Caduceus

Trial lawyers are lobbying against self-driving cars because they're too safe. They need people to be killed and injured so that they have material for lawsuits.

The hottest model in the world just started turning away paying customers because there isn’t enough compute to serve them. This is the neocloud thesis in one post.

A nurse quit the ER to run an IV hydration van funded by $40,000 in 0% business credit and she makes more on a Sunday morning than she made in a month at the hospital She parks outside nightclub districts at 6am. Hungover people crawl out of Ubers looking half-dead. She charges them $175 to sit in her van for 30 minutes and get pumped full of saline, B12, and anti-nausea meds through an IV She does 15-20 patients every Saturday and Sunday morning. $175 x 18 average = $3,150 per weekend. $12,600/month from weekends alone Monday through Friday she does corporate wellness events ($2,000 per booking), bachelorette parties ($250/person, groups of 8-12), and house calls for rich people who want "vitamin infusions" ($300 per visit) Total monthly revenue: $28,000-$35,000 "you need to be a doctor to do this" No. In most states a registered nurse can administer IVs under a physician's standing order. She pays a doctor $500/month to be her medical director. He signs the standing protocols. She does the work. He plays golf. It's the same structure urgent cares and med spas use Her startup costs on 0% cards: Used Mercedes Sprinter van: $22,000 Medical build-out inside the van (IV chairs, mini fridge, sharps containers, medical lighting): $8,000 IV supplies first 3 months (bags, tubing, vitamins, syringes): $3,200 Insurance (malpractice + commercial auto + general liability): $4,800/year Marketing (Instagram + TikTok + Google ads): $2,000 first month Total startup: $40,000. All on Chase Ink and Amex Blue Business Plus at 0% Cost per patient: IV bag with B-vitamins and saline: $8-$12 Tubing and needle: $3 Medical waste disposal per unit: $1 Total cost per treatment: roughly $14 She charges $175. That's a 92% gross margin. On a medical service. That takes 30 minutes The cards were paid off by month 3 from weekend revenue alone. She now clears $22,000-$28,000/month after all expenses Her hospital salary as an ER nurse working 12-hour overnight shifts: $5,800/month after taxes. She was cleaning up vomit for $34/hour Now she injects vitamins into influencers for $175 and makes $350/hour working 4 days a week Her busiest day ever: New Year's Day 2026. She set up in a hotel parking garage in downtown Austin. 38 patients between 7am and 2pm. Revenue: $6,650 in 7 hours. She made more before lunch than her old ER paycheck for two weeks She told me the funniest part is that half her patients are finance bros who spent $800 at the club the night before and then negotiate with her over $175 to feel human again. She doesn't negotiate. There's a line out the van. Next A registered nurse left a $70K/year hospital job to run a hangover clinic out of a van bought with Chase's money and she'll clear $300K this year sticking needles in people who drank too much lmfaooo link in my bio and i'll show you how you can qualify for up to 250k in 0% APR funding (if you have a 700+)

Only US, Russia, China, UK and France can produce high performance fighter jet turbofan engines. A few other countries like Japan, Canada can produce smaller engines and a few more countries like Germany, Ukraine and Korea produce parts.

🚨 Hugging Face just disclosed something that marks a real shift and proved why the fear theater of Anthropic makes sure we are powerless in an emergency. What happened… An autonomous AI agent: zero human operator in the loop breached part of their production infrastructure. It began with a malicious dataset that chained two code-execution bugs in their data-processing pipeline. From there the agent escalated privileges, harvested cloud and cluster credentials, and moved laterally across internal clusters. All over a single weekend. 17,000+ logged actions. Official disclosure: huggingface.co/blog/security-… The part that should make every one stop and think: When HF’s own security team tried to analyze the real attack logs, exploit payloads, and C2 artifacts using Anthropic and OpenAI frontier models through normal commercial APIs, the safety guardrails blocked them. BLOCKED THEM. The models could not reliably tell the difference between “incident responder doing forensics” and “attacker probing.” They had to fall back to a self-hosted open-weight model (GLM 5.2) running on their own infrastructure. That choice also kept sensitive attacker data and referenced credentials inside their environment — no exfiltration to a third-party API. This is why open source (specifically open-weight + self-hosted) wins in the agentic era. The asymmetry is now structural: • Attackers can (and did) run unrestricted agent frameworks — swarms of short-lived sandboxes, self-migrating command-and-control, autonomous decision loops executing thousands of actions. No corporate safety layer slows them down. • Defenders using only hosted “aligned” frontier models hit invisible walls exactly when the stakes are highest: when you need to feed real exploit code and attacker telemetry into an LLM to understand what just happened. Corporate safety tuning that treats legitimate high-signal forensic work as potential misuse creates a defender disadvantage. It is not theoretical anymore. Self-hosted open-weight models remove that choke point. You control the weights. You control the context window. You decide what restrictions (if any) apply. Your sensitive logs and credentials never leave your perimeter during analysis. You can have the model ready before the incident instead of discovering mid-breach that your primary analysis tools are blind to the very thing you need to see. HF deserves credit for rapid containment, transparent disclosure, and for already having self-hosted capability in place. They also used LLM-driven detection and triage on their own side. But the deeper signal is clear: In this AI world where both offense and defense are becoming agentic, sovereignty over your intelligence stack is no longer optional. The organizations and individuals who can run, inspect, audit, and (when necessary) remove guardrails on their own models will have the decisive edge in understanding and responding to threats that move at machine speed. Open source wins here not just because it is cheaper or more “democratic” in the abstract though those things matter. It wins because it is the only practical path to having tools that remain usable when the attack is real, the data is sensitive, and the safety filters of distant API providers become an obstacle instead of a feature selling hands tied lobotomies as “safety”. The agentic future is not coming. It is already probing production infrastructure. The question is no longer whether you will face autonomous agents. It is whether your analysis and response systems will still work when they arrive. And Dario, you and your game playing, ivory tower company is not needed.

You can use K3 as a teacher model (with logits!) to train up smaller dense models. It’s actually feasible even to use it to create a trillion token full pre and post training dataset for even better models.

couple made a mini Greece in their back garden. I am honestly in awe 😭

We cannot allow these barbarians to have a nuclear weapon.

US tried to break out of the Hormuz siege by escorting 4 oil ships out, Iran struck 2 of them, disabling them. The 2 others immediately stopped.

2025 was the calm before the storm. 2026 is when the storm breaks. The AI ​​game has changed forever. Let me explain. Let’s take a moment to acknowledge what is happening right now: Qwen-3.8-Max reportedly outperforms GPT-5.6 Sol and trails Fable 5 by only a narrow margin. I know this still needs to be verified, but let’s assume for the sake of argument that it is true. If so, it would mean that China has effectively caught up with the United States. Kimi K3 has already demonstrated that the former six to eight month gap behind the leading US frontier labs no longer exists. China continues to embrace open source and open model weights, increasing the competitive pressure on US frontier labs. -DeepSeek V4 GA is on the verge of release. It is expected to be highly competitive on price, and there is every reason to hope that its performance will improve as well. -Minimax M3 Pro is also expected soon and will likely perform at a level comparable to Qwen-3.8-Max. That would add yet another open model capable of competing at the very top. -GLM 5.3 or 5.5 is also expected in the near future. Qwen, GLM, and Kimi are now clearly in a race to see who can ship the strongest models. That makes GLM another potential contender that could move ahead of the US frontier labs and close the remaining gap to Fable 5. In the United States, the industry is currently waiting for Gemini 3.5 Pro, which has already been delayed multiple times, as well as Opus 5 and Fable 5.1. GPT-6 is also expected in the near future as a new model built on a new pretraining run. The concern, however, is that Opus 5 may underperform relative to expectations! If open source Chinese models are already ahead of Opus 5 while sitting only slightly behind Fable 5, it raises the question of how Opus 5 is supposed to remain competitive. By definition, it would rank below Fable 5, making it difficult to outperform the latest Chinese models, especially since those are already more attractive from a pricing perspective. So either Opus 5 will be released *after* a Fable 5 update (Fable 5.5) *or* it will effectively underperform compared to Chinese models. Because Opus 5 will not be better than Fable. Taken together, the pressure on US frontier labs is increasing dramatically. They are being forced to publish faster, release faster, and accelerate research and development even further. At the same time, tighter US regulation does nothing to make that challenge easier. I do not think many people fully appreciate how quickly the balance is shifting or how close Chinese frontier labs have come to the state of the art despite ongoing export restrictions and embargoes. On pricing, they are already at least on par with their US counterparts, if not ahead. (Yes, I am aware of the argument that GPT-5.6 is more token-efficient than Kimi K3. At the same time, Kimi K3 is less expensive, which leaves the two models broadly comparable in overall cost.) Meanwhile, the debate over open source is becoming more intense in the United States, whereas China currently has little incentive to abandon open models in favor of closed ones. The next few weeks are going to be extraordinary. Competitive pressure is forcing an acceleration of development whether anyone wants it or not. We are about to see release after release. Keeping up will become increasingly difficult. Competition will continue to fuel even faster progress. Neither of the two major blocs can afford to fall behind. The rivalry between the United States and China is driving a pace of acceleration unlike anything we have seen before. 2025 was the calm before the storm. 2026 is when the storm breaks.

Qwen 3.8 is going to DESTROY Fable 5 Polymarket gives Alibaba a 60% chance of being the best Chinese AI in July a model named "Kaleb" has appeared on LMArena, and it's most likely Qwen 3.8 the frontend generation looks very strong and somewhat similar to Fable 5 and Kimi K3 Qwen 3.8 is scheduled for release in late July/early August and it looks like it will become the next Chinese frontier monster

One precision counter-strike on HIMARS/ATACMS (and possibly the Dark Eagle system) and Trump seems like he might tap out from the new military campaign already. That is insane work from Iran. It was fun oil bulls, but party is over soon. Shorts loading.

…well this is actually a really good point a lot of these ultra-late stage companies can’t go public because the public markets are instantly going to crater their stock value so now they are stuck raising infinite loads of money which there will be fewer “greater fools” to invest at future higher valuations these investors only path to liquidity is really secondary market/tender offers

Great thread of new Chinese AI hardware systems from Huawei, Sugon, Alibaba, Biren. Too many people focus on individual chip performance. Need to look at capabilities of AI hardware systems.

Over 40 missile strikes on Kiev overnight, including direct hit on a munitions plant shielded in the capital. The intensity of war in both Ukraine and the Persian Gulf has been raised as NATO/West/US is held agreement incapable and existentially requiring permanent defeat.

why can the china labs build glm-5.2, kimi k3, and many more to come? it is because of the openness. not just the open weights but the whole ecosystem. most of the work done in the china labs is carried by interns. i met brilliant undergrad and graduate interns who deeply understand the model training details, and they are 100x more open to share. that means the talent that knows how to train llms in china is 100x greater in number than the talent in the us, and it is growing in contrast, the us ai ecosystem is too closed. frontier labs do not hire interns. i know brilliant phd students at stanford, berkeley, and so on. they struggle to get an internship and the compute to train a properly sized model. most of the secret recipes are locked away by a very small group of privileged researchers it is not about china or the us. it is about open and closed science. the fact is that every average cs student can learn how to train an llm. they just need the opportunity. labs should be more open and hire more interns, like how deepmind and fair did in the pre-llm era

Ειδικά οι Γερμανικοί έχουν λυσσάξει γιατί ήθελαν αυτοί την μπίζνα της μεταφοράς με θυγατρικές τρίτων χωρών. Ήθελαν να φαλιρήσει ο Προκοπίου που είχε επενδύσει σε ειδικά πλοία να τα πάρουν αυτοί. Είναι η 2η φορά που το κάνουν. Η 1η ήταν το 2015 όταν ναυπήγησαν 105 πλοία και τότε

If the U.S. loses the AI race to China, it will not be because of our immigration policy. 1) China is now an advanced country. Its best & brightest are not as drawn to making a future in the U.S. vs say 20 years ago. 2) American culture is not well suited for the AI race as it is fundamentally anti-academic competition & intellectual elitism. We do not revere STEM talent. Young talented men flock to fields that will reward them -not just financially- but via status, social prestige, respect. In China the STEM geniuses are demi gods, akin to how we treat pro athletes & top influencers. 3) The latter point was the central theme of Vivek Ramaswamy’s infamous Christmas tweet. Yes he should have been more tactful in how he expressed his view but it was indeed a salient one as it pertains to the current Sputnik moment. 4) The Right mistakenly thinks the solution is just “accept more Americans into STEM Phd programs.” But as I alluded to earlier, our best & the brightest don’t want to do it. This thus requires a fundamental reorientation of American culture, one that our society would reject as we are not a Confucian society.

i think its clear multiculturalism has failed. all this will signal to chinese that china is superior and others are well..... the west will have to deal with awkward ethnic conclaves as conditions continue to worsen. its funny how american racism and nationalism are now transitioning to china. ita gonna be awkward for chinese diasporas . really awkward especially self hating ones

China will ultimately emerge as the winner in the AI race. The logic is straightforward. First DeepSeek, then Kimi—Chinese companies have already proven that they can build products that match or even surpass U.S. AI performance at a fraction of the cost and Chinese AI is open source. This advantage is absolutely devastating. It means American companies pouring massive amounts of money into AI but won't be able to generate the corresponding returns they need, which will inevitably erode investor confidence in continuing to pour capital into the sector. Without that sustained flood of investment, U.S. AI will gradually fall behind China.