but more persistent and intractable than optimists hope.
Planning, at its core, is a search problem. You search among different paths towards the goal, predict the outcome (reward) of each path, and pick the path with the most promising outcome.
I suspect that both Substack and OnlyFans may become a Girardian bubble. Everyone joins after hearing from the few people who succeed; the thousands who slink away after months of effort netted them a mere 17 subscribers usually do so quietly.
I read that last month, and visited Venice while I was reading it, and I was struck by the extent to which medieval Venice was a well-oiled political economy machine.
Q3 2024 had the rate 8.9% so there is no way it can reach 30% in 2027. I was way too optimistic at a time when EV enthusiasts thought I was horribly pessimistic.
Waymo now has a service that looks and feels like this in San Francisco, 8 years earlier than I predicted. But it is not what every one was expecting. There are humans in the loop.
There’s still plenty to worry about with respect to the environmental impact of the great AI datacenter buildout, but a lot of the concerns over the energy cost of individual prompts are no longer credible.
Our Share of Night by Mariana Enríquez - I was struck to the bone by this book. I thought about it for many weeks after I read it and even had several dreams about it.
The Rattle Bag edited by Seamus Heaney and Ted Hughes - In an age of LLM-generated poetry and machine learning curation (even coming from yours truly sometimes), it's refreshing to have real people who love poetry curate it and show you what you need to read to touch grass.
The Tombs of Atuan - My only regret is that I came to Ursula K. LeGuin so late in life, and this might be my favorite of all of her books I've read so far.
I picked this book up, and I liked it. On the surface, it's about wizards and magic, but there's more to it. Every now and then, Ursula Le Guin sprinkles in some wisdom.
Some of his essays are thought-provoking. His reflections on education, nationalism, and other topics are not only stirring but also showcase literary brilliance. The man could write.
It's stimulating work as well as easy to read. Although parts of South Asia are not commonly known to have an argumentative gene, Amartya Sen convincingly shows that Indian history is deeply rooted in an argumentative tradition. He masterfully weaves evidence and tales to make his point.
We ought to be especially careful in the cases where what we delegate to a device, app, agent, or system is an aspect of how we express care, cultivate skill, relate to one another, make moral judgments, or assume responsibility for our actions in the world
A simple and tested method for communicating and connecting with anyone, Supercommunicators is an exploration of the psychology and neuroscience of communication — and how to use those skills at work, home and in life.
My first book, which spent three years on New York Times bestseller lists, takes us to the thrilling edge of scientific discoveries that explain why habits exist — and how they can be changed.
My second book is about the science of productivity, and how to succeed with less stress and struggle, to get more done without sacrificing what we care about most.
And what they found was if that search space, if the sky is an ocean and you're looking for fish, how much of the ocean have we looked at, and it turns out to be a hot tub. That's how much of the ocean that we've looked up. We've dragged a hot tub's worth of ocean water up and there was no fish in it
So, we'll call that the indirect Fermi paradox and there absolutely is no indirect Fermi paradox for the most mundane of reasons, which is money. There's never been any money to look.
But I felt like what this paper showed was that the burden of proof is now on the pessimists. So, that's why we called it the pessimism line. Throughout history, there's been alien pessimists and alien optimists, and they've been yelling at each other, that's all they had to go with.
And so, what we found was basically that there have been 10 billion trillion habitable zone planets in the universe. And what that means is that those are 10 billion trillion experiments that have been run. And the only way that we're this whole process from a biogenesis to a civilization has occurred is if every one of those experiments failed.
In which case, those kinds of collisions are so rare that you would expect one in a trillion stars to have planets. Instead, every star in the night sky has planets.
Best books I read in 2024: Just a note that these books weren’t all necessarily published this year (though some were). Father Time: A Natural History of Men and Babies by Sarah Blaffer Hrdy
Best books I read in 2024: Just a note that these books weren’t all necessarily published this year (though some were). Father Time: A Natural History of Men and Babies by Sarah Blaffer Hrdy Just as Deadly: The Psychology of Female Serial Killers by Marissa A. Harrison
Best books I read in 2024: Just a note that these books weren’t all necessarily published this year (though some were). Father Time: A Natural History of Men and Babies by Sarah Blaffer Hrdy Just as Deadly: The Psychology of Female Serial Killers by Marissa A. Harrison Eve: How the Female Body Drove 200 Million Years of Human Evolution by Cat Bohannon
This "memorize, fetch, apply" paradigm can achieve arbitrary levels of skills at arbitrary tasks given appropriate training data, but it cannot adapt to novelty or pick up new skills on the fly (which is to say that there is no fluid intelligence at play here.)
Their words now
OpenAI's new o3 model represents a significant leap forward in AI's ability to adapt to novel tasks. This is not merely incremental improvement, but a genuine breakthrough, marking a qualitative shift in AI capabilities compared to the prior limitations of LLMs.
Passing ARC-AGI does not equate to achieving AGI, and, as a matter of fact, I don't think o3 is AGI yet. o3 still fails on some very easy tasks, indicating fundamental differences with human intelligence.
o3's improvement over the GPT series proves that architecture is everything. You couldn't throw more compute at GPT-4 and get these results. Simply scaling up the things we were doing from 2019 to 2023 -- take the same architecture, train a bigger version on more data -- is not enough.
Earlier this year I started writing a piece on why “hire great people and get out of their way” is such terrible, dangerous, counterproductive advice to give anyone in a leadership role.
Either can work. Both have tradeoffs and implications. If you try to import either philosophy wholesale, it will break in unexpected ways; if you try to mix and match, it will probably be an unfettered nightmare.
You should ALWAYS have as few employees as possible. Always. Hiring more people should never be the first lever you reach for, it’s what you do after exhausting your other options.
A korrent is a belief a person has stated in their own words: one
sentence stating the claim, backed by a quote and a source, kept at
korrents.com.
Under a name here, the quoted block is what they actually said.
The korrent beneath it is the claim those words support, in
korrents' wording — tap it to see the record, its source, and who
else holds it.
Nobody here wrote their own korrents. They are compiled from public
statements, and a person can change their mind, which is recorded too.
About the English under a post
Some people here publish in a language other than English. Where they
do, this site shows a machine translation beneath the post, in
this typeface — the site's own, not theirs.
The post itself is never changed, moved or hidden: what is set in the
serif above is exactly what the person published, and it is what to
quote them on. A translation can be wrong in ways that matter,
especially about tone.
Only the post's own words are translated. A quoted post, a linked
article and a belief on korrents.com
are left in their original language.