It's pretty clear there is little working consensus on what a network effect is. Definitions I've now run across: - A derivative of Metcalf's law - A scale effect - Value of the product grows as a function of the number of users - The previous definition but with a flywheel effect - A flywheel effect - Company growth and value are the function of an actual interconnected network (telco, ISP, social network, github etc). where the interconnections define the value I guess words can mean whatever you want if you don't care about communicating.

Thx for the thoughts. I think the pushback would be great models are available from 5-10 players who have been neck and neck forever, so seems like inference margins will be quite low over time (although I hear your point that this isn't a certainty), and that LLMs don't seem to present a distribution threat for now.

Our clinical trial of AMIE, a patient-facing AI system, was just published in The Lancet. I want to explain what *I* think this study ACTUALLY means. It's about a fundamental question: how do we know that clinical AI works? 🧵⬇️ xitter.cf/Google/status/21083245…

I’m so excited to launch GPT-6 today with intelligent UI and the ability to start answering while still thinking! This is the result of many months of work from our team and so many amazing collaborators. With intelligent UI, the model can choose how to present information in the way that’s most useful to you, e.g. an interactive explanation of how a violin produces sound or a calculator to double or triple a recipe. We wanted to give the model a richer way to express itself while keeping the experience fast and familiar, and the answers comprehensive and factual. I especially love it for shopping :) We built a library of native, streamable components and a compiler that renders the interface as the model generates it. We trained the model to make thoughtful design decisions with these components, including when to use something interactive and when plain text still works best. GPT-6 can also start answering while it continues to think and work in the background. It builds up its answer as it goes, so you don’t have to wait to start reading useful content. This feels especially good for research queries. I remember talking about wanting this two years ago when we were working on deep research, and I’m so happy we finally get to ship it! We trained the model to take your waiting time into account and build an answer across multiple partial responses. Each should add useful information based on what it knows and has found so far, while keeping the full answer cohesive and accurate. Let us know what you think!!

Caution with the word "cure" is right, but incomplete if building trust and conquering cancer are paramount. In the AI ecosystem, Jensen Huang has described Nvidia's role as funding what venture capitalists will not and building what no one else will. Nvidia and the frontier AI labs could play a similar role in the cancer ecosystem. With global stature and expansive networks of capital and talent, they can reshape the landscape like few organizations can. They can attack gaps left by misaligned incentives, question old assumptions the field has molded itself around, and anchor startups and investors with a public roadmap. Today, AI is bolted on rather than assigned a central role. What if we reimagined the ecosystem around what humans and AI each need and where each excels? This essay reflects five years of researching cancer from the outside and aims to start a discussion about where we can go from here. It is long because cancer is a systemic problem and needs an interdisciplinary perspective. AI and cancer research share surprising parallels. In both fields, many ideas sound compelling in theory but die in reality. In cancer, findings may not translate to patients while AI methods may not transfer across models. Even success can be contextual. Pancreatic cancer illustrates the gulf between progress and cure. Over the past year, a drug from Revolution Medicines has invigorated oncologists by demonstrating unprecedented efficacy against RAS, a protein family implicated in roughly 20% of cancers and historically deemed undruggable. Small molecules usually need a pocket, a contoured groove to wedge into and hold, but RAS proteins are smooth globes with shallow ridges. The drug is a genuine breakthrough and can buy patients more time with loved ones, another birthday or graduation, or maybe the birth of a grandchild. Everyone who contributed deserves our gratitude for years of devotion. Nonetheless, progress is measured in months, not years or decades. The Phase 3 trial reported a median overall survival of 13.2 months against 6.7 months on chemotherapy. A gain of 6.5 months. Curing cancer is hard because no one fully understands human biology. Many aspects remain inscrutable despite centuries of remarkable insights, and even clarity at the population level may not impart understanding at the individual level. Humans are devilishly heterogeneous. What helps Shaq may not help you. What triggers lung cancer in an Asian American woman who never smoked may not affect a young man in Europe. As Stanford geneticist Michael Snyder puts it, a sick you is closer to a healthy you than to a sick me. His longitudinal study of 109 people found many health measurements stay stable within a person and vary widely between people. Population ranges matter, but your baseline may matter more. Cancer itself is heterogeneous. It is not a single disease but a set of them, varying by tissue of origin, oncogenic drivers, and the immune context of each patient. What succeeds against a blood cancer like leukemia may fail against a solid tumor like pancreatic cancer. Many cancers arise from genetic damage accumulated with age or through exposure to carcinogens such as tobacco smoke and ultraviolet radiation. Viruses and other pathogens cause an estimated 10% of new cases worldwide. Technology should accelerate our understanding of cancer, but AI fatigue in the life sciences is warranted. Countless deaths and failed experiments already condition scientists and doctors to distrust breezy optimism. On top of this, past efforts often exacerbated workloads instead of freeing time for patients. Skepticism has been amplified by crypto bros and growth marketers who breathlessly hype AI news, but also by AI experts who fixate on flaws. The worst critics are right about the wrong things. In tech, the classic pitfall is being right about limitations but wrong about their permanence. No technology is perfect, particularly one endeavoring to emulate an entity as opaque as the human mind. By fixating on flaws, you will find plenty but also miss how primitive AI models could morph into sophisticated systems, transforming the world and producing trillions in market value. The consolation prize is playing 3D chess in the Hall of Infamy with Paul Krugman, who in a 1998 essay titled "Why most economists' predictions are wrong" proclaimed that the Internet would not have a greater impact on the economy than the fax machine. He won the Nobel Prize in economics ten years later. Pessimism is natural but frequently conflates difficulty with impossibility. It underestimates the rate of innovation and how small improvements can compound. Surrendering to the problem as impossible merely guarantees no solution. Goals, especially audacious ones, demand belief. This mindset empowered us to invent the airplane and put a man on the moon, despite declarations of impossibility. Of course, we accomplished both before curing cancer for a reason. All three require discovery and engineering, but machines can be simulated, reset, and retested. No two patients are the same. No patient can be reset. Impossible metamorphoses into possible with belief, a refusal to accept demoralizing evidence as final without first exhausting every creative route. Belief is the enzyme, even if the transformation stays slow and encumbered by setbacks. Tech needs to give our life science colleagues more reason to believe. No one understands the constraints better than the clinicians and scientists who devote themselves to easing suffering and prolonging lives. Earning trust demands openness, not guardedness. The strategy, as with any seemingly impossible task, is to speak plainly and honestly about where we want to go, where we are, and what we must do. In the best case, the mission is to cure cancer, even after it has spread. In the worst, the goal is to reduce every cancer to a manageable condition, the way chronic myeloid leukemia (CML) is now largely controllable with a daily pill instead of killing most patients within five years. Today we stand far from such a world. Cancer kills nearly 10 million people each year, with the most aggressive cases stealing our friends and loved ones within months of diagnosis. Towering mountains and deep rivers block the way. Promises of a cure are deceptive because the probability of failure is staggeringly high. There can only be pursuits, not promises. But CML and breakthroughs like PD-1/PD-L1 immunotherapy hint that the terrain is not impassable. Similar leaps are what we seek. The path is forged sometimes by serendipitously connecting to an old trail or wandering long enough to discover one, other times by methodically chipping away, but always by pressing forward, stone by stone and plank by plank. It is reasonable to say the odds of curing cancer are infinitesimal. It is reasonable to say AlphaFold changed the world but did not solve protein folding. It is reasonable to say virtual cell models, at this time, fall short of the hype and may not even represent the right cells in the tumor microenvironment. It is reasonable to say AI models do not capture the biochemical and biophysical dynamics of one human organ system, let alone the whole body. It is reasonable to say AI trained on animal models and immortalized cell lines may not learn the reality of patients. It is reasonable to say small, unstratified studies may obscure human heterogeneity. It is reasonable to say much of clinical work is deciding what to ignore, but vignettes and multiple-choice questions spoon-feed scenarios where most details count. It is reasonable to ask whether models overfit to benchmarks rather than reason clinically. It is also eminently reasonable to say AI today covers a sliver of drug discovery, that no magic AI wand can wave away clinical trials. That beautiful binders for the wrong targets are worthless. Or that even if we develop a cure, we cannot designate it as such until years pass without signs of relapse. Daphne Koller details many of these issues in a superb essay, "Drug Discovery Has No Magic Wands." She decomposes drug discovery into three stages: disease-to-mechanism, mechanism-to-drug, and drug-to-patient, and argues AI largely concentrates on the second. My primary quibble is the decoupling of an interdependent system, illustrated by rocatinlimab. The antibody modulated the OX40 pathway as intended, but the mechanism may have let HHV-8, a typically dormant herpesvirus, escape immune control and cause Kaposi sarcoma in some patients. Koller, a distinguished expert in AI biology, later covers the systems-level principle in the essay, and the three-stage abstraction is valuable for conceptualization. But it sidesteps a core tension: when is it logical to separate the disease from the patient? Yes, the reasons to disbelieve are numerous. But stark differences separate today from yesterday and support grounded optimism. AI is more sophisticated than ever. Yes, it may hallucinate and sound no smarter than a babbling child, but it has also achieved superhuman feats in math, cybersecurity, and weather forecasting. In chip design, AlphaChip from Google DeepMind has produced novel layouts that now ship inside the hardware training frontier models like Gemini. In healthcare and biology, progress remains tangible even when verification is delayed and expensive. In early 2026, The Lancet published results from a randomized trial of over 105,000 Swedish women. AI-supported mammography raised sensitivity nearly seven points to 80.5% without reducing specificity. Computer-aided detection could not do this well a generation ago. EAGLE, a model developed by researchers at Memorial Sloan Kettering and Mount Sinai, predicts the presence of EGFR mutations from routine pathology images. EGFR status helps determine treatment in lung adenocarcinoma, the most common histologic subtype of lung cancer. In a prospective silent trial published in Nature Medicine, an EAGLE-assisted workflow would have reduced rapid EGFR testing by up to 43%, with predictive performance comparable to the existing rapid test. While AlphaFold did not solve protein folding, its database of over 200 million predicted structures helps scientists sharpen hypotheses and prioritize experiments on a scale infeasible with prior approaches. Rentosertib marks another milestone. AI helped both identify TNIK, a kinase previously investigated in cancer, as a fibrosis target and design the compound. Phase 2a results were published in Nature Medicine last year, and a 320-patient Phase 3 is now registered to test efficacy. The rate of innovation is accelerating. A jar of peanut butter enjoys a longer shelf life than frontier models. The field is also expanding. Billions in capital have flowed into AI-native startups, and researchers worldwide are contributing to cancer research and drug discovery. Cancer is a systemic problem, yet careers reward specialization. Frontier models can now track research and draw connections across fields, allowing specialists to answer questions and explore ideas outside their expertise. The reasons to doubt are many, but the reasons to believe are not few. Ultimately, we are all allies. No one wants to stay trapped in this world of cancer, not even the staunchest AI critic. Not believing is reasonable. The burden of proof belongs with the optimists. Meanwhile, we can all help enumerate key gaps and draw the map: - Which feedback loops between models and the real world remain neglected? - Which promising targets lack assays fit for automated screening? - What verification systems are missing? - What new modalities and instruments do we need? - Which datasets could help us understand and stop metastasis? - Which dogmas are most likely wrong? The cancer ecosystem is rich with expertise and infrastructure. Cancer Grand Challenges, for example, funds interdisciplinary teams to tackle intractable problems. Open Targets integrates genetic, transcriptomic, and clinical evidence into scores ranking target-disease associations. The opportunity is to complement these initiatives by designing for humans and AI from the outset. People thrive when creating and deciding, but dislike rote tasks. AI can automate menial work and analyze across domains, but needs supervision. The open secret of AI is how data and evaluations can dictate performance more than model architecture. Without data, you cannot train AI. And without evaluations, you cannot steer it. Some proposals: 1. Pilot an AI reference design in five rare cancers. Reference designs are blueprints for building complex systems. Nvidia develops them so partners can deploy and customize AI infrastructure without solving the same problems from scratch. Similarly, Nvidia and frontier labs can give disease foundations the template and resources for connecting talent, knowledge, and experiments into a learning system accelerated with AI. More than 300 million people live with roughly 7,000 rare diseases, over 95% of which lack an effective treatment. Piloting in five rare cancers focuses effort and tests what transfers. Chordoma is a bone cancer diagnosed in about one person per million each year. The Chordoma Foundation provides much of what the reference design could emulate and extend. The foundation supplies tumor tissue, disease models, and wet-lab capabilities such as target validation and drug testing. It curates priorities, resources, and knowledge, lowering the barriers for researchers to contribute. It fills gaps left by academia and industry and shares nonconfidential results, including negative ones. Models, like humans, trained only on success may struggle to learn what fails. Concretely: - Build the disease knowledge graph. Frontier labs seed it with claims from the literature, each linked to evidence and work reliant on it. The graph maps open questions, foundation priorities, and assets such as cell lines. People access it through ordinary AI assistants or AI-generated personalized interfaces. - Frontier labs donate compute and embed engineers. Foundations oversee implementation and tailor each graph using a shared schema and canonical identifiers. Corrections can then propagate across cancers and organizations. - Grant access. Model Context Protocol, or MCP, connects AI assistants to the graph, data, and tools the foundation permits. The foundation's priorities, judgment, and resources can then guide research. - Attract ideas with prizes. In tech, competitions can efficiently concentrate energy. ARC Prize drew 1,430 teams with a $1M purse. The Chordoma Foundation created a $500K challenge for binding TBXT, a key driver of chordoma, and included free biophysical evaluations. Anthropic and AWS backed a $50K hackathon for one child's rare disease. Frontier labs can supply compute and awards. - Turn hypotheses into experiments. The design specifies how an idea moves from the graph through expert review to an assay or disease model. Labs run the experiment and return results to the graph. 2. Seed open validation. The WHO says more than 7,000 rare diseases exist while the FDA says more than 10,000. How many are there? When experts disagree, how do people determine the truth? How about machines? In science, facts are born as claims and mature into truth through validation. Most truths endure while others are refined or overturned as new data, methods, or cohorts emerge. Consider MDA-MB-435, a melanoma cell line mistakenly used to study breast cancer before cell line authentication became more common. The origin was contested for years. By the time it was resolved, decades of breast cancer findings based on melanoma cells had already accumulated. Outdated or overstated claims can misdirect years of research, and the risk rises with automated scientific loops. Humans, however, are rewarded for novelty and not for the unglamorous work of validation. As such, we lack a public system and incentive structure for tracing what supports each claim and what depends on it, and for propagating updates through the literature. Open source offers a precedent, where code has a public history and unresolved problems remain visible. The frontier labs could seed open validation, starting in the same five cancers. The knowledge graph gives models and researchers a place to check claims. - Machines analyze, humans judge. Models revisit claims as evidence changes, trace dependent work, and highlight conflicts. They flag studies missing key controls or details needed to assess or replicate the work. Experts decide what the evidence supports and what merits retesting. Selected items become open issues in the graph. - Version control each claim. The graph preserves how a claim and its evidence change over time. Models can see what earlier work relied on and what the evidence supports now. - Fund validation. Pilot one validation per cancer, prioritized by what depends on it. Foundations can later apply to validate others. Frontier labs provide the capital. Publish every outcome, including negative and inconclusive ones, plus time and cost, and update the graph. - Standardize claim checks. AI assistants learn to query the graph before relying on a claim, recording the version used. 3. Scale up biological datasets. Koller sees a data chasm between what we can measure in patients and how drugs act inside cells. The average adult has roughly 30 trillion cells across hundreds of types and many states, each of which may respond differently to the same intervention. A TBXT drug may work in one state, for instance, yet not another. The chasm is enormous. As she writes, "The largest cell atlases assembled to date, now spanning hundreds of millions of cells, remain orders of magnitude too small to cover this space." The answer is not always data, though. In glioblastoma, a beautiful drug against the right target can still fail unless it crosses the blood-brain barrier. The answer is also not indiscriminate scale. One billion pulse readings cannot explain why lung cancer develops. The answer is creating and scaling the right data. For causal understanding, Koller argues for perturbational experiments to measure how a living system changes after an intervention. Observational datasets offer a pragmatic complement, though cannot alone establish causality. Drawn from patient tissue, blood, and other specimens, they help scientists identify patterns across tumors, generate hypotheses, and design experiments. While not the whole problem, data remains a central bottleneck because gaps can hide mechanisms and risks. Addressing this means scaling data both up and out. Scaling out, or integrating more of what exists, might have connected OX40 modulation to the risk of Kaposi sarcoma sooner. Yet gaping holes remain in cancer datasets. The Cancer Genome Atlas (TCGA), a landmark molecular map completed in 2018 and spanning 33 cancer types, demonstrates what scaling data up can achieve. It reshaped cancer genomics and has enabled thousands of researchers to deepen our understanding of cancer. These gaps puzzled me, but the explanation became clearer the more I studied biomedicine. It mirrors a pattern in AI, where research incentives and interests favor discovery over dataset construction. Practical burdens also hurt. Specimens are difficult to obtain, clinical records are messy, and assembling multi-site cohorts can take years. This leaves disease foundations and others to fill the void. As a result, many valuable datasets remain unbuilt. Take lung cancer in never smokers (LCINS). LCINS accounts for up to a quarter of lung cancers worldwide and would rank as the fifth-leading cause of cancer death on its own. In the US, up to 20% of lung cancers occur in never smokers, or people who have smoked fewer than 100 cigarettes in their lifetime. Among Asian American women with lung cancer, more than half are never smokers, and the figure nears 80% among Chinese American women. Proteins perform and regulate cellular activities, but some can drive tumor growth and survival when dysregulated. Many cancer drugs target such proteins. Yet most LCINS datasets stop at DNA and RNA, both imperfect proxies for protein abundance. RNA, the closer of the two, poorly predicted the abundance of more than one-third of 13,000 proteins across eight cancers. Proteomics, the measurement of many proteins, can reveal molecular subtypes while nominating prognostic biomarkers and therapeutic targets not inferable from DNA and RNA alone. In colorectal cancer, proteomics helped reduce a broad set of driver candidates to the most promising ones. In a 406-tumor lung adenocarcinoma atlas, protein patterns marked a group at higher risk of recurrence, including stage I tumors whose biology resembled advanced disease. Armed with proteomic insights, the field can enhance treatment strategies based on protein tests. Certain subtypes of lung cancer, for example, can overexpress the c-Met protein without corresponding DNA alterations. As a result, genomic profiling may fail to identify patients who could benefit from a c-Met-directed drug. LCINS is not a single disease. Beyond tumor biology and treatment, patients vary in genetic susceptibility and carcinogen exposure. Capturing this heterogeneity requires both scale and representative sampling. The 406-tumor atlas included 242 tumors from never smokers, but only seven came from the United States, leaving little power to resolve subgroups within American LCINS. Dedicated proteomic LCINS cohorts typically profile about 100 tumors and top out at 150. At those sizes, a subgroup representing 5% of LCINS patients yields about five to eight patients, compared with fifty in a cohort of 1,000. Recurrence analyses are constrained further because only a subset will recur, leaving perhaps just a few events. This also assumes the cohorts capture real-world heterogeneity, which may not hold. A 1,000-tumor atlas of LCINS adenocarcinomas linked to outcomes is better scaled for subgroup analysis. Focusing on one histology prevents cross-histology differences from dominating protein clusters. Adenocarcinoma is the most common form of LCINS and already has extensive genomic datasets and tissue collections. Unlike many missing oncology datasets, this atlas could rely mostly on archived specimens, decreasing the need for new biopsies. What remains is largely capital and execution: auditing specimens for sufficient tissue and consented use, standardizing workflows, and linking each tumor to treatment, recurrence, and survival. Nvidia and frontier labs have both the footprint and the motive to scale datasets across borders. International atlases fit awkwardly within national mandates, but AI models expected to generalize across human heterogeneity need global data. They also have the capital and could share costs across public datasets. The OpenAI Foundation, for example, plans to invest at least $1 billion over the next year across four areas, including life sciences and curing diseases. Given the disease burden of LCINS, frontier labs could start with this atlas, then ask disease foundations which molecular, imaging, and longitudinal datasets they need but cannot build alone. 4. Make clinical trials cheap, fast, and informative. In pharma, Koller observes an aversion to failure and a herd mentality. Novel targets per year have dropped from roughly 100 in 2015 to about 30 in 2024 while 37 targets each carry more than 50 programs. My diagnosis differs. When bets can run $100 million or more, caution is rational. Every pharma executive aspires to cure cancer, but must worry about job security like all of us. Evidence-based thinking cuts both ways. As CEO, are you promoting the executive who blew $100 million on an unproven target, or the one who shipped a profitable program against de-risked biology? As a patient, do you prefer the trial backed by human evidence or the untested one? As Koller points out, we cannot engineer around disease progression. To get 24 months of follow-up, we must wait 24 months. Cost, however, is different. We can make trials faster to launch and cheaper to run while maintaining robust ethical guardrails. Decades of tech history prove what falling costs make possible. More people can create and participate, and innovation blooms. More trials mean more ideas tested, perhaps more of them designed for human heterogeneity. Failures are inevitable, but learning from them could deepen our understanding of cancer. Twitter unfortunately cut this idea short, so the next essay will flesh it out. I believe certain labs could benefit from building public health infrastructure the way governments built highways. In the end, not overpromising is the right first step. State the mission, flag the gaps, and do what others will not. Stone by stone, plank by plank, it is possible.

2/2 Second, on the messaging around AI.  I do not agree that my messaging has been disproportionately negative.  In fact it has been about equally balanced between risks and benefits: I’ve written one major essay about each, and even in interviews where I discuss the risks, I make sure to frequently mention the incredible benefits as well as proposing possible solutions to the risks (short clips from my interviews that end up on social media tend to be disproportionately negative, as that gets clicks).  In fact, I wrote Machines of Loving Grace because I didn’t feel the AI industry was painting an inspiring enough picture of how the technology could radically transform the world for the better.  The bulk of the essay is devoted to refuting skepticism of AI’s potential in health and biology, and showing why I think it will actually be possible to cure most human disease in ~5-10 years, as crazy as it may sound to ordinary people and frankly to biologists as well (I used to be one!).  And, if you read my most recent essay (Policy on the AI Exponential), I discuss concrete proposals for how to streamline the FDA process to make sure the deluge of AI-accelerated drugs isn’t slowed down by the regulatory process.  I feel the urgency here: I lost my father to Hepatitis C only a few years before the development of direct-acting antivirals (sofosbuvir), which cure 95% of patients and probably would have cured him. I do agree that the public has a negative view of AI (and that this is a big problem), but I don’t think it is primarily caused by me or any other AI leader warning about AI’s risks.  I think it is fundamentally a crisis of trust.  I think that ordinary people don’t trust companies, governments, or the tech industry and always suspect that we are cooking up some new way to screw them over.  The causes of this go back decades and AI is just the latest iteration of it.  I don’t think that a glitzy marketing campaign with a positive spin (which some have advocated that Anthropic do) is the way to win back that trust — at this point, saying that AI will cure cancer is more a cliche than it is inspiring, and most people think it is deceptive.  The thing that will work is *actually curing cancer*.  I think by far the most accurate criticism of AI companies including Anthropic is that we haven’t yet delivered on our big promises to benefit the world.  That is totally on us, and I think it’s the criticism you should be making, instead of all this stuff about messaging and marketing. We are however doing our best to fix this: Anthropic is ramping up its efforts very quickly in biology and medicine, and we hope to have incredible results in the coming years and some early glimmers in the coming months.  When we’ve actually accomplished something real, the whole world will hear about it, as loudly as possible, you have my word on that.  But until then I don’t want to make empty promises, and in the meantime I feel compelled to speak honestly about the very real risks of AI and how to address them.  Honesty is the right thing on the merits, and in terms of public credibility and trust it is no worse than, and may in fact be better than, an approach that ignores or distracts from risks which people instinctively understand are real.

Wow, seems like Google is buying Spirit Airlines' enterprise data for $10m (outbidding Mercor at $7.5m). Basically includes every internal document, email, workflow, and codebase for a once $6B company. Honestly, $10m for 34 years of operational data really seems like a steal.

“We’ve learned and struggled for a few years here figuring out how to make a decent phone,” he said. “PC guys are not going to just figure this out. They’re not going to just walk in” — Ed Colligan, Palm CEO, on iPhone “There is nothing Tesla does that we cannot also do” — Sergio Marchionne, Fiat-Chrysler CEO “Personally, I think reusability is a dream. You shouldn’t present things that are not realistic. They’re not supermen, so whatever they can do we can do.” — Richard Bowles, Arianespace

Was just as excited as everyone else to read this cool new paper, and it felt like a trip down memory lane! XM seems to be the same as IMLE, and the motivation and insights are also similar. Having worked on IMLE for years, I'll provide some context for my followers 👇. 1/n

Cancer is a difficult problem.

We’re building Silicon Valley outside Silicon Valley.

Applications for the latest Standard Capital Series A funding cycle are now open! Apply by July 21, hear back by July 31.

Announcing a new division of Midjourney called "Midjourney Medical"

Barbell strategy for killing it in an age of superhuman AI: Simultaneously get as close to AND stay as far away from AI as humanly possible. 1. Get close — play with AI models, use them to help you think, ask them to teach you about the world, get them to help you create, work with them to write code, understand what makes them tick, embed them into your everyday life, have fun. 2. Stay far away — learn to tell stories, make eye contact, build a team, lead with courage, connect far-flung ideas, build lifelong friendships, debate persuasively, think forbidden thoughts, handwrite ideas, confess your fears, fall in love. Spend less time trying to master mental transformations that are purely mechanical — building spreadsheets, analyzing trades, balancing accounts, writing code by hand, following playbooks, searching for needles in haystacks. These are the emerging no-man's land, squarely the domain of AI. Venture to the extremes. That’s where all the fun is anyway.

"It's not going to be a one person job." 😏 @TeamLou23 expects to see multiple bodies on Wemby throughout this series.