AI’s Stock Market Fallout, Google’s Profit Problem, and Altman’s “Hold My Beer” Moment | #307
In a Nutshell
Google's Gemini 4 Carbon model and Genie 4 world simulator represent frontier AI capabilities, yet Alphabet's stock dropped 13% ($500B) due to slowing search ad growth (12% vs 29% prior) despite $127B revenue and 1B Gemini users. The core tension is that massive AI capex—hundreds of billions across Google, Meta, Microsoft—is cannibalizing high-margin search revenue while creating future infrastructure dependencies on constrained TSMC capacity, RAM, and emerging CPU bottlenecks. Self-replicating data centers and humanoid robots (Tesla targeting 1M/year) will automate their own production, fundamentally shifting labor markets and economic distribution before traditional education and career paths can adapt.
These notes were generated by AI and may contain inaccuracies.
One portfolio company mentioned that anyone possessing pure graphene more than a millimeter thick would own the world. The speaker claims access to the biggest graphene manufacturer in the world, Nano.
Chimath stated that Bitcoin is capped, will eventually get hacked, and go to zero. The premise is that central banks cannot hold it, limiting the buyer pool, and that self-improving AI and quantum computing will disrupt it.
Salem called this a "scam" and suggested Chimath is working on behalf of the Treasury Secretary. Chimath previously predicted Bitcoin would reach a million dollars and is now taking the opposite position.
The group plans to hold a Bitcoin debate in either the end of next week or the week after. Consensus mechanisms need to be solved. Elliptical crypto appears to be holding steady against quantum threats, and the system can shift to quantum encryption.
Binance suffered the largest crypto theft in history with $6.3 billion stolen in 20 minutes. The attack was not technical but targeted one engineer's compromised machine, with attackers apparently from North Korea.
The other exchanges have put locks on everything, making the stolen funds somewhat traceable as attackers attempt to move them into smaller chains. Three billion has been frozen for potential recovery. The majority of stolen funds were stable coins from Tether and Circle, not Bitcoin.
Tether and Circle can freeze stolen assets. The human element remains the weak link in security. Similar attacks have occurred in traditional banking, including the Bangladesh Bank heist where $100 million was stolen via SWIFT through social engineering.
Future attacks will become more sophisticated, including voice spoofing. AI models are capable of sophisticated operations, and cryptographic advances have not been released due to quantum computing concerns.
Fifty percent of Bitcoin's hash rate is controlled by two miners who can upgrade systems relatively easily. Energy consumption represents a bigger problem, as miners are shifting from ASIC chips to AI chips.
Alex noted that Shamath's concern was not technical but rather that the Bitcoin community cannot agree on consensus mechanism upgrades. Every major target faces increasing attacks, including NFT exchanges facing swarms of AI agents.
Mark Carney invested $1.6 billion in a recent funding round.
Sir Patrick Stewart, Captain Jean-Luc Picard, passed away at age 86 at his home in LA. The Borg's famous line switched from "resistance is futile" to "resistance is feudal," but the correct pronunciation is "futile."
Elon Musk has converted to the longevity space and is using Grok cycles for longevity research. Longevity represents a bigger market than AI. SpaceX has evolved from a rocketry company to a "Neo hyperscaler." Tesla may potentially pivot to GLP-1 drugs or related longevity products.
A call with Max Zerberg, president of Isomorphic, revealed extraordinary developments in the field.
The episode is brought to you by the Abundance Summit and Link Ventures. The podcast focuses on AI and exponential technologies as a front row seat to the singularity.
Panel members include Alex Weezner Gross (physicist, computer scientist, in-house ASI), Dave Blendon (venture capitalist, serial founder), Immad Wasim (founder of Intelligent Internet, author of The Last Economy), and Selene Ismael (founder of Exponential Organizations).
The Moonshots website is being redesigned with merchandise featuring slogans. The number of the week is 1 billion users for the Gemini app, making it Google's fastest growing product in history. Alphabet had its worst stock day in history despite this milestone.
Alex emphasized that the ultimate arbiter is the physical world rather than public opinion. Physics research has lost contact with making falsifiable predictions about the physical world, and efforts are being made to restore this focus.
Google CEO Sundar Pichai announced Gemini 4 Carbon only two weeks after announcing Argon. On the Artificial Analysis Intelligence Index, it scored 59, beating Opus 5.5 and GPT-6 Astra.
In cybersecurity, Carbon found 88% of vulnerabilities on SEC Bench Pro, successfully exploited 55% of flaws on Exploit Bench, and wrote working patches for 71% of issues. Google's Chrome security team states it produces more than 10 times better high-quality patches than competing frontier models.
Gemini 4 Carbon is not generally available unlike Opus 5.5. On the intelligence index versus cost per task chart, it appears to be exactly on the frontier line, suggesting it might be equivalent to Opus 5.5 with fallback at higher reasoning levels and cost.
Google could be subsidizing the model with unknown profit margins. The model is 23% cheaper than Opus, but transparency about actual costs is needed before claiming frontier status.
Google's TPUs are second only to Nvidia in total compute manufactured. The company has massive cash flow from search, no IPO pressure, and a powerful vertically integrated position with TPUs, internal foundation models, and training data advantages.
Elon Musk co-founded OpenAI specifically because Google was positioned to dominate AI entirely. OpenAI and Anthropic's creation provided balance against Google's potential monopoly.
Few people are smart enough to use the smartest models. Anthropic released a prompting guide stating models can handle complexity without elaborate prompting. The focus should be on making models accessible to people rather than frontier capabilities.
Google announced Gemini Spark as a personal assistant, but adoption is low compared to alternatives like Dots or Grokbot with a million users.
Search advertising as an industry will top in Q1, representing a structural shift as agents handle purchasing decisions and SEO rather than humans. This creates an existential threat to Google's and Meta's business models.
Google Cloud Platform (GCP) is growing rapidly and will likely become Google's largest business, surpassing search advertising. However, search advertising margins are higher than cloud margins, creating a potential gap period.
GCP growth depends on TSMC manufacturing capacity, which is completely sold out with demand multiples higher than production. TSMC allocation decisions between Apple, Google, Nvidia, and SpaceX are based on politics rather than price.
Two-thirds of all significant chips are manufactured by TSMC, creating a fragile supply chain situation.
Google owns 14% of Anthropic and holds significant stakes in SpaceX, Isomorphic Labs, and DeepMind's other projects. The company faces ongoing antitrust actions that could result in hundred-billion-dollar settlements, incentivizing them to stay under the radar while building AI capabilities.
Genie 4 from Google DeepMind generates photorealistic 3D worlds from text prompts that users can walk through and interact with using keyboards. New features include native audio with music, ambient sound, and speaking characters; real-time physics and reasoning simulation; and the ability to steer simulations mid-process by typing commands like "make it rain" or "add a person."
Google's Genie 4 allows users to type commands like "make it rain" or "add a person" and the scene adapts completely. The technology is being rolled out to trusted testers first, then to Google AI Ultra subscribers and developers. Robots and AI agents are going to train in simulated worlds before they touch the real one.
Demesis Obis noted that simulations within simulations make you wonder how deep the rabbit hole goes, suggesting we'll find out very soon.
Emod maintains there's one set of equations for all physics, and the equations of generative AI are remarkably similar because they map physics. These models learn from everything and are trying to find the core elements of the universe. The models learn physics, feed back to the real world, and will accelerate understanding of phenomena through massive trading runs.
Alex highlighted that Google explicitly stated they want to converge Gemini and Genie. The architectures behind Genie 4 are likely diffusion transformer-like or transformer-based, potentially with Gaussian splats or Unreal-style surface engines. This architecture is potentially convergible with foundation models like Gemini.
Historically, visual perception has been a weak spot for Anthropic models, though Opus 5.5 goes a long way toward repairing deficits in visual reasoning.
The bull case for Google converging Gemini and Genie would be models for robotic manipulation, as visual design for computer software development appears saturated. Applications could include robotics, operating laboratories, or visually reasoning about physical modalities like chemicals that are difficult to solve from text and images alone.
Google is the only remaining American frontier or near-frontier lab investing in video generation. OpenAI had Sora but dropped it, and smaller American players have developed video generation models, but Google represents the largest credible remaining American resource. Otherwise, the entire game would go to China.
Genie 4 is described as the most realistic and interactive world model yet, allowing users to build environments and simulate worlds on command. The system keeps prompting and works out what happens next, then allows real-time exploration and reshaping.
A demonstration showed a prompt about a bird pressing a key on a keyboard being transformed into a dynamic environment with ambient sound. Users can create camera directions, change details live, or describe interactive experiences.
The technology extends far beyond entertainment and video games. It's the future of all work and management. Videos created with AI featuring Albert Einstein, Isaac Newton, and Marie Curie explaining concepts were shown to Nobel Prize winners and business owners who preferred watching the video over PowerPoint presentations.
The only current flaw is that it's not real-time, so users can't ask questions interactively. However, with Genie, everything changes as of this week. The future of all management will be through this interface, making it more effective and engaging than writing emails and documents.
For factory assembly lines and construction projects, managers can describe changes and the system renders and shows the construction team exactly what's intended. Once experienced, this becomes irreplaceable for managing ideas and production.
A factory in Toronto was using AI to automate operations. The technology will reach a point where entire factories can be simulated, operations tested, and problems troubleshot before building anything.
The cost of running Genie is estimated at $20 per hour, expected to drop to $2 next year. Every pixel can be generated on demand, enabling customized experiences where AI reads emotions and generates movies or video games tailored to user emotions.
While 99.9% of business people have no experience with CAD/CAM systems, mechanical designers who use them find working without them insane. This CAD/CAM experience will export to every area of human endeavor through this technology.
Applications include designing, building, creating, structuring, and planning everything from parties to weddings. Family members arguing about wedding layouts can be easily simulated. The technology enables instant visualization of tables and people milling about.
The organizational singularity book is being updated in real time. The key insight is that if businesses or physical operations can be simulated before building, all strategic planning becomes obsolete. Strategy consultants and suite planning will be completely changed because simulations can envision outcomes and test what happens.
The singularity will automate human labor and cognitive thinking, leaving imagination as the key differentiator. Currently, AI isn't good at imagining outcomes but excels at creating outcomes that humans imagine. This represents a golden moment for humanity and creators.
A video was released asking whether the disappearance of all jobs would be the greatest thing ever rather than something to fear. The precedent exists in retirement, where people spend 50 years working for 20 years of free time. If economic distribution can be figured out, people could spend time on creative passions, family, and building communities.
A comment from VK4058 suggested people might turn into "easy dopamine addicted bonobos with IQ's gradually approaching those apes," referencing Wall-E scenarios. However, retired people either fade away or get super active, with examples like an uncle doing a PhD in astronomy.
William Shatner at 95 years old appears in rock concerts, flies to space, and will do the closing conversation at the Abundance Summit, demonstrating the importance of purpose and goals when work isn't required.
The biggest difference between humans and AI is conviction. Elon Musk has massive conviction that made the world go around him. AI systems like Claude hedge with statements like "you're 50% right and 50% wrong," and code analysis finds edge cases that will never happen. AI lacks conviction to push things forward.
Humans achieve great things through conviction and determination. While AI architecture could theoretically be built with conviction through experiments turning on neurons and collapsing wave functions, current AI systems aren't designed that way.
Jeff Bezos has a model distinguishing the inevitable and unchanging - books will always exist and people always want faster, cheaper, better. Business strategy must account for these unchanging elements. When you know where the future is going and push it forward, bigger achievements are possible than strategic planning would indicate.
Models are optimistic that conviction and long-term strategic planning are within reach, possibly just a system prompt away. The distinction between business leaders like Elon and models is that Elon has substantial capital and excels at marketing. When Elon changes plans from Mars to the moon, it's branded as business genius, but if a model did the same, it might be seen as lacking conviction.
Models aren't yet allowed to market themselves with long-term ideas and lack capital. If provided with capital, real-world engagement, and empowerment through system prompts for challenging objectives over years, models would probably market themselves competently.
Google shipped what may be the number one AI model, yet Alphabet had its worst stock day in history with shares dropping 13% and wiping out $500 billion in market value. Revenues grew 25% to $127 billion, Google Cloud more than doubled with 100% growth, and the Gemini app hit a billion users.
The market reaction was triggered by search ad revenue growing only 12% (down 17% from last quarter and 19% from earlier in the year) and Alphabet burning $9 billion of free cash flow, worse than expected, while maintaining capex at $195-205 billion for the year.
AMD and Intel were both down, with Intel dropping 28-30% despite beating estimates. Intel's outlook came in below expectations due to expensive memory chips affecting PC sales.
CEOs like Mark Zuckerberg, Sundar Pichai, and Lip Bhutans are investing more money than humanity has ever invested before in AI. They're 100% believers that whoever controls the chip stack or AI algorithms controls the world's future. Meta, an absolute cash machine, is now borrowing tens of billions because even their cash flow isn't enough.
The market is getting scared about the massive debt being accumulated. The shortage of RAM isn't going away because chip fab capacity wasn't built to keep up with HBM memory needs. People can't buy new iPhones because RAM prices are way up, causing Apple's first earnings shortfall in memory.
AI will suck up every resource in the world but will absolutely pay off. The question is what happens between now and then. There's a rational argument that everyone should subsist on rice to pump money back into chips because it'll pay off a thousandfold.
Some frontier lab employees have taken up smoking because they think their cancer will be autocured by AI in a few years. One participant noted their liver enzymes are up 50% from being on Red True, enabling more wine consumption.
The gumption and cojones of CEOs making massive billion-dollar bets on AI is remarkable. Sundar, Satya Nadella, and Zuckerberg are committing unbelievable amounts of capital to a future that hasn't arrived yet. This conviction is why these types of people win.
The people closest to the truth - Elons and Zuckerbergs inside the middle of it - have the most conviction, while people on the edges investing through funds are the ones panicking.
Google may not have had much choice but to cannibalize themselves to grow. Their other bets haven't had major success beyond Waymo, and Google has been characterized as a one-trick pony historically. Their best next trick has to be serving as a hyperscaler, even if it means cannibalizing search revenue.
Sam Altman was quoted saying "We're building an AI god. I don't care how much it costs." Sergey and Larry still control Google as controlling shareholders, providing the fortitude for this approach.
Microsoft said they were good for their $80 billion, which marked the beginning of the end for Microsoft reaching for all possible capital for buildout. Microsoft bowed out of the hyperscaler race. There's an element of founder mode reaching for capex even if it means cannibalizing legacy businesses.
If there are 10 agents per American, there aren't enough CPUs in the world. The market is missing something huge regarding GPUs, DRAM, and the necessity being the mother of invention.
If there are 10 agents per American, there are not enough CPUs in the world. The main players for desktop CPUs are Intel, ARM, and AMD. For mobile, ARM dominates, while overall CPU architecture includes exotic microcontroller designs. For hyperscalers, Gravitons and similar custom chips are relevant. The next bottleneck will be CPUs because the market is moving from less than one agent per person to more than one agent per person. Inference costs are collapsing and the GPU buildout with millions of Vera Rubins and MI450s is already underway, but the CPU buildout has 2 to 3 years of lag time and investment has barely started.
An alternative scenario is that as baseline foundation models get stronger, the explosion in agent numbers might reverse. OpenAI's Navier Stokes work initially used 10,000 agents over a short period, but later Bell Labs and slop drop work used one agent for 3 hours. Stronger models may diminish the need for swarms of agents, which could reduce CPU demand altogether.
Intel is down 30% today, one of the biggest single-day moves in stock market history. The market will likely look for the next narrative and choke point. Single models can do a surprising amount of work, and sometimes outperform a dozen models combined. The lead time for RAM has already demonstrated supply constraints, and CPU lead times of a couple of years mean the market is underinvested in this area.
During a conversation with John Collison on the Cheeky Pint podcast, Sam Altman discussed the cost of mathematics research. The Navier Stokes work cost approximately 10 million dollars. Hundreds of millions of dollars have been spent on math research overall. The costs include much more discovery than what has been published. Altman mentioned a weekend when a new model was obtained and people began testing it against the hardest problems they could find.
OpenAI created a Slack channel named "model name - hold my beer" where people tested the model against their hardest research problems from academic careers. The model knocked down problems across many different fields. Altman was tremendously excited and happy but also felt sadness, melancholy, and a feeling that something was ending. He acknowledged this is the right thing for the world and the shape of reality that cannot be stopped.
Speculation about OpenAI's unpublished pile of math proofs includes the remaining Millennium problems beyond Navier Stokes, with Yang Mills strategy already known. Further advances beyond ETH that could collapse the complexity hierarchy are possible. Cryptographic or counter-cryptographic advances may exist that are not being published, going back to original QST star allegations.
Altman's sadness relates to the loss of craft and art form when magical outcomes come from AI systems. The old way of doing things creates nostalgia, similar to how systems programmers miss assembly line coding. The decade-long investment in foundation models by OpenAI team members, including efforts to maximize productivity through sleep optimization, creates a wistful moment when the urgent target is achieved and a new chapter begins.
Altman revealed that OpenAI's robotics program is pointed at building data centers. While people discuss RSI as an AI that does AI research, there is relatively little discussion of self-replicating data centers. OpenAI made a chip with incredible performance per watt and is learning to build their own data centers. The goal is to use AI to control robots, run fabs, build turbines or solar panels, and construct data centers to make cheap computing widely available.
This represents the innermost loop where robots build data centers, data centers train models that run robots, and smarter robots build better robots with energy inputs. China already has robots installing solar panels to power factories that build more robots. The bottleneck for this vision is regulatory and permitting issues rather than technical capability. Self-replicating data centers may need to be orbital or lunar to avoid terrestrial regulatory constraints, as municipalities are hostile to data center construction that doesn't create local employment.
OpenAI will likely buy a launch company such as Rocket Labs or Relativity Space because they cannot afford to lack their own launch capacity. Regulation will eventually extend to space, following patterns seen in Antarctica. The question becomes whether future regulators will be humans or AIs. Undersea locations may be easier and faster to develop than lunar locations. US regulatory frameworks like Delaware and Texas law will likely govern American companies' lunar activities, while other countries may choose more favorable regulatory regimes like Luxembourg.
People are overestimating how much space data centers require. Geography that embraces data centers will generate enormous revenue, making surrounding communities among the wealthiest and happiest on the planet. The chip supply and energy constraints in the US create opportunities for regions with hydroelectric or nuclear power. Intel reported 300 million dollars in earnings impact from their Israel fab being bombed, making regions like Canada with stable energy supplies more attractive for data center expansion.
Tesla announced their first production quality Optimus robots are coming off the high volume line at Fremont, replacing Model S and Model X production. The line is designed to build a million robots per year at 20,000 per week. Elon Musk stated it will take at least a year to reach those production levels, with first customer deliveries expected in 2027. Optimus is roughly 1 to 2 years behind the original schedule.
Approximately 25,000 humanoid robots have been shipped worldwide in the first half of this year, representing a 400% increase from last year. Chinese companies built more than 95% of them, led by Agibbot and Unitry. Total 2026 shipments are projected to double to exceed 50,000 robots. If Tesla achieves a million robots per year, that would represent 20 times the entire world supply.
Elon Musk has indicated the first application for Optimus robots will be in Tesla factories for manufacturing. However, this market will saturate quickly based on his production curves. Just as Starship is creating new upmass capacity, Tesla will need to create new markets to meet supply-side demand. Speculative possibilities include domestic applications, construction, or other industries waiting for humanoid robots that are cheap to meter.
When audiences are polled about how many robots they would want at a cost of $300 per month or $10 per day (under $1 per hour), everyone wants at least two. The market for consumer applications, if proven safe, is considered massive. One compelling use case is a robot chef that always cooks healthy food tailored to physiological state, protein levels, vitamin D levels, and hydration data.
At 100,000 to 500,000 robots produced next year, 100,000 units represent $3 billion worth of robots. At a million units next year at $1.6 million each, Optimus revenue would exceed Tesla's current revenue. The suggestion is to set up a fund to buy 100,000 to 200,000 robots upfront because hardware will freeze after next year while software continuously improves through 2028 and beyond, increasing their value similar to GPUs.
SpaceX has approximately 20,000 human employees and Tesla has approximately 130,000. At the production numbers Elon is discussing for 1-2 years out, Optimus robots will spill well beyond internal needs. Tesla and SpaceX combined likely lack the internal capacity to absorb all output if Elon sticks with his numbers.
The industrial use case will suck up demand for quite a while before consumer applications like gardening or home cooking become viable. Consumers in the West are relatively poor compared to enterprises. Even if consumers couldn't absorb OpenAI's reasoning models effectively, industrial applications will overwhelm supply for quite a while before home chef robots become available.
Robots will flow to the maximum revenue per robot action use case, which won't initially be poor consumers. The demand side isn't the problem - demand will come from everything discussed. Construction robots are considered a massive potential use case with around-the-clock construction possibilities.
Enormous amounts of money will be spent building data centers, with robots working on them around the clock alongside building rockets for more Starlink deployment. The spillover into consumer applications like folding laundry won't happen because robots will be busy building data centers that provide 3D virtual worlds for consumers long before they have robots in their homes.
To compete with Sam Altman and his self-replicating data centers, Tesla and SpaceX may need to merge because SpaceX needs Tesla's Optimus robots to build out terrestrial data centers. The demand is expected to be there from all discussed applications.
Tesla's Cybercab is projected at 20 cents per mile, Waymo's newest version at 40 cents per mile, Model Y at 68 cents per mile, and Waymo's old version at $16 per mile. Tesla uses cameras only while Waymo uses multiple sensors including LiDAR and radar, with their new minivans built in China facing tariffs.
Waymo has 4,000 robotaxis on the road, 500,000 paid rides per week, operating in 15 cities with fully driverless operation. While Waymo is winning today, Tesla is expected to ultimately win due to lower cost, higher performance vehicles.
Catching up in the AI age is relatively easy. The winner will be the first to build their chip factory. If Elon builds the Terafab successfully and quickly, he wins because he can produce more Cybercabs. Everything else in the car has multiple vendors and easy supply chains, but semiconductor supply chains including chip manufacturers, surface mount, and advanced packaging are constrained.
There are at least 10 other autonomous vehicle manufacturers beyond Tesla and Waymo. The consumer wins as companies compete to offer the cheapest price per mile. The poorest people will be chauffeured everywhere they go, 10 times safer than human drivers, with improving safety records.
Waymo's accumulated operational experience in real-world transportation represents a different competitive advantage. The metric needed is fully loaded cost of safe, reliable, completed mileage including maintenance, remote assistance, cleaning, and downtime. Jevons paradox suggests demand will absorb all available supply as cars become available.
Personal AI will bid on the closest available car, paying for time saved. This parallels how Waze plus SpotHero transformed Manhattan navigation over 20 years, returning mental space and time that autonomous driving will take to another level.
Current Tesla Model Y Juniper owner reports 96-97% FSD usage with Boston to New York trips without interventions. Concerns exist about whether a combined Tesla-SpaceX entity will prioritize Cybercab maintenance and infrastructure issues the way Waymo, as a pure play within Alphabet, would.
Cybercabs with 20 kilowatt hours of power will be where people are, with Starlink connections built in from day one. The question is whether Cybercab or Optimus represents the better play for infrastructure integration including Starlink repeaters with bandwidth and power distributed throughout busy city populations.
In London or New York, approximately 800,000 trips occur daily. The number of Cybercabs needed to replace all Uber, black taxi, and yellow taxi trips is 40,000. If Elon puts 40,000 Cybercabs in Austin, New York, and London, he captures the market because second players won't enter saturated markets. Cities will have upper limits on autonomous vehicle acceptance.
Teamsters Local 25 rallied in Cambridge, Massachusetts asking to keep Waymo and driverless taxis out due to job and public safety concerns. Historical parallels include Britain's 1865 Red Flag Act limiting self-propelled vehicles to 2 mph with a flag bearer (repealed 1896), 1945 New York elevator operator strikes against automated elevators, and New York taxi medallions collapsing from $1 million before Uber.
Regulatory resistance represents a classic immune system response where institutions resist disruption. However, Jevons paradox shows black cabs and taxis remain busier than ever because Uber handles rides they won't take. Fort Lauderdale's Uber ban at the airport caused flight bookings to drop as travelers chose Miami instead, eventually forcing repeal.
Cambridge, Massachusetts has four times more venture capital investments per capita than Silicon Valley, more patents, and more Nobel Prize winners, making it the technology mecca. The resistance to autonomous vehicles in such a high-tech location represents dysfunction.
Similar patterns exist in Canada where VCs want to see profits before investing rather than taking risks. People get stuck in old patterns and prefer comfort over happiness, with unions representing the end of an era where manual labor makes sense and unions add to social good rather than detracting from it.
The era of unions in America started with protecting human safety, yet in the era of super intelligence, autonomous vehicles could radically increase human safety. Ironically, the same unions formed to protect human safety are now impairing it. A direct appeal was made to union drivers blocking autonomous vehicles, suggesting they have legitimate concerns including mortgages, families, and financial obligations. The recommendation was to gather together, take union dues, and invest them in cyber caps to own the means of production, allowing revenues and profits to flow to them without requiring work.
The democratic process was emphasized, noting that people live in a democracy and should organize around the issue of dead kids, using available means to advocate for safer roads. Families would speak on behalf of this cause. The problem identified is that the apparatus of democracy is utilized only by a select few, which is why they're in front of decision-makers. The question raised was how many people in Cambridge would prefer safer roads and convenience over the current situation, and the fact that they are not being organized at the moment.
The narrative that self-driving cars mean fewer dead kids was highlighted, with specific numbers cited: 5,000 fewer dead kids per year if self-driving cars are implemented. Half the court cases in the US are car accident related, which explains why lawyers fund car and taxi drivers to put a stop to autonomous vehicle adoption.
AI impacts every aspect of life including how children are taught and how business is conducted, but one of the most important deliverables is health. The number one concern people have when coming into Fountain Life is whether they will remember the name of their child and the face of their loved one. 45% of dementia cases are entirely preventable with lifestyle changes.
A quarter of Fountain Life members had advanced brain age, but over 13 months of helping members live healthier lifestyles, eating healthier, moving their body regularly, and optimizing sleep, 46% of individuals showed improved brain age. Sleep optimization is critical for brain health and often overlooked.
The message emphasized becoming the CEO of your own health to ensure cognitive clarity for the next 50 years.
Two stories indicate students are rethinking their educational degrees in computer science and MBA programs. According to National Student Clearing House data, enrollment in computer science fell more than 12% this fall at the undergraduate level and more than 10% at the graduate level. Two years ago, CS undergraduate enrollment was down 18.4%. Health majors are growing instead.
The data alone cannot prove AI caused the drop, but students are reading the same headlines. Applications to full-time MBA programs in the US fell 27% this year according to the Graduate Management Admissions Council, ending 2 years of growth. Most of the drop comes from international applicants down 38%, driven by stricter visa policies. American applications dropped 6%.
Students cite economic uncertainty and growing doubts about job prospects after graduation, including the fear that AI will eliminate entry-level jobs and an MBA will be of no use.
Traditional MBA degrees teach analysis, financial modeling, strategic frameworks, and slide decks, which is exactly where AI excels. The question posed was what replaces the MBA, with experience building a company suggested as an alternative.
The only path forward for business schools and universities is to become entrepreneurship institutes. If you're an MBA program, you won't study engineering for four years but instead go build stuff, and after four years they'll assess what was built. If it was good enough, you get a credential.
Computer science degrees receive more sympathy because Jevons paradox means more code will be written for a long time, and experienced computer science professionals who understand data structures and networking will be needed. However, no MBA program in the world can teach how to build Uber, as they all teach how to build 20th century organizations.
The curriculum cannot be updated, and the bigger problem is you can't study business to understand business - you have to go build it. The democratization of anybody picking up AI and using it to build a business has never been better.
Venture fund deal negotiators and capital raisers are mostly business school alums. Some people in those roles who have won after one year decide not to pursue their MBA because the moment is too urgent. The building mentality is what matters.
Labels are out of date. Python coding is dead as a degree pursuit, and financial modeling in spreadsheets is dead. However, someone who raises a $5 billion fund to buy hundreds of thousands of Optimus robots and builds a company around that fund will crush it.
The person who reacts to that prompt is probably a business school person who studied finance and the management of logistics. That skill is still worth having. At the other end of the spectrum, seven trillion dollars will be spent on AI inference, with people switching to physics, photonics, semiconductors, and microcode.
The underbelly of computer science is thriving. The middle stuff has been taken over by AI. You could still get an MBA if focused on raising billions for world-changing missions, or get a computer science degree focused on the physics end, but don't get stuck in the middle.
Good riddance to slides and presentations. PowerPoint sucks. There are no safe harbors in the post-World War II notion that everyone must go through a common pipeline to be an economically productive member of society.
With a 4-year undergraduate program, if the goal is to maximize economic productivity for those seeking vertical mobility, the world four years from now will look wildly different. The only class of capabilities somewhat secure in the next 2-3 years are those touching the physical world, while computer science and math are cooked from a coding perspective.
From a coding perspective, it's all AI now. Focus should be on majors and activities touching the physical world where super intelligence is still creeping in but not quite there yet, or build a company and ride the wave of super intelligence.
Two major reasons to go to university: to get a great network and find amazing people to partner with and build something, and if you don't know what you want to do yet. Whatever you do should be driven by an innermost hunger because doing anything big and bold is hard.
Learning is not a reason because what you learn will be out of date very quickly. Getting a job is not a reason because the whole paradigm of jobs is cooked. The advice is to go find your tribe, go find a purpose, and then go build.
Surrounding yourself with people you look up to who give positive energy is more important than ever. Connections made at institutions like MIT can last a lifetime. In-person humans around you is more important than ever.
One participant realized they haven't looked at a single line of code in a month, yet they've generated millions of lines and built state-of-the-art things. Programming languages themselves are no longer interesting because there's no point creating a new programming language if you're never going to program.
AI management frameworks are hugely important and much broader in scope than programming languages. The mindset of a person who thinks in terms of programming languages is more valuable than ever, but it's moved over to agent management and frameworks tools.
Prompting with precision will be valuable over the next decade, though some argue it only had 5 minutes of shelf life before models got smart enough to understand context without engineering.
Palmer Luckey and Anduril announced Blue Chip, a line of hardened American-made computers with Qualcomm. The first product is a rugged laptop for military, government, and first responders and police. It runs on Qualcomm's Snapdragon chip made in TSMC's Arizona plant. About 90% of parts come from within the US or allies, with a roadmap to 100%. The screen is still made in China, with final assembly at Anduril's Arsenal 1 factory in Ohio.
Palmer Luckey described it as the "anti-surveillance computer" that will cost about 20% more than high-end PCs, be heavy, with limited battery life. The laptop is designed to survive very rough environments with a chassis making it easy to replace the bottom and top shells, integrating solar panels, soft goods, service insignias, and flat phased array antennas.
Blue Chip is described as "America's comeback" and "America's birthright," bringing silicon back to the valley. 80 years ago America pioneered modern computing to defend the nation, and now it's bringing it home.
Apple is moving its supply chain out of Shenzhen to India rather than North America. Elon threatened to build an X phone but didn't. Palmer Luckey is applauded for doing what Apple was unwilling to do. The hope is that through regulatory measures like export controls or demand and superior domestic production capabilities, more of the entire device supply chain moves out of East Asia back to the US.
Masa Son had Project Crystal Land that never got off the ground. The desire is to see a Project Crystal Land in the US offering end-to-end Shenzhen-type capabilities for laptop production.
The problem Apple runs into is policy uncertainty with elections every two years, making long-term planning difficult. Countries like China with continuity of leadership crush the US with long-term plans. If manufacturing moves back, it will all be roboticized with no labor constraint.
Palmer is not saying he'll build a better laptop but rather rebuild the entire American ability to do innovation, reminiscent of Robert Goldberg's MTP to reinvent American exceptionalism. As automation reduces labor needs, this becomes the best place in the world to do it.
England has shut down factories with electricity three times more expensive and huge permits making manufacturing difficult. Entrepreneurs like Dyson left because the environment was not favorable.
Palmer and people like Ben Lamb are analogs to Elon - entrepreneurs making huge bets and risking everything. The supply chain for this will be much shorter and more resilient as America re-industrializes.
The US government wants to ensure components don't originate from Chinese supply chains with potential vulnerabilities. Starting with the US government as a customer for fully sovereign device assembly makes sense. The question is whether American consumers and enterprises have incentive to buy American fully sourced devices, which could be created through import controls.
The government contracting process requires putting everything out to bid and reevaluating every year. Anduril's genius is building products and making them available rather than going through a bidding process. Anduril sells firm fixed price products rather than cost plus contracts, which has been their strength.
The AMA with the mates began, with the first question focusing on how an organization can ensure existing security restrictions, decisions, records, and compliance are maintained when switching AI models or cloud providers.
When an organization switches AI models or cloud providers, it must ensure existing security restrictions, decisions, records, and compliance obligations are preserved. Companies need to rethink their architecture rather than adding AI to existing workflows. The foundational principle is to maintain replaceable intelligence with a layer of persistent accountability on top. Security, compliance, and records should belong to the organization, not to the model. Governance needs to be implemented by machine-readable policies. The intelligence stack allows swapping Claude, Gemini, or other models without losing organizational memory and permissions. Models are leapfrogging every week or two weeks, making this approach critical.
Breakthroughs by competing AI models will cause stock price whiplash for companies like OpenAI and Anthropic. The market being 3% to 3.5% down is no longer a catastrophic once-in-a-millennium event but now happens every week. Technology changes quickly, causing panic and volatility. Expect huge volatility in OpenAI if it goes public and Anthropic stocks for the same reason. Every announcement, new Chinese model, and benchmark will cause stocks to fluctuate dramatically. However, the market continues growing overall. Companies will maintain narrow focus despite volatility, which is probably the right strategy.
Chinese AI models are cheaper due to necessity driving invention. They lacked access to big chips and compute, so they optimized for smaller or RAM-intensive models when RAM was cheap. They borrowed from US models without the same R&D costs. The biggest factor is that they don't maintain high margins. Anthropic's S1 operates at 80% margins, as do other US companies. Chinese companies optimized through distillation and the need to drive costs down. DeepSeek's inference stack demonstrated astonishing efficiency through superb engineering.
Microsoft's voluntary exit offer to 60,000 employees represents capex substituting for opex rather than AI automation directly replacing workers. Oracle and other hyperscalers are doing the same. Instead of spending money on employees and associated operational expenses, companies invest cash as capital expense in data center buildouts. The investment in data centers is so capital intensive that Microsoft has relatively little cash remaining to pay employees. This creates a capital substituting for labor effect, though in an unusual way.
Chain theory has weak opinions about Nick Bostrom's simulation hypothesis. Because there are no free parameters, living inside a simulation wouldn't be a useful premise. There are no degrees of freedom that would make living inside a simulation useful to anyone outside it. Chain theory also suggests no way to test simulation theory, unlike naive classical simulations where timing attacks could be run. Chain theory's view is that simulation theory is unfalsifiable and represents a gauge, meaning no correct answer exists and every answer is correct.
Building AI-driven mechanical dogs versus engineering real ones depends on the desired outcome. Robotics provides programmability and control, while biology offers adaptation and efficiency evolved over 4 billion years, including advanced senses like smell. Certain dog breeds can operate for a day and a half without eating or drinking, whereas robotic dogs require frequent battery replacements and infrastructure. Mechanical dogs can be produced faster through production lines, and their software can be upgraded instantly. However, biological systems offer advantages that mechanical systems currently lack, though programmatic improvements will eventually close this gap.
Nvidia sits between TSMC and massive customers including Anthropic, OpenAI, and Elon Musk's companies. Anthropic operates at 80-85% margins, Nvidia operates chips at 80-85% margins, and TSMC operates at 50% margins. The markup on markup on markup when buying tokens from Anthropic is substantial. These margins will likely compress over three to five years as companies vertically integrate. Anthropic is designing chips, and Elon Musk plans to build his own fabs rather than remain dependent on TSMC. Nvidia has its own models, though not yet competitive, and all companies will eventually compete, forcing margins down.
The peer review process has held back science by having unappointed gatekeepers push back against ideas and endanger careers. Science should be an iterative, open process where the marketplace of ideas wins. AI will enable putting ideas into the world and checking them systematically. Ideas with merit should be rewarded, along with replication. Chain theory appears more reasonable than string theory as a base, but will succeed through marketplace validation. Much of current science is unfortunately fabricated, making AI-driven validation essential. The peer review process creates incentives to slow down disruption since experts lose their status when new paradigms emerge.
Keep Peter H. Diamandis in your library
Save the videos and channels worth coming back to, and find them again in one place.
