I keep landing on the same question. We feel hungry all the time and there is more food on the planet than at any point in human history. We feel overwhelmed all the time and there is more information than at any point in human history. The two are not separate problems. They are the same problem, which is that we crossed out of scarcity and into suffocation somewhere along the way and we did not notice.

I had dinner at a friend’s place last month. Six starters before the main course arrived and I ate all of them because they were in front of me and they were good, and somewhere in the middle I lost track of whether I was hungry. The meal kept arriving and I kept saying yes because that is what we do at dinner. I walked home feeling force-fed, and I kept thinking about Paracelsus, who said five hundred years ago that the dose makes the poison. Nothing is toxic in itself. Water will kill you if you drink too much of it. Oxygen will kill you if you breathe too much of it. The difference between medicine and poison is quantity, and we have built a civilization on quantity without asking when quantity became the problem.

I think about this with sugar. Sugar used to be the most precious thing on the planet. The Greeks called it Indian salt and they would not have known what a croissant was. The British spent centuries at war partly to control its trade. In Bengal where I grew up my grandmother kept a small jar of it on the highest shelf and we used it sparingly because it was expensive. Now the same sugar is in everything. The bread, the cereal, the sauce, the supposedly healthy granola bar that is mostly sugar held together with oats. Diabetes is one of the leading killers of our age. The thing that is killing us was once the most expensive thing we could buy. What was a luxury became a weapon because of how much of it we now eat, and that is what Paracelsus was warning us about, and that is what our grandmothers understood without needing a 16th century alchemist to explain it.

Information is the same story, just faster. When I was growing up in Bengal, the way you learned anything was that someone older told you, or you read a book you had to walk to the library to find. The information had a shape and a weight and a cost. You had to earn it. Now the information arrives in your pocket whether you asked for it or not. Every time I take out my phone the screen lights up with things I did not request. The world is throwing facts at me the way the dinner table threw starters at me, and I have started noticing that the more information I consume the less I retain and the more opinions I read the less I think. The ancient Greeks inscribed the answer at the temple at Delphi. Meden agan. Nothing in excess. They carved it into stone because they understood that absence rarely kills. Abundance does. The body and the mind were not built for overabundance.

Socrates used to say that contentment is natural wealth and luxury is artificial poverty, and the line has lived in my head for years. We were wired by millions of years of scarcity. We do not know what to do with infinite supply. In our evolutionary history, infinite supply meant a tree full of fruit that was about to be eaten by something else, so we gorged and we saved and we kept gorging and saving long after the supply stopped being scarce. We call the gorging lifestyle. We call the saving investment. The body just knows there is food in front of it. The brain just knows there is information in front of it. The rest of the system is screaming because this is not the world it was built for.

Seneca wrote two thousand years ago that the man who craves more is the one who is poor, not the man who has too little. Seneca was writing from a Roman villa with running water and fourteen slaves, which is to say the disease of excess predates capitalism, predates the internet, predates almost everything we blame it on. The Stoics figured out that the problem has always been wanting too much. They built their whole philosophy around the idea that the good life is the life calibrated to the amount of things you can actually use, and the life with the most things is rarely that. I think about this every time I open my phone. Unread notifications, unread emails, unread newsletters. I live in the unread, and the unread is growing faster than I can shrink it.

The Delphic priests had it right. Three maxims at the temple. Know thyself, nothing in excess, certainty brings ruin. The middle one is the one we have completely lost. You can see the absence of it in diabetes, in information overload, in the obesity epidemic, in the attention crisis, in the loneliness epidemic. All of them just different costumes for the same disease of more. The remedy is the same one that worked for the Stoics and the grandmother in Bengal and the Greeks at Delphi. It is not less ambition or less drive or less curiosity. It is the muscle of stopping, and that muscle has atrophied in us because the world punishes stopping and rewards continuation.

Epicurus said something I have been turning over for months. We should remember that what we now have was once among the things we only hoped for. The line reads like a prayer at first, like something a wise man murmured to his friends, and then you live with it and you realize it is one of the most violent sentences in the tradition. The thing you once prayed for is in your pocket. The thing you once walked a hundred miles for is in your refrigerator. You are still not satisfied, and you will not be satisfied, because the capacity for being satisfied got replaced by the capacity for being dissatisfied, and the replacement happened so gradually you did not notice. A kettle does not notice the water is boiling until the steam is in your face.

I have not figured it out. I ate six starters at dinner, I read forty newsletters this morning, and seventeen browser tabs have been open since last week that I will not return to. Moderation is a practice, not a destination. It is something you do today, fail at tomorrow, and do again the day after. The world is not going to get any quieter or any less full. The world is going to get louder and fuller. The only question is whether you will build the muscle to put the phone down before the seventh starter arrives, or whether you will let the abundance eat you. The abundance will eat you if you are not paying attention, and we are not, most of the time. That is the whole problem.

Tell me where you land on this. I am @troysk704.

I keep landing on the same frustration. The people who have the most to say about what is good are usually the people who know the least about it, and the more I trace why this is the case the more it points back to the same place, which is the gap between who is talking and who has actually done the work. It started for me in a HSR co-working space last year where I sat close enough to read a guy’s screen in a Uniqlo polo and he spent three hours on a single slide for a millet snack brand. The line on it was something like this is not just a snack, it is a return to ancestral wisdom and a quiet rebellion against industrial food and a philosophy of eating the way your grandmother did. I could not stop thinking about it. The product was roasted ragi tossed with ghee and salt, the kind of thing every grandmother in Karnataka has been making for free her whole life, and a roomful of people had decided that the way to charge a hundred rupees for thirty grams of it was to invent a mythology around it. The move only makes sense if nobody in the room has ever been hungry enough to just eat the ragi plain, which is probably true, and the part that should worry us is that the same playbook is being applied to everything now. Every commodity wrapped in the costume of a movement so the price can go up and the buyer feels like they joined something.

A friend showed me three Shopify stores last month, all selling the same bundle of incense sticks made in a factory near Mysore. The first one was calling it a wellness ritual at four hundred rupees for a hundred sticks, the second a mindfulness tool at five hundred, the third ancient aromatherapy at six hundred. Same product, same factory, three different costumes. The whole thing was a Canva logo and a Razorpay key, no maker, no factory floor, no actual ancient wisdom, just a target demographic and a Pinterest board. An influencer burned it in their reel and the agarbatti borrowed a mood for the price of a sponsored post. That is the whole economy in a single transaction.

MrBeast Burger launched in fifty states at once and it was not even a real company. It was a virtual brand running out of other people’s kitchens, a child of a service that builds delivery-only celebrity culinary brands, which is a sentence that should not exist but does because the influencer had enough reach to make the marketing work without the supply chain having to make anything real. You get a burger with his name on it that is not his, made by people he has never met in a kitchen he has never visited, and the whole illusion holds because the platform gave him the audience and the audience gave him permission to be the brand. The burger might not even be a burger anymore. It might just be a notification that something happened.

The pattern is the same every time. The influencer is the last mile of the supply chain, not the source of the thing, and the marketing copy needs them to look like the source so the machine can pretend it is making culture when it is just moving boxes. Calling it out would mean admitting that the curator is a courier, and a courier is not a curator. Those are different jobs. One needs taste and the other needs a follow count, and we have been confusing the two for a decade.

Culture had gatekeepers once. I am not romanticizing that era, it had plenty of problems of its own, but the gatekeepers at least had to have spent years inside the thing they were gatekeeping. You could not become the voice of a generation at twenty-three if you had not made anything yet. An editor developed a real ear through years of actual editing, a gallerist learned to spot a bad show on sight by walking enough rooms, and a critic earned the right to an opinion by reading more than anyone was paying them to. Their opinions carried weight because they had paid the cost of being wrong many times. The cost is invisible on a feed. The feed only rewards momentum.

The feedback loop got hacked. The old one was slow and imperfect and full of bias, but creators put work out and audiences responded and critics interpreted and over time some kind of consensus formed about what was good, and the people who had done the work had room to be heard. The new loop collapsed all of that into one signal, which is engagement. Engagement does not care whether you have read the book, it cares whether you can hold attention for fifteen seconds, so the people who win the new loop are the people who are best at performing the appearance of expertise, which is not the same skill as having expertise. We have lost the ability to tell the difference because the loop rewards the performance more than the thing itself.

Every endorsement an influencer makes is compromised the moment it leaves their mouth. Their livelihood depends on the next post, the deal pays for the post, and honesty would cost them the deal. The whole machine runs on the gap between what is being said and what is true, and that gap is where you and I are supposed to live, on the receiving end, scrolling through it.

Most influencers are not malicious. They did not set out to do this. The problem is that the system picked for being visible, not for being good at anything, and now we have a culture shaped by the most skilled performers of attention in the room, most of whom have not actually tried the things they are shilling. They have no firsthand experience with the place or the book or the restaurant or the routine. They have only the brief they were handed and the check that came with it, and looking alive is the only thing they have been paid to do. The recommendations feel hollow because they are hollow.

I call influencers the flu on purpose. The flu does not kill you on its own and it does not rob you. It lowers your defenses so the real infections can take hold, and that is what is happening to culture right now. The influencer is the gap in the wall, the lowered defense, the thing that lets a hundred low-quality signals pour through and settle in the lungs of whatever used to make the conversation worth having. After long enough on the low-grade infection you stop noticing the symptoms because the symptoms become the weather. Then one day you cannot remember what it felt like to trust your own taste because you have been outsourcing it to a feed for so long that your taste is just a list of things other people told you to want.

Do not take an influencer’s word for it. Not because they are lying, most of them are not, but because the system that produces them does not allow for the kind of judgment you actually need. The system rewards volume and visibility and being relatable, and none of those are the same as knowing what is good. The shortcut feels free but it is the bug, and the moment you take it you have already become a node in the network that is feeding on your attention the same way the feed is feeding on theirs. The whole thing is a closed loop, and you are inside it whether you wanted to be or not.

Find me on Twitter if you want to argue with any of this. I am @troysk704.

Eight months ago I posted something small on LinkedIn that I did not think much of at the time, and it has since turned into one of those posts that comes back with interest because the world has caught up to what I was quietly thinking. I had just done my routine thing of putting a chunk of money into US stocks which I do every few months, and on that particular day I thought I would try something new and ask an LLM for recommendations, which was a hypothesis I was about to abandon because the picks were bad. I remember MKTX and TRU and MRK in there and the honest part is I do not remember the rest because I closed the tab and went back to code, and when I came back I did the only thing that has ever reliably worked for me which is to ask what I would do with the money if I were not allowed to use any tool at all and had to think like a human being for fifteen minutes.

What came out of those fifteen minutes was a line I posted without much ceremony, which was that I had decided to put most of the new investment into GLP-1 related companies, and I said it the way you would mention a restaurant you wanted to try, and underneath that casual sentence was something I had been chewing on for at least a year. I had been reading the retatrutide literature, the actual trial data rather than the press releases that summarize it, and what I was reading did not feel like a small improvement on existing drugs because semaglutide and tirzepatide are already remarkable and you do not usually get a third step that is meaningfully larger than the second. The bet was small in the scheme of things but it was deliberate, and the conviction I was putting into the trade was the same conviction I had been carrying around about my own body and my own hopes for the next ten years. Putting my money where my body was.

I have not taken retatrutide yet because it is not approved, and the molecule has already helped me in a way that has nothing to do with my blood sugar, because it has clarified what I think the next decade of medicine is going to look like and I have arranged my reading and my small portfolio around that clarity. A kind of help that does not require a prescription. The TRIUMPH-1 data came in a few weeks ago and it confirmed the bet, and I wrote about the science back in May. The point of this post is the knowing, not the science. The knowing arrived first and the data arrived second, and that order matters to me more than it probably should.

What I want to say now is that retatrutide is going to help millions of people, and I am not saying that the way a guy on a podcast says it when he is pumping a stock. I am saying it the way I mean it when I say the air in Delhi is going to help millions of people once the city stops burning its fields. My friend has been avoiding mirrors for five years, and one morning she is going to look at one without flinching because a molecule did its work, and my uncle whose knees have hurt since his forties is going to take the stairs without thinking about which leg he leads with, and a teenager whose liver is inflamed from a decade of sugar is going to wake up at twenty with a clean scan and never know the difference between a body that was saved and a body that was always going to be fine. None of those people are going to credit a triple agonist in their private thoughts, and the molecules that do this for them do not care whether we feel guilty about needing them, and that indifference is going to be one of the most compassionate things that has happened to public health in my lifetime.

If you want to push back on any of this I am @troysk704.

I keep seeing AI systems get vision capabilities and the pattern is always the same which is you drop an image in and the model describes what it sees and this has been working for a while now even for open models and even for video and so when I read about a new technique from DeepSeek that adds vision to an AI system my first reaction was so what because vision is not the bottleneck anymore. But then I actually read the paper and it turns out they are not just adding vision and they are changing how the AI sees and that is the part that matters because the old approach to visual reasoning asks the model to describe what it sees with words and if you want to count the number of strawberries in a bowl the model has to say something like there are strawberries clustered on the left and a few scattered near the rim and some are partially hidden behind others and some have their stems visible and at the end you cross your fingers that it got the number right and you cannot really check its work because the whole process happened in a black box of language. The new approach does something different and it is so obvious in retrospect because when you or I want to count things in an image we do not write a paragraph about it and we point at each thing with our finger and we count one two three and we are done and instead of describing the image like a poet the AI points like a human and it gets to use visual primitives like coordinates and bounding boxes and trajectories as part of its thinking process which makes it more accurate and also faster because describing stuff with words is a wasteful intermediate step.

The numbers are genuinely surprising because this free system uses about ninety percent fewer visual tokens than most frontier models and it still matches or beats them across seven benchmarks and the paper explicitly excluded their own in-house benchmarks which matters because the easiest way to win a benchmark game is to just invent a new benchmark that you happen to be good at and they did not do that and they let the existing benchmarks speak for themselves and the results held up. This is free and open research which means the technique can be added to existing models and it is described as a blueprint rather than a finished product and I think that is the right framing because the real contribution is not a specific model checkpoint but a way of thinking about visual reasoning that makes the model both cheaper and better which is the kind of tradeoff you do not see very often.

The mechanism behind it is called policy distillation and the idea is that you have a bunch of expert AI models where one of them is great at drawing bounding boxes and another is great at tracing paths through a maze and you train a student model that learns from all of these teachers at once so when the student says here is what I would try the teachers say here is what I would have done instead and after enough iterations the student internalizes all of these different visual skills into a single model. This is why they call it distillation because you are condensing the knowledge of multiple specialists into one generalist and the result is a model that can not only answer questions about images but can trace its own reasoning visually and show you where the crown connects to the octopus and you can see the path it took instead of just trusting the final answer and when something goes wrong you can find the mistake and fix it which is a huge step toward AI systems that we can actually understand.

There are limitations though because the model does not automatically use this kind of pointed thinking and it needs a word as a cue to start reasoning visually and bounding boxes work well for people but if you are trying to count blades of grass or strands of hair then the low resolution becomes a problem and thin structures are always the thing that breaks these visual systems and the topological reasoning does not generalize as reliably as you would want when you show it something completely new. But I feel like this is one of those papers that shifts how people think about the problem and that is happening a lot lately in AI research because we always assumed that making models smarter meant giving them higher resolution images and more pixels and more data and it turns out that sometimes the right move is to give them better tools for thinking about what they already see and less can be more when the less is designed well.

This is also part of a broader shift that I care about which is that the large AI companies are going to IPO and they will become ventures that need to maximize profits every quarter and the open models with free weights become more important every time a company closes off another piece of the stack and this paper makes those open models better for free and it describes the method in enough detail that other researchers can build on it and that is the kind of research I want to see more of.

Find me on Twitter if any of this connects. I am @troysk704.

I was listening to an interview with David Reich the geneticist who runs the ancient DNA lab at Harvard and he said something about the Indian caste system that stopped me because it is one of those things that you think you understand until you realize the actual evidence is sitting in everybody’s DNA and it tells a story that is nothing like what you were taught. The standard model of Indian history is that the caste system is ancient and maybe even eternal and that it grew organically out of the Vedic period and has been slowly evolving for three or four thousand years but the genetic data says something much more specific and much more startling. What Reich and his team found is that almost everyone in South Asia today sits on a gradient between two ancestral populations which they call Ancestral North Indians and Ancestral South Indians and if you plot everybody on this gradient it looks exactly like what you would see if you sampled African Americans because African Americans are also a population in the middle of mixing between two very different source populations with some people having more European ancestry and some having more West African ancestry and the gradient tells you that mixing happened recently and then stopped. In India the mixing happened and then it froze.

The data shows that the three primary ancestral groups of India came together around the time of the decline of the Harappan civilization about 3800 years ago and there was a local hunter-gatherer population and a farming population and then the steppe pastoralists from the Yamnaya culture who had migrated across Eurasia and when these three groups met there was a period of convulsive mixing where they formed the two mixed populations we call Ancestral North Indians and Ancestral South Indians and then the mixing stopped about 2000 to 3000 years ago and the reason it stopped was the caste system which locked people into endogamous groups that did not intermarry across boundaries and this freezing is visible in the genome because instead of collapsing to a single blended population like what happened in Europe after similar mixing events you see a preserved gradient that has been stable for two thousand years. The DNA gives you a snapshot of the moment the system crystallized and that snapshot has not changed since because people who were in different caste groups simply stopped exchanging genes with each other and the proof is that the gradient is still there just as sharp as it would have been two millennia ago.

The caste system is not a slowly evolving organic thing in this reading and it is not an ancient unchanging tradition either because what the genome shows is a specific historical moment when a society that was mixing freely suddenly locked itself into a rigid structure and preserved the genetic snapshot of that moment for thousands of years and you can read it in the DNA of every Indian alive today even people who think caste does not matter anymore because the signature is in the frequencies and the proportions and the way the ancestry clusters and it does not lie. The same data also reveals a strange exception where the Patel community from Gujarat sits off the main gradient because they have additional ancestry from Central Asia which is a reminder that the freeze was not absolute for every group and that the history of South Asia is even more complicated than the already complicated story the genomes tell.

If this kind of thing interests you find me on Twitter. I am @troysk704.

I keep seeing people in my feed say that this or that is solved or cooked and there is a certain glee to it that I cannot shake because it is not a neutral description of progress and it celebrates obsolescence in a way that feels personal and I started looking at the people posting this and I realized the same people cheering are often the same people being priced out of the thing they are cheering for and I started wondering if this is really a class thing and not a technology thing because the people cheering have a cushion that absorbs the impact of what they are celebrating and they have a government and a safety net that half the world does not and science has always had a blind spot for the people at the bottom and to even be in this conversation you had to have had the means to get an education in the first place. The original promise of AI was that it would solve big problems like climate and disease but that has quietly shifted to a simpler goal which is replacing labour because even the most exploited people had one piece of leverage which was being needed and the people running these companies keep announcing that this leverage no longer counts for anything and the venture capitalists say in five years this and that will be solved and they say it with the confidence of people who have never had to worry about what being solved means for them. What happens to someone when you remove their economic utility is that the whole deal of modern life was that you get a chance if you work hard and even that deal was already thin and AI raises the ceiling of what one person can produce and it raises the barrier to entry at the same time because you need to pay for access or have the hardware or hope that the model weights stay open and the knowledge workers say with AI not by AI as if that distinction matters to someone signing a cheque because they will always choose the token cost over the salary cost. I watched a mathematician I respect say that if your goal is to achieve something lasting in his field that will not be possible for much longer and I think about how the people who never learned to code feel like the obstacle has finally been removed because they do not have to deal with programmers anymore and the AI will not argue back and will tell them their ideas are brilliant and I think about how all the training data came from everything everyone ever wrote or made with no choice in the matter unless you knew to put up a sign and how the world leaders were told that falling behind is extinction and once that argument is made there is no stopping the funding. The money goes in and the data goes in and the leverage goes and the quiet pleasure of making something with your own attention goes and you end up in a system you opted into by existing and it is hard to see a way out and it is not a good place to be regardless of who you are.

If any of this connected come find me on Twitter. I am @troysk704.

I have this pattern where I take a supplement for months or sometimes years and then I stop taking it just to see if I can tell whether it was actually doing anything and I have done this with every supplement I have ever taken because the idea of paying for something that is not working bothers me more than the risk of feeling bad for a couple of weeks. I have been taking Vitamin D and Omega 3s for a long time and I have cycled off them more times than I can count and every time the same thing happens where the first few days I feel fine and I tell myself that this time I have finally caught the placebo in the act and then around day seven things start feeling a little heavier and by day ten I am waking up with that familiar low hum of sadness that I have gotten to know very well over the years and by day thirteen every thought finds its way to something bleak and the filter through which I experience everything has shifted and the world looks grey and I cannot find the energy to care about the things that usually pull me out of it. And I always know somewhere in the back of my mind that it is the supplements and I always wait a couple extra days anyway because I want to be sure.

When I start taking them again it is never dramatic and there is no switch flipping but by evening I notice that I laughed at something and it was the first genuine laugh I had felt in days and the next morning I wake up and the world looks normal again and it is such a clear signal that I feel stupid for needing to prove it to myself again. The thing about Vitamin D is that it is not really a vitamin the way we think of vitamins because it functions as a steroid hormone that binds to receptors spread throughout the prefrontal cortex and the hippocampus and the limbic system and it directly regulates the expression of tryptophan hydroxylase 2 which is the rate-limiting enzyme that converts tryptophan into serotonin so without enough Vitamin D your brain literally cannot manufacture enough serotonin no matter how many SSRIs you throw at it because the enzyme that makes the neurotransmitter is not being turned on. And Omega 3s work through a completely different but complementary pathway where they integrate into neuronal membranes and change how signals propagate across synapses and they produce specialized molecules that actively resolve inflammation in the brain and they modulate the HPA axis so your cortisol response does not stay stuck in the on position and they shape the gut microbiome in ways that signal back up the vagus nerve to influence mood. So Vitamin D turns up the serotonin synthesis and Omega 3s quiet the inflammatory noise and stabilize the neural architecture and when you take them together they cover both sides of the equation in a way that neither can do alone.

I think the reason I keep running the same experiment over and over is that supplements work slowly and invisibly and you cannot feel the slope of a gradual improvement the way you can feel the shock of a sudden decline so every few months I forget what the decline feels like and I need to rediscover it. I am back on both supplements now and I will stay on them until the next time I convince myself that they are probably placebo and I stop again and feel terrible again and learn the same lesson again because that is just how I am wired and I have made peace with it. I think about how the sun is the original source of all this because Vitamin D is what your skin makes when light hits it and your brain uses it to make serotonin and without enough light the mind starts to grow things it should not and sunlight is the disinfectant of the mind in a way that is literal and not just poetic.

If any of this hit close to home find me on Twitter. I am @troysk704.

I have been thinking about GLP-1 drugs for over years now and the framing that finally clicked for me is that they do not just suppress appetite or reduce blood sugar but that they fix the energy cycle itself because obesity and metabolic disease are fundamentally disorders of failed energy regulation where the body’s signals for when to consume and when to burn and when to store have all gone haywire and what these drugs do is rebuild the signaling infrastructure one receptor at a time. Semaglutide activates the GLP-1 receptor which tells your brain you are full and your stomach to empty slowly so the energy-in side of the equation gets corrected but the energy-out side remains untouched because GLP-1 alone does not make you burn more calories. Tirzepatide added GIP agonism which improved how your body handles glucose and distributed the energy more efficiently but again the output side was mostly passive because you cannot exercise your way out of a broken metabolic set point and the body will just lower its energy expenditure to compensate. What makes retatrutide different is the third receptor.

Retatrutide is a triple agonist engineered from a GIP backbone to activate GLP-1 and GIP and glucagon receptors all in a single molecule and the key innovation is glucagon because glucagon is the hormone that tells your body to burn stored energy and generate heat and activate thermogenesis and it is the missing piece that converts these drugs from appetite suppressants into actual metabolic reprogrammers. The preclinical data showed that in mice retatrutide reduced food intake like the other drugs but also increased energy expenditure through glucagon receptor activation and that is why the weight loss numbers are not just incrementally better but categorically different. In the phase 2 trial published in the New England Journal of Medicine the 12 milligram dose produced an average weight loss of 24.2 percent at 48 weeks and the weight loss curves had not plateaued which meant people were still losing weight when the trial ended and they had to stop measuring. In the phase 3 TRIUMPH-1 trial announced just a week ago the numbers grew to 28.3 percent average weight loss at 80 weeks and in the extended cohort that continued to 104 weeks the average reached 30.3 percent which is the territory that until now required bariatric surgery and the permanent restructuring of your digestive tract.

Nearly half of the participants on the highest dose achieved 30 percent or greater weight loss and almost two thirds of them dropped below the BMI threshold for obesity entirely including 37.5 percent of those who started with class 3 obesity where your BMI is over 40 and the medical risks compound every year. The waist circumference reduction averaged 24.1 centimeters which is nearly ten inches and the systolic blood pressure dropped by 14 millimeters of mercury and the improvements in triglycerides and non-HDL cholesterol and C-reactive protein were all clinically meaningful and the liver fat reduction reached 82 percent which means retatrutide resolves metabolic dysfunction-associated steatotic liver disease in the vast majority of patients because the drug is not treating individual symptoms but fixing the underlying energy imbalance that causes all of them. And there is early preclinical evidence that retatrutide-induced weight loss reduces tumor engraftment and delays tumor onset in pancreatic and lung cancer models because metabolic health and cancer risk are connected in ways we are only beginning to understand.

The side effects are gastrointestinal like nausea and diarrhea and vomiting which are manageable with dose escalation and there is a strange one called dysesthesia which is abnormal skin sensations that resolves for most people during treatment and a dose-dependent increase in heart rate that peaks around 24 weeks and then declines and the discontinuation rate at the highest dose was 11.3 percent which is higher than placebo but reasonable for a drug that achieves weight loss rivaling surgery. The molecule itself is a single 39-amino-acid peptide conjugated to a fatty acid side chain that binds to albumin for a once-weekly injection and it was designed to be 8.9 times more potent at the GIP receptor than native GIP and slightly less potent at GLP-1 and glucagon receptors and this balance is the engineering achievement because too much glucagon agonism would raise blood sugar and too much GLP-1 agonism would cause intolerable nausea and the optimal ratio is something you can only find through iteration and human trials.

What excites me most is not the weight loss numbers but what they represent for healthspan because obesity is the single largest driver of preventable disease and it accelerates aging through inflammation and insulin resistance and cardiovascular strain and joint degeneration and the thing about retatrutide is that it does not just make people weigh less but it changes the metabolic trajectory of their entire body and the blood pressure improves and the liver heals and the inflammation drops and the knees stop hurting and people who could barely walk start moving again and this is what it means to extend healthspan not by adding years at the end of a life of decline but by preventing the decline itself from taking hold. The first generation drugs semaglutide and tirzepatide were already transformative but retatrutide with its glucagon-mediated energy expenditure is the first drug that addresses both sides of the energy balance equation and if the long-term safety data holds up it will be the closest thing we have to a genuine metabolic cure.

If this resonated find me on Twitter. I am @troysk704.

I spent the past few days reading the DeepSeek V4 paper and the design is genuinely beautiful because the team built a 1.6 trillion parameter model with a 1 million token context window that matches the top closed models while being incredibly resource constrained with a team forty times smaller than OpenAI and without access to the top NVIDIA chips and the ridiculous thing is they open sourced everything including a paper that reveals all the infrastructure details that the closed labs treat as top secret. The fundamental problem with large language models is that every time the model reads a new token it has to compare it against every token before it and at a million tokens that number of comparisons becomes astronomical and the GPU memory required to store the intermediate results which is called the KV cache explodes into gigabytes just to maintain context for a single conversation. The standard approach to this problem has been to throw more compute at it but DeepSeek did not have that luxury so instead they asked a different question which was what if the model did not have to look at everything in the first place.

The result is a hybrid attention architecture with three parallel pathways that work together. The first is called compressed sparse attention which groups tokens into small chunks and merges their information into a single denser representation so instead of remembering four individual tokens it stores a compact summary of all four and reduces the sequence length by a factor of four. The second is called heavily compressed attention which is far more aggressive and groups something like 128 tokens or an entire paragraph into a single representation and at that point the sequence becomes so short the model can afford to look at everything at once. The third is called sliding window attention which keeps the most recent tokens completely uncompressed with full exact fidelity. Between the compressed and heavily compressed layers there is something called the Lightning Indexer which acts like a built-in search engine that rapidly scores all the compressed blocks and selects only the small subset that most likely matters for the current context and everything else is skipped entirely. It is the difference between trying to remember everything perfectly and trying to remember the right things at the right time.

But solving the attention problem still leaves the problem of scale because when you stack dozens of layers at the trillion parameter scale the signals flowing through the network start to amplify like a microphone placed too close to a speaker and you get these catastrophic feedback loops that crash the training run. The industry standard way to handle this has been residual connections which act as bypass lanes but DeepSeek’s paper explicitly mentions that at over a trillion parameters even the newer hyperconnections start experiencing these spikes. So they introduced something called manifold constrained hyperconnections which forces the residuals to behave like a doubly stochastic matrix where every row sums to one and every column also sums to one which means the total signal is always conserved and can never amplify because the math literally forbids it. To apply this constraint at scale they used the Sinkhorn-Knopp algorithm which runs about twenty iterations of row and column normalizations before each layer and through aggressive low-level GPU optimization they shrank the overhead of this entire process to only 6.7 percent of runtime.

They also replaced the industry standard optimizer AdamW with a custom one called Muon which works in two phases where it first makes big rough adjustments to get the system close to convergence and then switches to tiny precise tweaks and this combination lets the model learn faster while staying stable. At the infrastructure level the model is so large it cannot live on one chip or even one rack so the layers have to be scattered across different racks and the bottleneck becomes communication rather than computation. DeepSeek solved this by breaking the data transfer into smaller sequential waves so that as soon as the data for the first wave arrives the GPUs start crunching while the data for the next wave is already traveling over the network and computation and communication are perfectly overlapped so the network latency essentially disappears. They wrote the code for this in TileLang using fused kernels and used a Z3 SMT solver to mathematically prove that the kernel code was correct because at this scale even a one in a billion error happens constantly.

The training itself used a curriculum where the model started with short sequences of four thousand tokens to learn grammar and syntax and then gradually stretched to sixteen thousand and sixty-four thousand and all the way up to the full million token window and they introduced something called anticipatory routing which uses slightly older snapshots of the model’s parameters to look past the noisy fluctuations and lock onto the underlying trend whenever it detects the early signs of a loss spike. The result is a model that achieves a perfect score of 120 out of 120 on the Putnam 2025 mathematics competition which is one of the most difficult undergraduate math competitions in the world and matches or beats models from Google and Anthropic and OpenAI across knowledge and reasoning and agentic benchmarks while running on roughly 27 percent of the compute that was required for their previous version and requiring only 10 percent of the KV cache memory. And the team put the model on Hugging Face for free and published the paper with all the infrastructure details and that is the part that is hardest to believe because a team that has every reason to hoard their advantages just gave everything away.

If this lands find me on Twitter. I am @troysk704.

I have been trying to understand when and how the shift happened from making things to making money from money because if you look at the post-war decades the American economy was built on manufacturing and a white male high school graduate could have a car and a house and three kids in college and a wife who did not need to work and that was the Pax Americana where prosperity was tied to producing things. But starting in the 1960s the cracks began to show because Germany and Japan rebuilt their manufacturing and started competing with American goods and by 1971 the United States ran its first trade deficit since the nineteenth century while inflation from Vietnam War spending was eating away at the dollar. In August 1971 Nixon closed the gold window which meant foreign central banks could no longer exchange their dollars for gold at thirty-five dollars an ounce and this ended the Bretton Woods system and turned the dollar into a pure fiat currency backed by the authority of the state rather than by a commodity.

The dollar needed something to anchor its value after gold was gone and the answer came in 1974 when the United States struck a deal with Saudi Arabia where oil would continue to be priced in dollars and Saudi petrodollars would be recycled back into US Treasury bonds and this created the petrodollar system which meant anyone in the world who needed oil also needed dollars first. Around the same time Nixon made his historic visit to China in 1972 but the primary motivation was not economic because it was about playing the China card against the Soviet Union and getting help to end the Vietnam War and China at that point was a poor country with little to trade and the real economic integration happened later under Deng Xiaoping’s reforms starting in 1978 when American companies began moving manufacturing to China for the labor arbitrage.

In 1979 Paul Volcker became Federal Reserve chairman and raised interest rates to twenty percent to kill the inflation that had been brewing since the 1960s and it worked for prices but it also crushed wage growth and manufacturing and triggered a debt crisis across the developing world and this is the moment when the balance tipped. After Volcker the financial sector began capturing a much larger share of profits and by the 2000s finance was taking forty percent of domestic corporate profits compared to ten percent in the postwar era while manufacturing employment peaked in 1979 and has declined ever since. Wage growth for the bottom ninety percent of earners decoupled from productivity growth around 1973 and has never re-coupled because productivity kept rising but the gains went to capital not labor and the top one percent saw their share of national income double from around ten percent in the 1970s to over twenty percent by the 2010s.

The mechanism behind this was straightforward which is that after the dollar was freed from gold there was no constraint on monetary expansion and after capital controls were removed money could move across borders freely while labor could not and after the financial sector was deregulated the people who moved money around could capture value without producing anything tangible. Michael Hudson calls this the shift from industrial capitalism to financial capitalism and what it means is that the person who builds a thing or grows food or writes code gets a shrinking share while the person who structures a deal or trades a derivative or collects rent gets an expanding share and this is how value creation changed from labour to money.

The practical consequence is that working hard will never make you rich in today’s world because the returns to labor have been deliberately compressed while the returns to capital have been inflated by decades of easy money and low interest rates and asset price appreciation. The people getting rich are not the ones who build things or grow food or write code because they are the ones who own assets and structure deals and sit on boards and collect rent and the system is designed to reward the movement of money over the creation of value because the rules were rewritten after 1971 to favor capital over labor. This is why crony capitalists keep growing because they understand that the game is not about producing more than the next person but about positioning yourself closer to the spigot of cheap money and once you see that the forty-year transfer of wealth from labor to capital was not a bug but a feature of the post-Bretton Woods system then the only honest response is to stop believing that hard work alone will save you.

If this resonates do give me a shout. I am @troysk704.