"Not Even Wrong" Podcast
Investing in fundamentally new concepts and engineering practices with large impact.
Lyrik “Der frühe Apollo” von Rainer Maria Rilke. Latenz zwischen dem was im Innern des Künstlers schwebt und was der Geistesfilter durchlässt. ‘…denn noch kein Schatten ist in seinem Schaun, zu kühl für Lorbeer sind noch seine Schläfe..’. Kunstschaffen ist die Verneinung von spezifischen Frequenzen des Künstlers mit der Schwingung der Natur. Der menschliche Geist hilft dabei, ist aber auch eine Hürde. .’und später erst wird aus den Augenbrauen hochstämmig sich der Rosengarten heben, aus welchem Blätter, einzeln, ausgelöst hintreiben werden auf des Mundes beben.’ Aus dem Gedicht ‘Über die Geduld’ ’wer die Frage lebt, lebt allmählich, ohne es zu merken, in die Antwort hinein.’
What’s the problem with Nvidia? Stock underperforms relative to business . Margin pressure due to memory cost. Vera Rubin will drive tokens/MW up and cost/token down. System level focus on value according to Ian Buck . Agentic AI increases demand for coordination and memory. Vera Rubin solving bottlenecks. Engineering vs. scarcity. Innovation happens either within Nvidia or startups. Latter have leg up because focus on point solutions. Key questions; 1/2 Memory bandwidth vs. capacity. 2/2 Interconnect, optical becoming integral part of product.
Episode, September 15 2026 III
Lyrik. ‘Über die Geduld’ von Rainer Maria Rilke. Lesung. Diskussion. ‘..leb die Fragen und du wirst allmählich in die Antworten hineinleben’. New Age. ‘Score takes care of itself’. Ziel auf dein eigenes Leben, dass im Einklang mit deiner spezifischen Natur voranschreitet. Dann wirst du gute Formen annehmen. Kein Geisteswahn. Keine Natur entfremdete Dekadenz. ‘Holzfällen’, Bernhard. ‘Solving problems is what make us happy’, Deutsch.
Excerpt from “Forge to Liquid” a memoir by Krim Delko. As I write, I post less. Interesting observation. Excerpt from the 1990s about Business Process Reengineering (BPR) and downsizing. Today’s discussion about AI and its impact on the workforce. If BPR was a machine gun for middle management, AI is a nuclear explosion in the enterprise. AI developed based on human intelligence which is pattern recognition, classification and iteration. AI can do much more since more access to energy and compute. Humans must pivot. Focus on job, not skills. Less human labor, more humane jobs. Change of corporate culture from employment to entrepreneurial jobs. Forge to Liquid. Antifragility.
Bucheindrücke “Die Welt von Gestern’ von Stefan Zweig. Eloquenz überschattet inhaltlich mangelhafte Ideen. Zum Beispiel, seine Kritik am ‘Rasseneinheitswahn’ seiner Zeit ist berechtigt, doch er sieht nicht, dass er selber der Versuchung von Einheitswahn verfallen ist. Die kompromisslose Zuneigung zur ‘reinen’ Kunst, seiner Religion ja sogar seiner Rasse scheint ihm akzeptabel. Er kann es nicht beidseitig haben. Entweder du verurteilst jeglichen Wahn oder sonst bist du genauso schuldig und/oder Opfer. Jeglicher Wahn zeigt früher oder später die Krallen des Unterdrückers im Namen eines -ismus.
Book discussion “The Sense of an Ending” by Julian Barnes. What is the evolutionary purpose of nostalgia, memories? They define life. Barnes plays around with memory and facts. They intermingle and you loose perspective of which one is more relevant. History is written by the words that transcend the scrutiny of facts and societal acceptance. The story keep changing with perspective, new introduction of facts and new emotions. No such thing as an average life. Most lives consist of addition but occasionally, there is multiplication. The more multiplication, the better. ❤️ What do we choose to remember? It’s what matters to us. The more the better. Life should be about things you want to remember. Tony’s life lacks love. Emptyness.
Book discussion “Independence Day” by Richard Ford. Intellectual finger food. Every sentence could be a book. Referencing Emersion’s Self Reliance. What makes a man? What makes independence? Grows. Slow. Frank Bascombe experiments. Postmodern man. Everything goes. Racial, Religious tolerance. Footprint for a liberal US. Bascombe is full of contradictions. He’s a realtor, low echelon of society. Do the right thing. Real estate is the most honest of all profession. People show who they really are. Looser and a winner. This book is neither win nor loose. It’s a draw. Looser as a man. No soul. No love. No family. Winner in material terms. Search for relevance in a world of irrelevance. Liberal trifecta: Anxiety, moral high ground and self loathing. All good ideas start with physical acts. Do things and then think. Bascombe is a harbinger of what is to come for liberals. No grounding. Lost soul. Unhappy optimist. What is happiness? Love. Family.
Reaction to World Labs conversation. Goal of world models is to predict one of the three: 1/3 Action. Robotics. 2/3 State. Simulation. 3/3 Observation. Video. Originate in model based RL. Implicit prediction doesn’t rely on representation of reality (doesn’t care what a chair is but knows what to do with it). Explicit, like Marble, describe real world. Best real world AI is vertically integrated. Requires hardware, chips. software and model. Robot learning pipeline; learn from video, spatial intelligence. Bootstrap robots and feed back real world interaction to world model. Iterate. FSD is model free, implicit model. Predict actions. State prediction is simulation. Why so many world model companies? Statistical models work for video generation. Spatial intelligence. Explicit understanding of real world (what is a chair)
❤️ AI will not reach higher levels of intelligence unless it also develops better morality. Morality and Intelligence are two sides of the same coin. Morality is based on respecting other people’s conjecture, particularly those, you don’t share. Error correction. Truth seeking. Popper. Only this level of morality can reach higher levels of intelligence. Intelligence is the computational part of solving problems. Lack of moral progress has brought down regimes in human history, Romans, Chinese, Soviets, Nazis and Imperial Japan. US system has been co-evolving morality and intelligence. If you want superior AI, focus on morality. Brett Hall refuting orthogonality thesis.
Book discussion “The Sun also Rises” by Earnest Hemingway. Human condition on the brink of boredom. Remedy is fight and love. Nature is life. Decadence of Paris contrasts with real life in the mountains of Pamplona. Authenticity. Matador is the antidote to cosmopolitan life. Brett falls in love with Matador., with real life. Meticulous about craft. What is a good bullfighter versus great? Techniques. Sports analogy. The craft is what counts. Life is an accumulation of training details that accumulate towards a purpose. Detail elevates athletic skill to poetic expression. The closer you get to the edge, the closer to real life. In order to understand a people you have to fight them and love their women. Culture. France. Spain. Culture is what people pay attention to, what gets them going. France is easy. Just money. Spain more complex. Convoluted emotional currents culminate in the bullfight. There is no word in Spanish for bullfight. ❤️ The entrepreneur is the matador of society. Exposure to true benchmark, nature and market.
State of affairs. Where do we stand? Three drivers for Tesla. 1/3 Higher demand driven by FSD adoption and gas prices. FSD is demand driver because it delivers convenience, safety and lowers cost per mile. 2/3 Cybercab launch initiates iterative process to scale transport as a service. Large market, cash flow generation. 3/3 Space X merger puts TSLA shares in new light. Longterm potential. Shorts covering. Nvidia driver of stock is Vera Rubin platform. Solving agentic inference bottlenecks. CPU and memory shortage. Lowers cost per token, watt per token and time to first token.
Book discussion “The Sportswriter” by Richard Ford. Peak post-modernism. The man has no qualities, no purpose and he is ok with that. Continuation of Musil’s ‘Der Mann ohne Eigenschften’. Frank Bascombe doesn’t even search for purpose. Content with irrelevance of sportswriting. Metaphor for the irrelevance everybody is obsessed about. What really matters is what happens, but then there is a sliver of imagination humans add to the literal which makes life interesting. Love is an illusion humans long for but do they really live it? Bascombe lives like a leaf in the wind. Deep, cerebral novel. Every sentence could be a book. Ford teases you with cynicism and then let’s go. There is no objective truth, right or wrong. But then there is something. Bascombe is searching for. Is search the purpose? Literature covers what science misses, love, hate, fight, fear. ‘Try to explain that with quantum physics.’ ❤️ Fear of being lonely is maybe the fundamental driving force of humanity. What is the purpose of a job? It’s not what you think it is. It’s like the ice hockey metaphor; ‘A game broke out during a fight’. ‘A career happened during the search for love.’
Tesla 45% and Space X 55% combined. TSLA robots based on vision and Space X agents based on Grok. Connectivity, energy and compute. Preliminary model for TSLA and SPCX merger. Based on fair value (TSLA 570 and SPCX 210), Space X gets 55% and Tesla shareholders get 45%. Not much immediate upside for TSLA from merger. Merger logic. Tesla physical AI based on vision. Space X workflow management based on LLM, Grok. Connectivity, energy and compute both for robots, Grok and third parties. What’s not priced in? Tesla Optimus, learning from real world interaction. Space X starship and orbital AI. Fully integrated AI company. Physical and digital, hardware, compute and energy.
Book impressions “The Sportswriter” by Richard Ford. Part 1. Purpose. Sports as a metaphor for irrelevance. Is it irrelevant? But people care? The guy with his car in the basement is happy, even though he is a toll booth operator. Is work really what defines us, or is it our imagination,? Factual vs. literal. Make room for mystery. What job is to be done? AI will replace skills, but not jobs. If your job is to drive safely, you’re done. If your job is to accompany somebody as a driver, you’ll be fine. Financial analysis is not about modeling, but about taking risks by understanding counterfactuals.
Inference engineering is a systems problem and Nvidia is well positioned to solve it. Interview. How does Nvidia stack at inference? System problem. Vera Rubin addresses memory, KV-Cache allocation, context, networking. Full stack co-design. Open weight models require system approach towards inference engineering. ASIC players as vertically integrated model providers such as SpaceX. Nvidia for everybody else. Vera Rubin solves KV-Cache coordination between memory and GPU through CPU, memory stack and networking. Co-design expertise compounds. Open source requires flexible, systems approach towards inference. What is inference? Solution to problem. Eventually, inference will be fully automated.
Book discussion ‘The River is Waiting’ a novel by Wally Lamb. How to recover from a big screw up in life, from addiction? Truth. Honesty. Discipline. What leads to addiction and wreckage? Lies. Weakness. Self righteous. Cocktail of disaster. When life hits you hard, you must draw the line. Sunk cost. Don’t cling to relationships that cannot work. Let it go. Corby doesn’t. Justice and punishment. Corby should have never gone to prison. But when there, he should not be treated specially. There is a space between law, morals, doing the right thing and practicality. Corby misses that space and ends up tragically. Moral high ground is not worth it. Nobody deserves special treatment and particularly not because of some sort of identity. Tom Wolfe style.
The problem with SpaceX is - what is the business? First earnings deck reveals key issue with Space X as stock and company. What is the business? Rockets, Starlink? Yes, but that’s a small part. Larger fraction is an AI infrastructure company. That is all about tokens/dollar, tokens/watt. It’s not exactly what SpaceX actually meant to be. Most of the capex and investments flow into AI and not rocket and communication. Financial performance will be driven by AI, adoption, model, infrastructure etc. Nvidia cake. Much better to separate those parts. Go public with a solid launch and Starlink business and have a separate AI infrastructure company. Combined, there will be issues with morale and investors will be confused for a long time. Not good.
Stephen Wolfram talks about AI at Ralston College. Human intelligence is a function of human embodiment and nature. Specific type of intelligence. What we know is a subset of what’s knowable and a function of our embodiment, constraints, physiology. Constraints around energy scarcity, life span, necessity to reproduce etc. AI will develop different intelligence because its constraints are different. More energy, faster iteration, more counterfactuals, no death. Humans are special insofar their constraints are unique to humans. AI will develop new science and technology because it has different embodiment and will venture in other subsets of what’s knowable. Learn from experience. Currently, AI is anthropomorphic because it learned from internet knowledge which is human knowledge. Similar episodes are here, Ian McGilchrist talks about AI and humanity and here, Tom Griffiths talks about the laws of thought.
Berkley Agentic Summit 2026. Jim Fan and Sergey Levine. Levine introduces test time compute to robotics. Reasoning. In principle everything can be ‘pixel in - actuators out’. But this would require lots of training. Reasoning draws analogies from previously seen situations to unseen situations. Jim Fan draws analogy of robot learning with Tesla FSD. Ambient data collection. Native. Internet video of humans performing tasks pre training learns statistical relationships between objects, depth and semantics. Fine tune with RL and world models. Robot imagines trajectories and gets rewarded. Humanoid for factor is best way to learn from humans. Continuous learning from experience.
Tesla needs a moment of urgency. FSD unsupervised and Robotaxi are too developing slowly. This is a ‘sleep on the factory floor’ type situation. Musk should put more priority on FSD unsupervised and Robotaxi. Benefits for society are immense, such as safety on the roads, more options for disabled and elderly people, more convenience for everybody else and lower cost of transport. Lobbing groups such as auto industry, tort lawyers and other professional grifters are fighting FSD for mass market transport as a service. Tesla should fight back, hire SWAT teams in regulatory, political, technical and PR.
Book discussion ‘Venetian Vespers’ by John Banville. City has character and Banville builds an edifice of emotions around the city of Venice and elevates to one of the main characters. What is the genre of mystery? A plethora of trajectories in a maize of uncomfortable events. Story doesn’t culminate in one event. Offers many overlapping questions. Floral, precise language. Verbal mosaic of descriptions that create a vivid portrait of the narrator’s situation. Banville skillfully crosses the lines of literature and imagery. It almost feels like watching the novel.
Tesla earnings call Q2 2026. Exhaustion. What we didn’t get: Robotaxi scaling. Unsupervised FSD for customers. Meaningful guidance on Semi. Profitably growing energy business (declining revenue per Gwh). What we got: FSD showing traction in demand generation. ‘People buying FSD with car attached’. Demand for cars is strong. Optimus training pipeline is what we expect; pre train - train from humans, train from video. Mid train - Optimus factory and skill generation. High capex is not issue, as long as existing businesses show promising cash flow growth. Flywheel; car business finances energy and FSD. Energy and FSD finance Robotaxi, Semi. Robotaxi, Semi finance Optimus, Terafab, 4680 and other battery/refining operation. This flywheel not working - stock is down. Two interesting things Musk said: 1/2 Robotaxi driven by march of 9s. What is the metric to decide ready for scale? 2/2 Capital efficiency vs. speed. Investing in Tesla is more about us. how much pain can we take? Company is fine and stock will go up.
What is the role of humans in the era of AI? Inspired by a lecture of Ian McGilchrist at Ralston College. To answer this question we have to be clear about 1/2 What is AI and 2/2 What is a human. McGilchrist tackles the latter question with emotions like loss, and love. I disagree. Human is defined by its goals, which is growth, reproduction and survival. What is AI? It’s iteration at scale. AI doesn’t have energy constraints like humans. No life span, physical and physiological constraints. That’s human, intelligence constrained by energy, space and life span. We developed a specific type of intelligence over millions of years to pursue our goals. AI can handle counterfactuals at much bigger scale. Mirror of what we are what we are not. Our role will not change; survival, reproduction and growth. AI is much better at scaling counterfactuals and finding solutions to problems that lend themselves to those type of problems (Moravec paradox).
Braucht es Authentizität? Was ist ein authentischer Mensch und wo fängt die Grenze zum Heuchler an? Ist ein Künstler, der seine Kunst ausstellt, schon ein Heuchler? Soziale Normen sind nützlich. Doch im Roman Holzfällen, greift Thomas Bernhard genau diese Normen an. Warum brauchen wir sie und was ist der Nutzen von Authentizität? Sie ist ein darwinistischer Imperativ. Ohne Authentizität stirbt eine Kultur aus (Soviet Union, Faschismus). Es braucht individuelle Experimente, die auf authentischer Expression wurzeln. Authentizität und gesellschaftliches Abstaubertum ist eine heikle Gratwanderung.
Buchdiskussion ‘Holzfällen’ von Thomas Bernhard. Abrechnung mit intellektuellen Eliten, den ‘professionellen Abstaubern’, die sich in den Treppenhäusern der Gedankenbürokratie besser auskennen als mit der Kunst. Bernhard sieht die Gefahr einer Anschwellung von institutionalisierter Dummheit und Korruption. Das gilt nicht nur für Kunst, sondern auch für andere Papierintellektuellen wie Finanz, Recht, Politik und sogar Sport. Verkrustung der Gesellschaft weiter fortgeschritten. Der Citoyen ist tot. Der Abstauber ist im Namen der Gesellschaft zum aktiven Schmarotzer geworden. Wenn Unternehmen von Beratern befallen sind, handelt es sich um ein Verwaltungsgeschwür. Holzfällen ist die Medizin dagegen. Bezug zur Natur, zum Echten. Weg von der geistigen Verzerrung der Wahrheit. Unternehmer und Ingenieure sind die Résistance.
Loop Engineering is how innovation works. Andrej Karpathy. Autoresearch, which is an application of ‘Loop Engineering’. It’s a formal description of how to interact with agents. Three key components. 1/3 Agent generates code 2/3 Agents evaluates code against formal criteria and 3/3 Agent evaluates whether the result is good enough or further iteration is required. Most important factor is the third, decision, when to stop iterating. What is success? What is optimal value. Here is where humans still will matter. Loop engineering is how innovation always works. Take SpaceX, they loop until Starship is functional or Tesla, looping until FSD delivers Cybercab and Robotaxi. The company of the future is loop engineering at scale. Key value creation is first, define value function (mission) and second, find solution to problem of scaling.
Book discussion “My Struggle” by Karl Ove Knausgaard. Autobiographical novel. You create your own character based on deep introspection into who you are and how you relate to things. 1/4 Your persona is defined by how you relate to things, places actions. ‘I am the guy who has those books on the shelf’. What you do is not what defines you. It’s how you relate to those things. And that can change. Post subjective 2/4 Objective truth? Emotional torrent (his true feeling about his father is sadness, not hate). You don’t control that. That comes from elsewhere. 3/4 Death is ephemeral. Comes and goes His father is dead to him until he actually dies. Then he comes alive in an emotional torrent. 4/4 Boxed in. Think across the edges of the box. To exit the box you have to learn how to relate. Knausgaard thinks about art, people, places. ❤️Those things are defined by how he relates to those things. Each item has its own universe based on how the observer relates to it. A post subjective, quantum interpretation of reality.
Jensen Huang on open source models and the need to build agents with proprietary workflow manager (harness) on top of open source models. Interview with LangChain. 1/5 Nvidia offers Nemotron, open weights. 2/5 Why does this matter? If you outsource your AI to Anthropic, they’ll steal your business. Protect your Alpha. 3/5 Intelligence cannot be outsourced. Future of company is proprietary harness around open source model. Harness (workflow), data and context is the essence of a company. 4/5 Anthropic is easy to use. If your business doesn’t have much alpha you can use Anthropic. Low ROI businesses tend to use Claude. High alpha companies will use Nemotron. 5/5 Anthropic high valuation predicated on replacing most of the low ROI GDP. Nemotron serve high ROI businesses. Agents for agents.
Discussing Luca Carlone’s presentation on robotics and memory. Most of what robot models do is memorizing (Mac Schwager). Better solve the energy problem and memorize as much as you can. Or find more efficient ways to learn and combine memory with task specific problems? Carlone. 1/3 Carlone comes from old school engineering, optimizing and Frankenstein robotics. Wants to pivot to data driven, end do end and learning based. 2/3 End to end versus engineered solution. Spacial intelligence augmented with foundation models. Add semantics to robot models. 3/3 Bitter lesson. Experience. Let model learn what matters, avoid inductive bias. Tesla FSD has spacial intelligence without priors. Goal is reduce time to market for real world AI. Robot learning pipeline; pre-train from internet scale video. Mid-train for skill acquisition and fine tune against benchmarks. Intelligence and scale manufacturing is key. New compute and chip architecture. Davison (Graph based design).
Book discussion ‘Flesh’ by David Szalay. Coming of age story of a man in Hungary pre and post Communist era. 1/3 Human condition. Deep and subtle described in raw, direct form. 2/3 Prince Myshkin meets the Brothers Karamazov. Morals, justice and what it means to do the right thing. 3/3 Life happens to István. Eventually ends up in cul-de-sac. You must take agency and responsibility. Not taking action when wrong things happen is the same as doing wrong.
Tesla Cybercab is neither Cybertruck nor ChatGPT. It solves a real problem at low cost. Waiting to monetize FSD through car sales, software and transport as a service. Cybercab is inverse of Cybertruck. Cybertruck is high cost for not enough value due to lack to battery technology. Cybercab is designed to deliver transport as a service at low cost. FSD unsupervised solves real problems such as parking and autonomous transport. Zero to one technology cannot be reasoned by analogy. Compulsion over planning. Solving new problems creates more interesting problems (innovation stack). Time to solve problem is not crucial, but being on the right path is.
Andrew Davison at ICRA 2026. Spatial AI. Representation. Similar to Ranjay Krishna, Chain of thought for spacial understanding through efficient representation. Is it effcient to go from pixel to torque? Every model depends on priors. The most important prior is the choice of training data and architecture. Reasoning in space. Tesla FSD exhibits spacial intelligence. Because it’s trained on native driving trajectories. Bitter lesson. Hardware for embodied AI . Graph network like. Distributed compute and memory. Our approach to robot learning pipeline. 1/4 Pre-training on large scale video data 2/4 Mid training for skill acquisition through imitation, behavioral cloning, teleoperation or corrective RL (Kumar). 3/4 Robot in loop. Robot as data collection agents with counterfactual (shadow mode). 4/4 Fine tune against benchmark.
Sergey Levine on robotic foundation models. 1/3 Pre train on large data, video, simulation, imitation, cloning. Goal is to learn about performing tasks in real world. 2/3 Skill acquisition through automated RL. Robots must learn skills by interacting with real world. 3/3 Robot in the loop. Treat robot for real world data acquisition. Feed specialist knowledge back into general model. Focus on mid training. Sample efficient RL. Fast learning from real world interaction. Correcting, not showing. Let robot find its own trajectory. Future vision (our take); video generation to generate trajectories and select paths. On policy trajectory in real time. Combine with automated RL. In order to learn, robot must have counterfactuals to select from.
Embodied intelligence must happen through space. Commenting on Vladlen Koltun’s presentation at ETH Zurich. Artificial General Intelligence is a bad problem to solve. Intelligence must be useful to humans. Embodied intelligence can be accelerated with cognitive intelligence such as Grok or GPT. But spacial reasoning requires spacial learning. Counterfactuals in space (I move here but should have moved there..). Tesla FSD has emergent spacial reasoning from end to end training. It can be accelerated through cognitive and simulation tools. Robot learning pipeline; 1/3 Pre training from video. 2/3 Mid training through direct skill development in simulation. 3/3 Robot in loop. Robot is data collection engine for real space interaction (dataset combining video feed and corresponding actuator data). Feed data back into simulation, improve model through counterfactuals (Robot did this..but model predicted that). Eventually, learning from experience. No rewards, just behavior.
Book discussion ‘Son of Nobody’ by Yann Martel. Where does Western culture come from? Bridging Homer’s Iliad with modern life. Idealized philosophy clashes with real life stories, both in ancient times as well as today. Glory and heroism meet suffering, quarrel and sacrifice. 1/3 History is theory laden. Winners write history. It’s good to get a different perspective. 2/3 Western culture is built on the idea of sacrifice and kings must keep that in mind. No unlined power. Life of common man matters as much as glory of kings. 3/3 War is bad. Troy is perpetual war, like today with Middle East. Absurd, Iliadesque absurdity of fighting with no purpose. Avoid war through deterrence because fighting war is horrible.
Keynesian economics is charlatanism because it can’t predict. Austrian economics can. Just look at US government debt. Any science discipline must develop models and tools to predict. Austrian economics predicts that a constellation with Fed, no gold standard, dual mandate and US congress dynamics will produce high government debt and high inflation. That’s exactly what has been happening in the US. Keynesian economics is a tool to justify government intervention and deficit spending. MMT (Modern Monetary Theory) is taking Keynesianism to the next level by instating perpetual deficit spending. Link to podcast that inspired this episode.
Robot learning pipeline. Mid training through correction, not showing. Aviral Kumar presentation at ETH. When mid-training and robot goes off into bad trajectory, create counter factual by brining robot back initial state and let it figure out a better trajectory. Pre-training is a collection of possible robot actions. Mid-training is teaching the robot how to execute task. But instead of imitation learning, correction learning is only brining robot back to initial state and let it learn how to do better. Improve reasoning. Paper.
Discussing Shuran Song’s presentation on UMI (Universal Manipulation Interface). Crowd sourcing robot data. Handheld device with camera. Easy to use, low cost and relevant for robot hand manipulation. Stanford culture is crowdsourcing, learning from scale data. UMI problem with implementation. Why should people use this? Robot learning pipeline. Pre train from video. Skill based manipulation (UMI and others) for mid training. Fine tune against benchmarks. Open loop. Robot in loop. To learn something you need counterfactuals.
AI must be safe and accessible. Deterministic explainability is not the overarching goal. Reaction to Gradient Decent podcast interview. The goal of Tesla’s self driving technology is to make superhuman safety accessible at low cost. Explainability is important, but not overarching goal. End to end training is about giving up representation but gaining in exploration and learning. Why stick to human level understanding? Iterate at pixel level and develop newer, better solutions for transport. Same applies to other forms of AI. The goal of AI must be to deliver more convenience, more safety, more access at lower cost.
Book discussion ‘Normal People’ by Sally Rooney. Psychological discovery of human existence. Subtle, everyday events lead to deep insights. 1/3 Socio economic background of Ireland in early 2010s. Zeitgeist pre Brexit and Trump 2016. Lack of purpose. Men must have something to fight for and it cannot be an abstract idea. It must be real. ‘Literature is a fetich for cultured people to take them on false emotional journey to separate themselves from the common people about whose lives they like to read'. They find meaning in abstract goals such as Justice because that is much cheaper and easier than engaging with real, solvable problems. 2/3 Identity. What does it mean to be ‘me’. Cannot define by others. Dangerous. Happens within. 3/3 Power of love. Love is power. In order to love you must submit to the other person which makes you extremely powerful and vulnerable.
Solve the data problem in robotics. CVPR 2026 Part 5. Jitendra Malik presentation. 3dfy everything paper. Take images from video, represent the objects in 3D and then learn in simulation to manipulate. RL and in the loop fine tuning, which is you use robot data and counterfactuals (shadow mode) to learn from robots about the physical world then feed back into model. Tesla FSD is solving for video reasoning. Jitendra’s approach solves manipulation. Trajectory for robotics learning at scale. Pre-train from video. Mid-train with skill specific manipulation, imitation, RL and simulation. Fine tune against benchmarks.
Not selling Tesla to buy Space X. Tesla catalysts near term. Space X longterm. Hot air, removed from fundamentals. Not Tesla. Driven by self driving technology, which is maturing and ready for scale. Space X driven by Starship, which is engineering and even science problem. Space AI requires successful reusability of Starship. Lots of iterations required. Similar to FSD five years ago. Once Starship solved, unit economics of space AI feasible. Tesla and Space X merger creates full stack enterprise AI for digital and physical work and transport. Merger at least 50/50.