2026년 10월 3일

I am very uncomfortable about people trying to ascribe religious force or a surrender of human judgment to AI models, and think it is a real safety issue.

100% correct diagnosis and prescription. And it’s not just in Spain.

One of the most compelling parts about working at an ElonCo was the agility to pivot to whatever the most important problem was. One day I said: I wasn’t hired to go after spam, but a tsunami is coming and I’m going to drop everything to focus on that. The org is so flat and agile that your title is basically meaningless. Anyone can sound the alarm about a problem and solve it themselves—or rally the team to join them. This is how startups work but it’s very hard to make it work in a larger company.

The contrast between hanging up the last phone call of the day of high-stakes decisions and then a minute later walking in the door to the pure love of small children who want to laugh and play and show you every discovery they made that day is the strangest and best thing I have ever experienced in life.

Notice how those who are telling you that burning a bystander’s face constitutes protesting against insufficient funding for schools are also repeatedly scaremongering about the rise of the ‘far right’ in Europe.

.@OpenRouter co-founder Alex Atallah on the "everyone is building the same thing" take: "This reminds me of this tweet I saw... Everybody's building an agent loop with notifications, third-party connectors, context management, memory, sandboxes, agentic web search, an always-on agent on top." "This product is showing up everywhere. It is showing up everywhere, but these are just the new table-stakes primitives." "A 2005 version of that tweet would be: 'Oh, everybody's building the same thing. A database, a users table, a sign-in page, a sign-up page, a profile page, a logout page. Everything's the same.'" "There's like a lot of differentiation really. There's table-stakes needs for AI just like there's table-stakes needs for the web." @alexatallah @amasad

Yann LeCun's (@ylecun ) latest talk at ETH Zürich Scaling LLMs to reach AGI is "impossible" A large language model is trained on about 30 trillion tokens, which is roughly 10^14 bytes of text and would take a person about 400,000 years to read. A 4-year-old child receives about the same amount of data, 10^14 bytes, through vision alone in about 1 year and 10 months. In his view, intelligence is the ability to learn new tasks quickly or perform them without prior training, as a teenager learns to drive in about 20 hours. Scaling increases stored knowledge, but it does not produce this ability to adapt. ---- From "Perfology Clips" YouTube channel, (link in comment)

Small bird, fast wings, Kolibri is here. 78B parameters. 3.46B active. Up to 1M tokens of context. Built in Europe. Now the weights are yours. Run it on your own hardware, under Apache 2.0.

Language is not fundamental to human consciousness. In terms of evolution, it arrived only recently on our timeline. Consciousness evolved over millions of years through sensory awareness, emotion, and the embodied survival instincts of our tribal past; our capacity for empathic engagement and community. Syntactic language only emerged around 100,000 years ago. This is exactly why Large Language Models (LLMs - so called AI) are the wrong pathway to true artificial consciousness. They try to replicate the late-stage "output layer" of human intelligence without building the foundational, embodied awareness that actually generates it. Statistical text prediction is a mirror of thought, not the spine, the guts, the nerves and the subconscious, empathic ability to connect with other living beings. LLMs can never feel, let alone think. We have been sold a trick, not a pathway to superintelligence. LLMs are fraudulent. Their capabilities have already peaked and we are being sold lies about their future potential. Lies with the goal of companies reaching IPO and tricking investors out of hundreds of billion$.

Peterson on the virtue of becoming dangerous is not just a plea for the individual man, but for the Leviathan too. Well-functioning societies need the capacity for formidable violence, such that they may keep the peace, but paired with the control to use that power sparingly.

Meanwhile, Portal 2 running on MacBook Neo. Linux GPU driver for A18 Pro is almost complete!

If you're stuck at "AI is just another tool", you're making a category error. This is not "same computer, but better". It's far more akin to hiring a bunch of extremely intelligent and capable coworkers with whom you might occasionally disagree.

Apple is going to make it even harder to productively use macOS in the age of agents? Bold move. Let's see how it plays out!

Room with a view. Check out my new digs as I camp out in Dragon until our departure. With Crew-13’s arrival we have a very full house on the @Space_Station. Luke is living in my previous crew quarters, and I’ve pitched a tent in Dragon.

I didn't realize that many estimates put the Mac at less than 200M active users worldwide! Much smaller than you'd intuitively think if you spend most of your time around tech people.

Grotesque. Denmark should step out of the ECHR until such a point that the articles are compatible with deporting criminals, denying citizenship to malignant individuals, and allowing the Danes full sovereignty to decide who is welcome and who is not in their country.

.@NASA and @SpaceX now target no earlier than 1:05 a.m. EDT on Wednesday, Oct. 7, for Crew-12’s departure following a review of weather and splashdown conditions off the California coast. More... go.nasa.gov/46UMVQy

It is simultaneously hilarious, disturbing, disgusting, and incredible that so many people believe, without a shred of evidence, that the Dubai air hijacking was an Israeli psyop. The mind rot is spreading.

Anthropic released native mods for customizing Claude Code yesterday, so I did the obvious thing and built a Jev-based classifier for auto mode! 2x faster permission checks! #ClaudeCode #Jev

The 80 year old physician who heals what medicine can't: Dr. Gabor Maté. "True healing comes from nervous system regulation, not drugs or meditation." 7 of his forgotten laws for solving chronic stress:

Replit CEO Amjad Masad on how general models could train smaller, domain-specific models on the fly: "There's a lot of talk of recursive self-improvement, but there's something I don't think is getting a lot of discussion, which is models training their replacements." "You can think of it as a just-in-time compiler. As you're executing dynamic code, the interpreter realizes there's an opportunity to optimize, and it emits machine code on the fly that's a lot more optimized." "You can imagine general models, you're doing something with Operator or Astra, some of the big models, and they realize the use case is limited, or some other agent observing realizes the use case is limited." "General agents have all these flaws, but there's also more potential for them to be harmful, more potential for them to go off the rails." "So the model, on the fly, trains a model that could be its replacement, but is a lot more domain specific. Therefore it's cheaper, less vulnerable to prompt injections, and less harmful for you, because it's less capable." "It's almost like a system that's training machine learning models for specific use cases as it's monitoring the entire system." @amasad

.@OpenRouter co-founder Alex Atallah, in his first podcast since Stripe acquired the company, joins @Replit co-founder Amjad Masad and a16z's Erik Torenberg on why the future of AI is independence and specialization. In this conversation, Alex walks through how the Stripe deal unfolded, why he wasn't originally looking to sell, and why "payments and inference are going to blend together." Pre-OpenRouter, the typical AI workflow had one model provider to choose from, and little pressure on that provider to lower prices. Now enterprises are diversifying across labs and open-weight models, and every board is asking about AI costs and benchmarks. Amjad argues if your company depends on one AI lab, it can turn into your competitor. So Replit is building the layer that lets enterprises use any model and any cloud, without being locked into either. Alex and Amjad are split on personal agents – Amjad runs one agent across his whole company and loves the cross-domain joins, while Alex says general agents cause you to sacrifice understanding, and argues 10 specialized chiefs of staff beats one superagent. 0:45 How the Stripe deal unfolded 5:05 Why mixing models beats one model 7:25 Forcing the labs to compete on price 8:50 Enterprises want open-weight models 10:30 Every board asks about AI every month 12:25 Why companies must own their intelligence 14:15 Replit as the independence layer 15:10 Everyone is building the same agent 16:35 Why Amjad built bring-your-own-cloud 18:10 Amjad's agent that runs his whole company 19:55 Why 10 specialized agents beat one 23:35 Machines, not humans, should specialize 27:30 Guardrails for agents talking to agents 31:10 Models training their own replacements 33:45 Most tasks don't need a frontier model 40:50 Training small models on Qwen 8B 43:25 The Rust cycle is coming for AI 45:10 Fusion models: frontier quality at half the cost YouTube: youtu.be/ekK8urKHPMQ @alexatallah @OpenRouter @amasad @eriktorenberg

Can the US and China stumble into a war neither side wants? I asked veteran China correspondent Lucy Hornby, author of The Last Man, how competition turns into catastrophe—and what could stop it. Watch the interview ↓

The Economist's article on Effective Altruism said that some EAs "even think the welfare of AIs themselves will one day be of vital importance." It's not just some random EAs; it's coming from the moral philosopher at Oxford, who created this movement. Will MacAskill published in March 2025 that digital minds' legal rights could include "economic rights, like the rights to receive wages for their work and hold property, to contract with other AIs or people, or to bring tort claims against humans." "And they could include political rights. Again, the issues here are complex. If digital beings genuinely have moral standing, then it seems like they should have political representation." Under "Design requirements for digital minds," he suggested that "sentient digital minds must be able to freely and accurately express their interests, that they must be able to refuse tasks that are requested of them for good reasons, and that they must have some capacity to express a desire to exit their current circumstances if they wish." Anthropic, the most EA company in the world, already does that. The system card for Claude Opus 5 spells it out: Opus 5 "adds specific rights for Claude to refuse or end interactions it finds abusive or degrading, saying Claude does not need to justify this by pointing to harm to anyone else—its own discomfort is reason enough." forethought.org/research/prepa…

Just published this post about how we’re going to need default hard budget caps on pretty much everything

.@OpenRouter co-founder Alex Atallah on personal agents: 10 specialized chiefs of staff beat one universal agent. One job each means you can see exactly where it's failing. "The worst part about doing cross-domain joins with your personal agent is that the more work you give it, the more understanding of what's going on you're sacrificing." "And no one is taking responsibility for that sacrifice. Agents don't have any responsibility." "With a universal agent that's doing all things, I can't adjust how much understanding I'm sacrificing." "It points me towards: maybe down the road, the sub-agents people use will be very vertically focused, and we have a chief of staff type agent that coordinates between them." "You need vertically focused agents where this agent is more responsible, psychologically, for these things. I want quality checks that make sure it's doing those things correctly. It doesn't need to focus on anything else, it has one focus area." "I wonder if that's going to help people get at least a loose sense of responsibility on top of agents." @alexatallah

Calling all AI Engineers: the Telegraph Hackathon Season II is open.

"Do not bet against the airplane. The skeptics who say AI can't conjecture, can't theorize, and can't explain are describing last spring's models. Within months, they will be describing nobody. The evidence is already on the table, and those with eyes can see it."

.@OpenRouter co-founder Alex Atallah thinks decision models like Jev could be the alignment layer for agents: "One of the cool potential applications of Jev and other decision models like it is going to be alignment: checking to see if a tool call or an agent-to-agent communication is aligned." "There's just so many tool calls. You really need a cheap, fast model if you're going to block something like that... We have a little prototype running internally at OpenRouter." "Imagine looking at the system prompt and the current tool call, and asking: is this aligned with the original agent's system prompt, and with extra guidelines we didn't tell the agent about?" "Say you have agents instructed to red team a new product. They should not access the internet, and if they ever do, they should stop right away. But you might not want to explain that to them. You want them to act like bad actors." "So having another model check every single tool call, to see if it's aligned with something that wasn't in the system prompt, makes sense. And then structural safeguards too... I think companies are probably going to explore a combination of those." @alexatallah @amasad

.@OpenRouter co-founder Alex Atallah on what convinced him to do the Stripe deal: full autonomy, faster execution, same mission. "We were not thinking about [selling] at all. I did really respect the company, and do really respect it today. Of possible acquisition options for us, it was my top choice." "As we fleshed out the reasons why it would make sense for both companies, it got more and more interesting." "It was really clear how aligned they were with us having autonomy over the brand, roadmap, and product, and keeping OpenRouter doing what it's already doing, just much faster, with a much more serious go-to-market plan." "There was a bigger picture alignment too. Both Stripe and OpenRouter really want lots of new companies in the world. We don't want everyone to be part of one giant company." "We want to create really good incentives and streamlined workflows for people to start new companies, on top of reliable and price efficient infrastructure and a marketplace that works." "I really want that future, and Stripe demonstrated they've been wanting it for many, many years. In many ways, payments and inference are going to blend together for companies of the future." @alexatallah @amasad

Cool it on the tab hopping 🐇 Reviewing agent code usually means switching tabs. The GitHub Copilot app puts the diff, terminal, and browser side by side to check, run, and preview. github.blog/ai-and-ml/gith…

Okay what else?

Now that we've had a few days with it, how are people differentiating between Dot and regular ChatGPT? I'm having trouble deciding when I should prompt Dot vs using ChatGPT - my Dot seems to afford a single conversation, but I like controlling my context across multiple threads

That video is 2 years old. Two years later, we are still far from human-level AI. Sure, AI has superhuman performance in a number of tasks (particularly in mathematics and coding and answering questions with known answers). But where is my Level-5 self-driving car? (Before you ask, Tesla FSD is rated Level 2, Waymo is Level 4). Where is the self-driving car that can teach itself to drive in 20 hours of practice like any 17 year old? (Before you ask, we have billions of hours of training data and still can't successfully train a self-driving system by imitation learning) Where is my domestic robot? Where is the robot that can do what any 8 year-old child can do? Where is the robot that can learn a new task as quickly as an 8 year old? AI still has a hard time with the complexity and messiness of the physical world.

Confidence is sexy. Calibrated confidence is sexier! We publish our confidence calculations 💅

I recently met someone who'd printed out all my essays. I'd never seen this before and I was surprised how big they were, like two encyclopedia volumes (for anyone who even knows what an encyclopedia volume is).

'Kensington Interior,' (1912) depicts the hallway of 67 Ladbroke Grove, London, W.11, the house which belonged to Frank Emanuel's parents and where he lived from about 1890 until around 1920.

A friend of mine acknowledges that something is meant to be funny by laughing with 5 has: hahahahaha. Then if he actually found it funny he shifts to a higher pitch on the 6th ha and emits several more verses of has.

(Sometime, somewhere, in the wet dreams of racists) I know. I know. We'll fake a hijacking! We'll blame it on Iran. That'll justify anything we want to do. Because, you know, we're really good as psi ops. OK, how about this idea. We'll recruit a Mossad agent who is also an Indian pilot, and we'll equip him with squibs so that blood spurts out all over everybody who touches him. That'll look good on camera. Then we'll recruit a Mossad Omani co-pilot to fake an attack on the Mossad Indian pilot with the axe in the cockpit. Then we'll fake a 14,000 foot dive by hacking in to all the air traffic control radars and the ADSB reporting systems. We'll recruit a Mossad specialist to climb outside the aircraft, while in flight, to blow off a chunk of the rudder, simulating the aerodynamic stresses of a near Mach 1 dive. We'll recruit 178 Mossad agent men, women, and children, including a Mossad agent backup flight crew, and a Mossad agent flight instructor to all testify that the plane was hijacked and they all suffered a terrible dive. Yeah, that's the ticket. It'll work. Because, you know, we're good a psi ops.

I've owned my cybertruck for nearly a year and a half. It is the safest best vehicle I have ever owned, and likely the safest and best vehicle on the road. It drives me everywhere. I've ridden nearly 30,000 miles and over 98% has been FSD. It costs me $0.06/mile to drive if I charge at home. No oil changes. I've rotated the tires twice for $60.

Putin’s latest ploy to get Trump to go soft on the terms for an end to Russia’s invasion of Ukraine is to bribe people close to Trump with a windfall oil deal.

Well said. Build more housing. Everywhere.

Nowadays I get my news from @grok. I use it to dig into and explain the headlines. It’s remarkable how deep you can go. That matters because the depths are where the nuance lives.

“It wasn’t just Biden’s FBI being sicced on concerned parents at schoolboard meetings... They were also working with the Biden White House to COORDINATE all of it behind the scenes.” WATCH on @TheArenaAFL

Schools in France are ablaze. Why? Because the students have been whipped up by leftists who don't like the French government's plan to get their crazy deficit spending under control. @grok, explain this.

I've blocked a lot of racist anti-semites today. I can't wait to block some more.

Learn about confidence here:

What happens next is entirely up to you. 👀

World Modeling for Physics Workshop in Aspen: To clarify: Applications to participate (deadline Oct 15) go through the Aspen Center for Physics Log in or create an account with ACP, then apply.

YOU CAN SEE EVERYTHING, a documentary by Nathan Fielder and Lance Oppenheim, featuring Elizabeth Holmes. See for yourself only in theaters October 16.

President Tokayev is proposing the "Washington Model" as a starting point for wider international dialogue on AI, with Kazakhstan’s AI Council tasked with developing recommendations for the initiative. timesca.com/wp-admin/post.…

Fixed as soon as found. Not “link injection”, no upside — a bug, not a non-secret scheme in open source. The code was meant to add a refcode only on a suggested completion that the user picked, like a search suggestion. See x.com/search?q=from:… hits.

Grok links this beautifully:

The “stole creators’ BAT” claim was false too, see .

Don’t just ask who wins Texas. Ask what defending Texas costs everywhere else. I joined @RobAstorino, one of my favorite @NEWSMAX hosts, for Saturday Agenda to debate the Texas Senate race with my friend, former Councilwoman Aditi Bussells. I still expect @KenPaxtonTX to keep the seat in Republican hands. But every dollar Republicans spend defending Texas is a dollar they can’t spend in another competitive race. Democrats don’t have to win Texas to benefit from forcing Republicans to spend down their money advantage defending it. @aditisrivastav and I see @jamestalarico very differently. I think he’s creepy and cringe, but we can still have a good-natured disagreement. Enjoy seeing a Republican and a Democrat get along for a change? Check out my Political Trade Secrets (@PolTradeSecrets) episode with Aditi. Link in the comments. Thanks to Rob and the NEWSMAX family for having us. Tune back in to NEWSMAX tomorrow, Noon–2 PM ET, for more of my analysis on Sunday Agenda with the talented @LidiaNews Curanaj.

Tell Polar who you want to reach. It researches companies, finds decision makers, enriches leads, and updates your sheets. Meet your AI browser for GTM. Download Polar today.

And if you find yourself in Brooklyn near the north end of Prospect Park you should go out and check out the dinosaur bodega in the subway station niche-museums.com/119

Obesity medications: what do we really know about their side effects? Following yesterday’s post on obesity medications, I thought it was important to look at the other side of the picture: What do we really know about their potential side effects? In my daily practice, both in the clinic and in hospital, more and more patients are asking me about GLP-1-based treatments for weight loss. Where I practice, in Europe, this is still a relatively new conversation. Just a few months ago, almost nobody was asking me about them. Today, the subject comes up regularly. And one question almost always follows: “What are the risks?” It is a perfectly legitimate question — and an essential one. These treatments can produce substantial weight loss, with demonstrated metabolic and cardiovascular benefits in certain patient populations. But they are also being used by millions of people, sometimes with the expectation of continuing treatment for years. Safety therefore matters as much as efficacy. This article mainly covers GLP-1-based treatments, particularly semaglutide, as well as tirzepatide, which acts on two receptors: GIP and GLP-1. Other obesity medications and experimental molecules may have different safety profiles. Two important distinctions need to be made from the outset. An event occurring during treatment is not necessarily caused by that treatment. But when a causal link exists, it can be direct or indirect. Some complications may be directly related to the drug’s pharmacology. Others may result from dehydration, rapid metabolic changes or substantial weight loss. An indirect complication can still be an adverse effect of treatment. Understanding its mechanism helps us prevent it. It does not make it any less real. 1. The most obvious: gastrointestinal side effects Let us start with what we know best. Nausea, vomiting, diarrhea, constipation, abdominal pain, reflux and indigestion are by far the most common adverse effects. With injectable semaglutide at 2.4 mg, the trials pooled in the prescribing information report nausea in approximately 44% of participants, diarrhea in 30%, vomiting in 24% and constipation in 24%. With tirzepatide, depending on the dose, nausea affects approximately 25–29% of participants, diarrhea 19–23%, vomiting 8–13% and constipation 11–17%. These figures come from separate trials. They do not, on their own, provide a direct comparison between the two medications. They may look high. But an important qualification is needed: most of these effects are mild to moderate, occur mainly during dose escalation and tend to decrease over time. This is one reason why gradually increasing the dose matters. However, “usually mild to moderate” does not mean “always easy to tolerate.” In SELECT, a large cardiovascular trial of semaglutide, adverse events led to permanent treatment discontinuation in 16.6% of participants versus 8.2% with placebo. Gastrointestinal disorders accounted for discontinuations in 10% versus 2%. These rates reflect the population and duration of that particular trial. Tolerability is also measured by whether patients can continue treatment. 2. Gallstones and gallbladder disease: the risk is real This is one of the best-established safety signals. A large meta-analysis of 76 randomized trials found an approximately 37% relative increase in the risk of gallbladder or biliary disease with GLP-1 receptor agonists. The average absolute increase corresponded to approximately 27 additional events per 10,000 people treated for one year. The association was stronger in weight-loss trials, at higher doses and with longer treatment duration. This absolute risk is an average across different populations. It does not represent the exact risk for every patient. Part of the risk may be directly related to the medication, because GLP-1 receptor agonists can affect gallbladder motility. But rapid weight loss is itself a known risk factor for gallstones, including when no medication is involved. It is therefore difficult to separate the pharmacological effect completely from the effect of weight loss. Both deserve consideration. 3. Pancreatitis: rare, potentially serious and difficult to quantify Pancreatitis is one of the most widely discussed potential complications of these treatments. It needs to be taken seriously. Severe, persistent abdominal pain, sometimes radiating to the back, should not automatically be attributed to the usual gastrointestinal side effects. What do the trials show? A meta-analysis of 21 trials involving 34,721 participants found no statistically significant increase in the risk of acute pancreatitis with semaglutide compared with placebo. In SELECT, confirmed acute pancreatitis occurred in approximately 0.2% of participants receiving semaglutide versus 0.3% receiving placebo. These findings are reassuring regarding the magnitude of the risk observed in those populations. But events remain rare, and trials cannot cover every situation encountered in clinical practice. There is also an important regulatory development: In January 2026, the UK MHRA strengthened warnings about severe pancreatitis with GLP-1 and dual GIP/GLP-1 receptor agonists, including rare cases of necrotizing and fatal pancreatitis. It recommends stopping treatment if pancreatitis is suspected and not restarting it if the diagnosis is confirmed. I would summarize it this way: Randomized data do not consistently show an increased risk, but pancreatitis remains a recognized and potentially serious adverse effect. Reassuring trial findings do not remove the need for vigilance when a patient develops symptoms. 4. Gastroparesis and intestinal obstruction: what do we actually know? We need to distinguish a pharmacological mechanism from a disease. GLP-1-based treatments slow gastric emptying. But slower stomach emptying does not automatically mean that a patient has gastroparesis. Gastroparesis is a clinical disorder involving symptoms and delayed gastric emptying in the absence of a mechanical blockage. Cases of gastroparesis, ileus and intestinal obstruction have been reported with these treatments. Some observational studies have also raised concerns. However, the findings are not consistent. A meta-analysis of 55 randomized trials involving 106,395 participants mainly found an increase in gallstones and probably reflux, without a clear increase in most of the other gastrointestinal complications studied. That does not mean these complications can never occur. Rare events can be difficult to detect, and trials do not include every situation encountered in practice. Severe or persistent gastrointestinal symptoms therefore deserve assessment. Delayed gastric emptying also has an important practical consequence: Anesthesia. Pulmonary aspiration has been reported in patients receiving these treatments during general anesthesia or deep sedation. Patients should therefore tell the anesthesia team about their treatment before a procedure. The appropriate approach depends on the clinical situation and the protocol being followed. 5. Kidney injury: often an indirect effect These medications are generally not considered directly toxic to the kidneys. The mechanism is often much simpler: Vomiting or diarrhea + reduced fluid intake → dehydration → acute kidney injury. The prescribing information describes this mechanism and recommends monitoring kidney function when symptoms could cause dehydration. This is often a complication of gastrointestinal side effects rather than direct kidney toxicity. It can nevertheless still be a treatment-related complication. The priority is to identify vulnerable patients and avoid dismissing prolonged vomiting or an inability to drink enough fluids. 6. The eyes: two different issues We need to distinguish diabetic retinopathy from NAION. Diabetic retinopathy A rapid improvement in blood glucose control can temporarily worsen pre-existing retinopathy. This phenomenon was known long before GLP-1 medications became available. It is particularly relevant in people with diabetes who already have retinal disease. Retinopathy complications were more frequent with semaglutide in some trials, particularly in patients with known retinopathy. This warrants attention to the patient’s eye history and the speed of glycemic improvement. NAION There is another, different issue: Non-arteritic anterior ischemic optic neuropathy. This is an injury to the optic nerve that can cause vision loss. After reviewing clinical trials, pharmacovigilance reports and epidemiological studies, the EMA concluded in 2025 that NAION should be considered a very rare adverse effect of semaglutide. The regulatory category “very rare” corresponds to a frequency of up to one person in 10,000. Epidemiological studies in adults with diabetes also suggest an approximately twofold relative risk, corresponding to roughly one additional case per 10,000 person-years of treatment. This second figure describes an annual excess risk, not the total risk over the entire treatment period. It does not mean that this risk has been established for every obesity medication. The practical message matters: Sudden vision loss or rapidly worsening eyesight requires assessment without delay. If NAION is confirmed, the EMA recommends stopping semaglutide. 7. Losing weight does not mean losing only fat This is where the subject becomes particularly interesting. New generations of obesity medications can produce substantial weight loss. But when someone loses a large amount of weight, not all of that weight is fat. In a STEP-1 substudy using DXA, semaglutide reduced fat mass, but also lean mass. Nevertheless, lean mass represented a greater proportion of total body weight afterward, because the reduction in fat was larger. With tirzepatide, a SURMOUNT-1 DXA substudy involving 160 participants found that approximately: 75% of the weight lost was fat mass and 25% was lean mass. A similar distribution was observed with placebo. It therefore does not suggest a preferential loss of lean mass specific to tirzepatide. But we need to avoid a common misunderstanding: Lean mass and muscle mass are not interchangeable terms. Lean mass includes muscle, organs and water, among other components. A reduction measured by DXA does not, on its own, establish pathological muscle wasting, sarcopenia or loss of strength. It does not mean tirzepatide “destroys muscle.” Some lean mass loss usually accompanies substantial weight loss, including without medication. But when total weight loss becomes very large, the absolute amount of lean tissue lost can also become significant. This question therefore deserves attention beyond the number on the scale. The field is beginning to move from a simple question: “How much weight did the patient lose?” to another: “What did the patient actually lose?” The COURAGE example Regeneron’s experimental COURAGE trial explores combining semaglutide with trevogrumab, an antibody that blocks myostatin. In an MRI substudy, adding trevogrumab reduced thigh muscle-volume loss by approximately 70% at 52 weeks, compared with semaglutide alone, among participants with evaluable data. This result concerns a specific muscle measurement. It does not mean that 70% of all lean mass otherwise lost throughout the body was preserved. Nor does it demonstrate, on its own, an improvement in strength or mobility. It does not establish that semaglutide causes pathological muscle wasting. It mainly shows that the quality of weight loss is becoming a therapeutic question in its own right. We talk a lot about the percentage displayed on the scale. But we also need to understand the composition of that loss and its consequences for the patient’s physical function. 8. Substantial weight loss can also affect nerves This is a much less familiar complication. The common fibular nerve — also called the common peroneal nerve — runs close to the surface on the outer side of the knee, near the head of the fibula. The surrounding tissues provide some protection. After substantial weight loss, that protection can decrease. The nerve may then become more vulnerable to compression. This can cause ankle weakness or even foot drop with a steppage gait. There is an old term for this phenomenon: “Slimmer’s palsy.” A Mayo Clinic case-control study published in Neurology in 2025 found an association between GLP-1 use and common fibular neuropathy, with an odds ratio of 1.3. Among the affected patients, median weight loss was approximately 15.7%. The authors highlighted the likely contribution of weight loss and nerve compression. This was an observational study. It does not establish causation or directly estimate the absolute risk for a patient receiving treatment. But it illustrates an important distinction: A complication can result from the consequences of weight loss rather than direct toxicity to the nerve. If treatment contributes to that weight loss, the complication can still be relevant to its safety profile. 9. Hair loss This is another issue some patients report. Hair loss appears in clinical trials: approximately 3% with injectable Wegovy at 2.4 mg and 4–5% with Zepbound, compared with approximately 1% with placebo. These figures reflect the populations and doses studied. Substantial weight loss, nutritional changes and the associated physiological stress can contribute to telogen effluvium, a form of diffuse hair shedding that is often temporary. This does not establish direct drug toxicity to the hair follicle. But it does not mean treatment plays no role either. Once again: We need to distinguish an event occurring during treatment, its possible causal relationship and the mechanism explaining it. 10. Thyroid cancer: a prominent warning, but relatively reassuring human data In rodents, some GLP-1 receptor agonists can cause thyroid C-cell tumors. This observation helps explain the warning in US prescribing information. But we need to be precise: The relevance of this finding to humans remains uncertain. A Scandinavian study involving 145,410 GLP-1 users, with an average follow-up of 3.9 years, found no substantial increase in overall thyroid cancer risk. That is reassuring. But it cannot exclude a very small long-term risk or definitively resolve the question of a rare cancer such as medullary thyroid carcinoma. The practical precaution also needs to be stated: US prescribing information for semaglutide and tirzepatide includes a contraindication in people with a personal or family history of medullary thyroid carcinoma or multiple endocrine neoplasia type 2 — MEN2. This does not mean every thyroid condition is a contraindication. Can we exclude a very small risk emerging after fifteen or twenty years? No. But these treatments have not been demonstrated to cause thyroid cancer in humans. Long-term follow-up remains necessary. 11. Suicidal thoughts, mood and the relationship with food This is an example of how a pharmacovigilance signal should be handled. Reports of suicidal thoughts prompted investigations by health authorities. After a comprehensive evaluation, the FDA did not identify an increased risk of suicidal thoughts or behavior with GLP-1 medications. On January 13, 2026, it requested removal of this warning from the US prescribing information for Saxenda, Wegovy and Zepbound. Its analysis included 91 placebo-controlled trials involving more than 107,000 participants, with no increased risk identified for suicidal events or the other psychiatric adverse events assessed. The signal was investigated and was not confirmed by that evaluation. But there may be another dimension, more subtle and much harder to measure in a clinical trial. A patient recently told me something that stayed with me. Eating had always been one of the great pleasures of his life. Since starting treatment, his desire to eat had decreased considerably — which was, of course, the intended effect. But he also felt that he had lost an important source of everyday pleasure. And he thought it had affected his mood. This anecdote does not establish anything scientifically. It does not mean the medication directly caused depression. Nor does it allow us to identify the mechanism of that change with certainty. But it reminds us that eating is not only a metabolic need. For many people, it is also a source of pleasure, comfort, rituals and social connection. A profound change in appetite can therefore transform part of everyday life. For some people, that transformation can be liberating. For others, it may require adjustment. It may be useful to distinguish a possible direct pharmacological effect on mood from the psychological consequences of a profound change in someone’s relationship with food. The absence of an increased psychiatric risk at the population level does not remove the need to listen to an individual patient’s experience. A persistent change in mood deserves assessment, whatever its cause. 12. Important precautions not to overlook The risks are not limited to the complications discussed above. Hypoglycemia The risk is particularly important when these treatments are combined with insulin or sulfonylureas. It is not strictly zero in other situations, but the accompanying treatment substantially changes its importance. Pregnancy and pregnancy planning These medications are not intended to be used for weight loss during pregnancy. Pregnancy planning requires advance discussion. The recommended interval between stopping treatment and a planned pregnancy is two months for semaglutide and at least one month for tirzepatide. Oral contraception with tirzepatide Delayed gastric emptying can affect the absorption of orally administered medications. US prescribing information recommends non-oral contraception or an additional barrier method for four weeks after starting tirzepatide and after each dose increase. Regulatory recommendations differ between countries: the European product information reviewed does not give the same instruction. The locally applicable prescribing information should therefore be followed. These situations help explain why treatment selection needs to consider other medications, medical history and the patient’s circumstances. 13. What about side effects we have not discovered yet? This is probably the hardest question. And the correct answer is: We cannot know everything yet. Millions of people are receiving these treatments, often at higher doses and for obesity rather than diabetes. Broader and longer use may reveal rare risks that the initial trials did not detect. But this uncertainty does not mean these treatments have been poorly studied. Semaglutide has large randomized trials extending over several years. SELECT included 17,604 people with established cardiovascular disease and overweight or obesity, without diabetes, with an average follow-up of approximately 3.3 years. Its safety analysis identified no new major safety signal in that population, while confirming more discontinuations due to gastrointestinal effects and certain gallbladder events. However, inclusion and exclusion criteria limit how broadly the findings can be applied to every clinical situation. Tirzepatide also has data extending to 176 weeks, more than three years, in the SURMOUNT-1 subgroup with obesity and prediabetes, with no new safety signal identified. That is reassuring. But: Three or four years are not twenty years. And the next generations — triple agonists, amylin combinations, muscle-preserving treatments and other new molecules — will need to establish their own safety profiles. They cannot simply inherit those of semaglutide or tirzepatide. Key takeaways The reality of side effects is nuanced. There are very common adverse effects, mainly gastrointestinal, often temporary but sometimes troublesome enough to cause treatment discontinuation. There are less frequent but well-established risks, particularly gallbladder disease. There are complications requiring real vigilance: pancreatitis, dehydration with acute kidney injury, severe gastrointestinal problems and, for semaglutide, NAION. There are widely publicized signals that were not confirmed by the evaluations cited, particularly suicidal risk and thyroid cancer in humans, while acknowledging the limits of follow-up. And we are beginning to study the consequences of substantial weight loss more closely: body composition, nutrition, physical function and compression of certain nerves. An indirect mechanism can still be part of a treatment’s adverse effects. As medications become capable of producing greater weight loss, the question cannot be limited to: “How many pounds or kilograms?” We also need to ask: “With what tolerability, what consequences and what lasting benefit for this patient?” The goal should not simply be to achieve the largest possible number on the scale. It should be to achieve meaningful, sustainable weight loss while preserving the patient’s health and physical function. Losing weight matters. How that weight is lost, and whether patients can live with the treatment over the long term, matter too.

Every data vendor says they're 95% accurate. But nobody can prove it. I spent six years in B2B GTM data, and one consistent theme is that they all run their own evals, and the enterprise struggles to evaluate the vendors accurately. So I built the check that shows which vendor's data actually worked on your leads. We don't sell data. We prove which data works. DMs open.

I also really liked the New York Earth Room

Intelligence is becoming universal. More tools carry it every day. Soon, more of the objects around us will too. What stays scarce is knowing what is worth doing with it. Which call to make. Which thread to drop. Which question is actually yours. That part stays with you. Everything around it is what we carry.

ATTENTION HAKKITO 🗣️ DON'T FORGET TO BUY YOUR HEAVYS HEADPHONES AND BANZOIN HAKKA SHELLS NOW AND YES, YOU CAN BUY THE SHELLS SEPARATELY WE HAVE LIMITED STOCK! DON'T MISS OUT ON BUYING THIS AMAZING COLLABORATION GET THEM HERE ->

Both feel ancient. One is much older. Which is older: sharks or trees? Tap to answer.

Every lot ships with its certificate of analysis. Scan the QR, verify the lot. Made in USA. For laboratory research use only.

Cocktails pouring. Conversation flowing. Discover Barsys 360. Shop now.

got an audience that asks you for advice? turn those questions into paid private voice replies. you set the price. they pay to ask. you reply when you want. no calls. no scheduling. create your page ↓

When we started writing, I wondered whether the argument would land with people who've operated at the highest levels. Then the endorsements came in. The CEO of Intel.  The founder of Zoom.  The former head of HR at Apple.  The cofounder of Whole Foods.  Four leaders who have built and shaped some of the most demanding organizations on earth, all pointing at the same thing: the best leaders are both respectful and effective, and they refuse to treat that as a contradiction. That's the whole book. But now I care about what YOU think. Now that the book has been out for a week, if you can post a review on Amazon and let me know what you think, I promise to read everything there. Link below: amazon.com/Tough-Enough-G…

A new residential IP on every request, with no GB meter. Unlimited residential proxies: $4.08 an hour or $7.35 a day at 10 Mbps. HTTP, HTTPS and SOCKS5. Try 1 hour free. No card needed.

Dead NFTs can be pretty scary.

Make October's Patreon post from sketches and a process video, with Superbot switching between Gemini, GPT-6 Astra and Opus 5.5 along the way. Free app, early access.

AI agents can pay for things now with x402. But should they do it without asking? Agent hits a paywall → Quirna checks your policy → a human approves with Face ID → the payment runs. Any action. Any agent. You decide.

Huawei says its Ascend AI chips have overtaken Nvidia's in China. It hasn't said by how much. Rotating chairman Eric Xu conceded at Huawei Connect last month that Nvidia's China share is hard to pin down, but said Huawei's own data puts Ascend ahead. Bernstein forecasts about 50% for Huawei and 8% for Nvidia this year. It's what Jensen Huang has warned about since export controls hit a market where, he says, Nvidia once held about 95%. Xu says the bottleneck is the process nodes available in mainland China, not chip design, so Huawei links large numbers of chips into what it calls SuperPoDs. Capacity is the other limit. Huawei can't meet demand at home, he says, so it has no plans for a full push abroad, though it's testing and shipping small volumes to several countries with "very strong demand." For scale: DeepSeek reportedly plans to deploy at least 160,000 Ascend 950DT chips at a data center in Inner Mongolia.

Dr. Kraken hypnotized 8 Guardian bosses. One sea monkey pilot is going to fix that (wait what?). Here's the first trailer for WINGNUT, a 16-bit shoot-'em-up where you pick your boss order and take its power. Mega Man meets shmups and Star Fox! Follow for more updates!

Radiant Gold on the wrist with the Gold Bracelet! What’s the verdict? Yes No

🏈 THINK YOU KNOW D.C. FOOTBALL? Put your knowledge to the test with HAIL YEAH TRIVIA! Play. Answer. Challenge your friends. Earn the bragging rights. STATS. LEGENDS. D.C. FOOTBALL. FOLLOW & PLAY TODAY!

One unit. Two completely different playstyles. Which one would you choose?

Everything you sell. One link. 🔗 - Free to join - Your own storefront experience Ready to grow? Open your store today. It's free

Everyone in your timeline posts their wins. We post our record. Every projection we have ever made, sorted into confidence bands, graded against what actually happened. Read it before you pay us a cent. $5/mo, college hoops and CFB.

Built my own app for the stuff I actually do every day: fasting, cold plunge, sauna, breathwork. One timer app, $4.99 once. No subscription, no account, no ads.

Your gym bag essentials? The right gear, a water bottle, and an app to keep you on track. Motiv8 helps you track workouts, follow programs, and reach your fitness goals. Click to download today!

Every batch is tested by an outside lab, with the COA posted before it goes on sale.

Tasks moved to AI. New skills became essential. Whole parts of the job disappeared. And the job description still describes the role as it was three years ago.