Luba Vangelova

They’re not even giving kids a chance to develop their brains or their social world. Hooking them as customers as early as possible.

Dermatology is wrong about the sun. And it's killing people. I'm a dermatologist. 226 publications. I should know. Avoiding the sun increases the risk of dying as much as being a smoker. We can fix it. For decades, dermatology's message has been simple: avoid the sun. Wear sunscreen. Seek shade. UV causes skin cancer. End of discussion. That message is incomplete and outdated. People are dying because of it. Lots of people. The evidence has gotten strong enough that the field needs to update it.🧵

American foods are cleverly changing their names because they no longer qualify as that legal food - Pringles used to be potato chips, now they’re labeled as “potato crisps” per FDA rules - McDonald’s Shakes are now called “shakes” (not legally milkshakes in some states) - Klondike Bar is no longer a chocolate shell, it’s a Chocolatey shell (not real chocolate) - Dairy Queen: All items are “treats” (no “ice cream” on the menu) - Oreo changed the spelling to “creme” (alternative spelling, not real cream) - Tyson changed spelling to “Wyngz” (It’s a processed chicken labeling trick) - Costco Blueberry Bagels labels as Imitation blueberry bagels (no actual blueberries) - Pearl Milling Company Syrup (formerly Aunt Jemima) Now ‘Original syrup’ not maple syrup. The first ingredients is corn syrup Our food is a science experiment Make America Healthy Again

Composting does not require a bin, a tumbler, a thermometer, or a schedule. The oldest soil improvement method in agriculture is trench composting — digging a hole, filling it with organic waste, covering it with soil, and planting on top. The decomposition happens underground where you never see it, never smell it, never turn it, and never think about it again. The soil organisms that do the work were already there before you started. 🪱 Every piece of composting equipment ever sold solves a problem that a shovel already solved. A tumbler aerates the material by turning it — soil organisms aerate buried material by moving through it. A bin contains the pile so it does not spread — the hole contains it because the walls are dirt. A thermometer monitors decomposition temperature — buried organic matter decomposes at ambient soil temperature without any monitoring because the organisms that work at that temperature are already the dominant population in the ground. The equipment is not wrong. It is just unnecessary for people who have a garden bed and a shovel. The build takes about 10 minutes: What you need: - A shovel or garden fork - Kitchen scraps, garden waste, shredded leaves, or any compostable organic matter - The garden bed you are already growing in How to do it: DIG — excavate a trench or hole twelve to eighteen inches deep in the garden bed. A trench running between two rows works well. A single hole between two plants works equally well. The shape does not matter. The depth does — twelve inches minimum keeps the material below the surface where decomposition is aerobic, odor-free, and inaccessible to surface-feeding animals. FILL — add kitchen scraps, coffee grounds, crushed eggshells, shredded leaves, garden trimmings, and any non-meat non-dairy organic material to the hole. Fill to within four to six inches of the surface. Mix a handful of existing garden soil into the scraps to inoculate the material with the decomposer organisms already present in the bed — this accelerates the process but is not strictly necessary because the organisms will migrate from the surrounding soil within days. COVER — backfill with the excavated soil, mounding slightly above grade because the material will settle as it decomposes. The soil cap seals the scraps underground where they are invisible, odor-free, and unreachable by raccoons, rats, and dogs. PLANT — wait two to four weeks for the initial decomposition to break down the freshest material, then plant directly on top or alongside. In an active garden, dig the trench between rows or at the end of a bed where nothing is currently growing and plant into that section the following month. In fall, bury the season's accumulated kitchen scraps across the entire bed and plant the whole area in spring. 🌿 What happens underground: Soil bacteria, fungi, earthworms, millipedes, and microarthropods colonize the buried material within days. The decomposition proceeds from the edges inward as the organisms work through the scraps. Nutrients are released directly into the surrounding soil in plant-available form — nitrogen, phosphorus, potassium, calcium, and a full spectrum of micronutrients. The organic matter that remains after decomposition improves soil structure — increasing water retention in sandy soil and improving drainage in clay. The timeline depends on temperature and material. In warm summer soil, soft kitchen scraps (fruit peels, coffee grounds, lettuce leaves) decompose to the point of being unrecognizable within four to six weeks. Tougher material (corn cobs, avocado pits, woody stems) takes three to six months. Shredding or chopping the material before burying accelerates everything because smaller pieces expose more surface area to the organisms. Three variations that fit different garden layouts: THE TRENCH — dig a trench twelve inches deep and twelve inches wide between two planting rows.

Your friend calls at 11:23 on a Tuesday. You wanted to ask him if his father has died yet. You wanted to ask him if he was okay, but you let it ring. On the last unedited medium, and the calls we don't return.

Let me tell you about a law most Americans have never heard of. Eighteen months from now, every new car sold in the United States will come with technology that watches you drive. Infrared cameras tracking your eyes. Sensors measuring your pupil dilation. Software analyzing your head position. Software analyzing your behavior at the wheel. If the artificial intelligence in your car decides you are impaired, your car can refuse to start. Or limit your speed to 25 miles per hour. Or shut off entirely while you are driving. This is not a proposal. This is federal law. It applies to model year 2027. Here is the case for the law. 🧵

If you don't understand this, you will not understand why LLM-based agents are irreparably failing for a general-purpose problem solving. An agent (by the way it was the topic of my PhD 20 years ago) to be useful, must be rational. Being rational means to always prefer an outcome that results in the maximal expected utility to its master/user. Let’s say an agent has two actions they can execute in an environment: a_1 and a_2. If the agent can predict that a_1 gives its user an expected utility of 10, and a_2 gives an expected utility of -100, then a rational agent must choose a_1 even if choosing a_2 seems like a better option when explained in words. The numbers 10 and -100 can be obtained by summing the products of all possible outcomes for each action and their likelihoods. Now here is the problem with LLM-based agents. The LLM is not optimizing expected utility in the environment. It is optimizing the next token, conditioned on a prompt, a context window, and a training distribution full of examples of what helpful answers are supposed to look like. Those are not the same objective. So when we wrap an LLM in a loop and call it an “agent,” we have not created a rational decision-maker. We have created a text generator that can imitate the surface form of deliberation. It may say things like: “I should compare the expected outcomes.” “The best action is probably a_1.” “I will now execute the optimal plan.” But the internal mechanism is not selecting actions by maximizing the user’s expected utility. It is generating a continuation that is statistically appropriate given the prompt and prior context. This distinction matters enormously. For narrow tasks, the imitation can be good enough. If the environment is constrained, the actions are simple, and the success criteria are close to patterns seen in training, the system can appear agentic. But for general-purpose problem solving, the gap becomes fatal. A rational agent needs stable preferences, calibrated beliefs, causal models of the world, the ability to evaluate consequences, and the discipline to choose the action with maximal expected utility even when that action is boring, non-linguistic, or unlike the examples in its training data. An LLM-based agent has none of that by default. It has fluency. It has pattern completion. It has a remarkable ability to compress and recombine human text. But fluency is not rationality, and a plausible plan is not an expected-utility calculation. This is why these systems so often fail in strange, brittle, and irreparable ways when given open-ended responsibility. They are not failing because the prompts are insufficiently clever. They are failing because we are asking a simulator of rational agency to be a rational agent.