{"generated":"2026-05-18T16:31:17+02:00","total_count":80,"site":{"title":"Philipp D. Dubach | Quantitative Finance \u0026 AI Strategy","url":"https://philippdubach.com/","author":"Philipp D. Dubach","description":"Exploring the intersection of Macroeconomics, Quantitative Finance, and AI infrastructure. Analysis of LLM unit economics, global monetary frameworks, and systemic liquidity."},"posts":[{"id":"c0e2ab45a80f4bd014ba9375c9cd2dc4","slug":"reconciling-enterprise-ai-revenue","title":"Reconciling Enterprise AI Revenue","url":"https://philippdubach.com/posts/reconciling-enterprise-ai-revenue/","description":"Four enterprise AI revenue figures span a 40x range. The $63.2B audit-grade floor is the only tier that defensibly underwrites $690B of hyperscaler capex.","date_published":"2026-05-17","date_modified":"2026-05-17","type":"Analysis","categories":["AI","Investing"],"keywords":["enterprise AI revenue 2025","AI capex revenue gap","$690 billion AI capex 2026","AI capex coverage ratio 2026","Menlo Ventures AI report 2025","$37 billion enterprise AI Menlo","Gartner $1.5 trillion AI spending 2025","IDC AI solutions spending guide","SaaSpocalypse paradox","audit-grade AI revenue","vendor census enterprise AI","AI model API revenue","OpenAI ARR 2025","Anthropic ARR 2025","Anthropic 40% LLM API market share","Microsoft Azure AI revenue","AWS AI run-rate","hyperscaler AI capex 2026","AI revenue reconciliation framework","disclosure tier framework AI","Tier A SEC AI disclosure","Tier B earnings call AI","NVIDIA Data Center segment AI","Broadcom AI semi revenue","Spread Index AI","AI capex sustainability","CoreWeave revenue 2025","Dell AI server revenue","SuperMicro AI revenue","Palantir AIP revenue","Baidu AI Cloud disclosure","vendor run-rate sum AI","BofA hyperscaler debt 2026","telecom 1999 capex coverage comparison"],"word_count":2700,"reading_time_minutes":13,"image":"https://static.philippdubach.com/cover-capex.png","summary":"Four enterprise AI revenue figures span a 40x range. The $63.2B audit-grade floor is the only tier that defensibly underwrites $690B of hyperscaler capex."},{"id":"a6c7bcda4bb854bac9d948cf28d1bd03","slug":"midyear-portfolio-review-the-rotation-worked-europe-didnt","title":"Midyear Portfolio Review: The Rotation Worked. Europe Didn't.","url":"https://philippdubach.com/posts/midyear-portfolio-review-the-rotation-worked-europe-didnt/","description":"Midyear review of the 2026 CHF portfolio: EM, small-cap and Japan paid off; the Europe overweight underperformed; CAPE rose to 42; the dollar didn't fall.","date_published":"2026-05-14","date_modified":"2026-05-17","type":"Analysis","categories":["Investing"],"keywords":["midyear portfolio review 2026","Shiller CAPE 42","S\u0026P 500 Shiller CAPE 2026","European equity valuation discount","Europe overweight underperform 2026","USD CHF outlook 2026","S\u0026P 500 concentration NVIDIA","NVIDIA 8% S\u0026P 500 weight","60/40 portfolio benchmark CHF","AI capex hyperscaler 2026","JPM S\u0026P 500 7600 target","Mike Wilson 7800 target","Germany fiscal pivot equities","emerging markets rotation 2026","US Treasury safety premium","IMF Fiscal Monitor April 2026","Bitcoin allocation portfolio 2026","small cap rotation 2026","gold UBS 5900 target 2026"],"word_count":1536,"reading_time_minutes":8,"image":"https://static.philippdubach.com/midyear-cover.jpg","summary":"Midyear review of the 2026 CHF portfolio: EM, small-cap and Japan paid off; the Europe overweight underperformed; CAPE rose to 42; the dollar didn't fall."},{"id":"c4b2a5c0e5dc7ce0bb23f3d223264c32","slug":"","title":"What Claude Thinks But Doesn't Say","url":"https://philippdubach.com/posts/what-claude-thinks-but-doesnt-say/","description":"Anthropic's natural language autoencoders translate Claude's activations into readable text. The method works. The press release skips three structural problems.","date_published":"2026-05-11","date_modified":"2026-05-17","type":"Commentary","categories":["AI"],"keywords":["Natural Language Autoencoders","Claude evaluation awareness","Anthropic interpretability","activation verbalizer","foundation model audit","Anthropic NLA paper","Claude interpretability","activation reconstructor","Claude Opus 4.6 audit","sparse autoencoders vs NLA","attribution graphs","mechanistic interpretability","model auditing reward sycophancy","Claude planning rabbit rhyme","warm-start NLA Opus 4.5","Goodhart law interpretability","Anthropic Mythos Preview audit","verbalizable activations LLM","AI safety evaluation awareness","Transformer Circuits NLA paper","GRPO interpretability","Anthropic May 2026 paper","AI model auditing cost","how do natural language autoencoders work","sparse autoencoders vs natural language autoencoders"],"word_count":2496,"reading_time_minutes":12,"image":"https://static.philippdubach.com/claude_understanding_image.jpg","summary":"Anthropic's natural language autoencoders translate Claude's activations into readable text. The method works. The press release skips three structural problems."},{"id":"725bf1abb4377745edfeb66c4a0b0b70","slug":"","title":"Two Anthropics","url":"https://philippdubach.com/posts/two-anthropics/","description":"Two Anthropics: the safety lab Dario founded in 2021 and the $380B frontier lab it became. Same organism, two narratives the company itself has to reconcile. Three scenarios for how the tension resolves.","date_published":"2026-05-09","date_modified":"2026-05-17","type":"Commentary","categories":["AI","Investing"],"keywords":["Two Anthropics","Anthropic 380 billion valuation","Anthropic Paradox","Anthropic moat","Anthropic safety thesis","Anthropic vs OpenAI safety","Dario Amodei race to the top","Anthropic Series G 2026","Anthropic Responsible Scaling Policy","AI safety lab business model","AI safety lab valuation","frontier AI lab valuation","Anthropic OpenAI 2023 CEO offer","Anthropic DoD First Amendment ruling","Pottinger chip export controls","AI lab pricing framework allocator","Anthropic safety vs commercial pressure","race to the top AI safety","Anthropic enterprise customers Pfizer United Airlines"],"word_count":2398,"reading_time_minutes":12,"image":"https://static.philippdubach.com/ograph/ograph-two-anthropics.jpg","summary":"Two Anthropics: the safety lab Dario founded in 2021 and the $380B frontier lab it became. Same organism, two narratives the company itself has to reconcile. Three scenarios for how the tension resolves."},{"id":"1a83c481c1617107787a548aae10f64b","slug":"","title":"The Anatomy of a Decentralized Prediction Market: Notes from the Polymarket Order Book","url":"https://philippdubach.com/posts/the-anatomy-of-a-decentralized-prediction-market-notes-from-the-polymarket-order-book/","description":"A 624 GB tick-level archive of Polymarket's WebSocket feed joined to the on-chain trade record reveals eight cross-sectional stylized facts and a measurement result: the public feed only recovers trade direction 59% of the time.","date_published":"2026-05-02","date_modified":"2026-05-17","type":"Project","categories":["Quantitative Finance"],"keywords":["Polymarket microstructure","decentralized prediction market order book","Polymarket WebSocket feed","CTF Exchange OrderFilled","longshot spread premium prediction markets","trade direction inference Polymarket","Lee-Ready algorithm crypto","Kyle's lambda Polymarket","prediction market wash trading","Polygon CLOB microstructure","decentralized finance market microstructure","on-chain trade direction","30 billion order book events","Polymarket research replication"],"word_count":2280,"reading_time_minutes":11,"doi":"10.48550/arXiv.2604.24366","image":"https://static.philippdubach.com/ograph/ograph-polymarket-microstructure.jpg","summary":"A 624 GB tick-level archive of Polymarket's WebSocket feed joined to the on-chain trade record reveals eight cross-sectional stylized facts and a measurement result: the public feed only recovers trade direction 59% of the time."},{"id":"9aa76202af06fee44c8926dd5e61c7e5","slug":"","title":"Karpathy's Software 3.0 Playbook","url":"https://philippdubach.com/posts/karpathys-software-3.0-playbook/","description":"Twelve lessons from Andrej Karpathy's Sequoia interview: Software 3.0, vibe coding versus agentic engineering, jagged intelligence, and why December 2024 was the inflection most people missed.","date_published":"2026-05-01","date_modified":"2026-05-09","type":"Commentary","categories":["AI","Tech"],"keywords":["Andrej Karpathy Software 3.0","Karpathy Sequoia interview","vibe coding vs agentic engineering","Software 3.0 paradigm","Karpathy never felt more behind","agentic coding workflow","December 2024 AI inflection point","LLM jagged intelligence","reinforcement learning verifiable rewards","context window as program","LLM as interpreter","agent-native infrastructure","Karpathy taste spec oversight","ghosts not animals AI","MenuGen Karpathy","10x engineer agentic era"],"word_count":1962,"reading_time_minutes":10,"image":"https://static.philippdubach.com/ograph/ograph-lessons-from-karpathy.jpg","summary":"Twelve lessons from Andrej Karpathy's Sequoia interview: Software 3.0, vibe coding versus agentic engineering, jagged intelligence, and why December 2024 was the inflection most people missed."},{"id":"b7524e3985d18a802967eb5c91afa454","slug":"","title":"F3ED Can't Call an Ace: Fixing a NeurIPS 2024 Tennis Model","url":"https://philippdubach.com/posts/f3ed-cant-call-an-ace-fixing-a-neurips-2024-tennis-model/","description":"F3ED, the NeurIPS 2024 tennis shot detector, mislabels 73% of single-shot serve unforced errors. A 23-line scoreboard OCR reconciler fixes them.","date_published":"2026-04-29","date_modified":"2026-05-04","type":"Project","categories":["AI"],"keywords":["tennis shot detection","tennis broadcast computer vision","tennis match analytics open source","fine-grained tennis event detection","automatic ace detection tennis","F3ED tennis model audit","tennis scoreboard OCR pipeline","score-grammar reconciler","TenniSet V006 benchmark","ATP Challenger video analysis","shot outcome classification","EasyOCR tennis scoreboard","YOLOv8x tennis player detection","CatBoost bounce detection filter"],"word_count":1903,"reading_time_minutes":9,"image":"https://static.philippdubach.com/ograph/ograph-tennis-vision3.jpg","summary":"F3ED, the NeurIPS 2024 tennis shot detector, mislabels 73% of single-shot serve unforced errors. A 23-line scoreboard OCR reconciler fixes them."},{"id":"0d4b4136a3dab40f3cee1d5198db2498","slug":"","title":"Inside PRAGMA: Revolut's Foundation Model for Banking","url":"https://philippdubach.com/posts/inside-pragma-revoluts-foundation-model-for-banking/","description":"Revolut's PRAGMA is a 1B-parameter encoder trained on 24B banking events. Reading the paper, comparing with Nubank's nuFormer, planning a rebuild.","date_published":"2026-04-26","date_modified":"2026-05-17","type":"Analysis","categories":["AI"],"keywords":["PRAGMA","Revolut foundation model","banking foundation model","nuFormer Nubank","transformer banking transactions","self-supervised learning transactions"],"word_count":1199,"reading_time_minutes":6,"image":"https://static.philippdubach.com/ograph/ograph-pragma-revolut.jpg","summary":"Revolut's PRAGMA is a 1B-parameter encoder trained on 24B banking events. Reading the paper, comparing with Nubank's nuFormer, planning a rebuild."},{"id":"46aeda2f9b3be552db93f8a5feb5a51a","slug":"","title":"The Moral Philosophy of Investing in Ignorance","url":"https://philippdubach.com/posts/the-moral-philosophy-of-investing-in-ignorance/","description":"Constraint arbitrage, the sidecar problem, and who bears the distributional cost of investing under ignorance. The final installment of Edge of Knowledge.","date_published":"2026-04-22","date_modified":"2026-05-17","type":"Analysis","categories":["Quantitative Finance"],"keywords":["ethics of investing under ignorance","constraint arbitrage fairness","Zeckhauser blame aversion investing","institutional investor career risk","Monday morning quarterback MMQ risk","sidecar investing ethics Gazprom","Munger Zeckhauser bridge quote","UU investing distributional consequences","Ricardo Waterloo investment unknown","fiduciary duty ambiguity penalty","market efficiency ignorance box","pension fund UU constraints","Buffett California earthquake authority ethics","complementary skills vs power investing"],"word_count":1518,"reading_time_minutes":8,"image":"https://static.philippdubach.com/ograph/ograph-moral-philosophy-investing-ignorance.jpg","summary":"Constraint arbitrage, the sidecar problem, and who bears the distributional cost of investing under ignorance. The final installment of Edge of Knowledge."},{"id":"a3f4c8b641a3232702af9bd414c9e583","slug":"","title":"Bet Sizing at the Frontier","url":"https://philippdubach.com/posts/bet-sizing-at-the-frontier/","description":"The Kelly Criterion assumes you know your probability of winning. In a UU world, you don't, and heuristics like Zeckhauser's Maxim B replace false precision.","date_published":"2026-04-17","date_modified":"2026-05-09","type":"Analysis","categories":["Quantitative Finance"],"keywords":["Kelly Criterion limitations","position sizing under uncertainty","Kelly Samuelson debate","Zeckhauser Maxim B","unknown unknowable investing","fractional Kelly half Kelly"],"word_count":1785,"reading_time_minutes":9,"image":"https://static.philippdubach.com/ograph/ograph-bet-sizing-frontier.jpg","summary":"The Kelly Criterion assumes you know your probability of winning. In a UU world, you don't, and heuristics like Zeckhauser's Maxim B replace false precision."},{"id":"a98cc90b93dab270ed8cd2af36483a25","slug":"","title":"Do Not Disturb My Circles","url":"https://philippdubach.com/posts/do-not-disturb-my-circles/","description":"AlphaFold cost under $1M to train. OpenAI spends $2.3B on inference. The chatbot era consumed the talent and compute that could have cured diseases.","date_published":"2026-04-13","date_modified":"2026-05-17","type":"Commentary","categories":["AI"],"keywords":["AI for science vs chatbots","Demis Hassabis AlphaFold Gemini","AlphaFold training cost","AI compute scientific research funding","Bell Labs monopoly research decline","AI chatbot misallocation","AI opportunity cost","AI brain drain academia","scientific AI funding gap","AI capital allocation 2026","DeepMind scientific AI","AI infrastructure spending science","Hassabis cure cancer quote","chatbot era capital misallocation","AI for science underfunded","fundamental research AI funding","AI resource allocation science commercial"],"word_count":2596,"reading_time_minutes":13,"image":"https://static.philippdubach.com/ograph/ograph-do-not-disturb-my-circles.jpg","summary":"AlphaFold cost under $1M to train. OpenAI spends $2.3B on inference. The chatbot era consumed the talent and compute that could have cured diseases."},{"id":"1525e4cb142a4b89ebe2c55eafd2cf51","slug":"","title":"The Geometry of Who Knows What","url":"https://philippdubach.com/posts/the-geometry-of-who-knows-what/","description":"When neither side can define the states of the world, adverse selection fears are misplaced. Zeckhauser's information matrices and constraint arbitrage.","date_published":"2026-04-13","date_modified":"2026-05-17","type":"Analysis","categories":["Quantitative Finance"],"keywords":["information asymmetry investing","adverse selection financial markets","sidecar investing Zeckhauser","Buffett California Earthquake Authority reinsurance","winner's curse investing","Aumann agreeing to disagree","constraint arbitrage institutional investors","limits to arbitrage institutional constraints","Monday morning quarterback risk","Akerlof lemons problem investing","complementary skills investing edge","Summers sidecar objection","Robb uncommunicable knowledge","fiduciary duty ambiguity avoidance"],"word_count":1733,"reading_time_minutes":9,"image":"https://static.philippdubach.com/ograph/ograph-geometry-who-knows-what.jpg","summary":"When neither side can define the states of the world, adverse selection fears are misplaced. Zeckhauser's information matrices and constraint arbitrage."},{"id":"a7a265c0adc963babb1491f16e0e9f00","slug":"","title":"Why Lilly's Weight Loss Pill Isn't a Peptide","url":"https://philippdubach.com/posts/why-lillys-weight-loss-pill-isnt-a-peptide/","description":"Oral semaglutide destroys 99% of its active ingredient per dose. Lilly's Foundayo skips the problem entirely. Inside the $70B oral GLP-1 pill race.","date_published":"2026-04-09","date_modified":"2026-05-14","type":"Analysis","categories":["Medicine"],"keywords":["Foundayo vs oral Wegovy","orforglipron vs oral semaglutide","oral GLP-1 weight loss pill 2026","oral semaglutide bioavailability SNAC","GLP-1 market size obesity 2030","Foundayo orforglipron FDA approval 2026","Foundayo weight loss pill price","Novo Nordisk vs Eli Lilly GLP-1 2026","oral weight loss pill GLP-1","small molecule GLP-1 receptor agonist","oral semaglutide Wegovy pill 25mg","GLP-1 manufacturing cost COGS","semaglutide patent expiry 2031 biosimilar","oral GLP-1 pipeline aleniglipron VK2735 amycretin","GLP-1 adherence persistence discontinuation","anti-obesity medication market expansion","Jevons Paradox statin analogy GLP-1"],"word_count":1764,"reading_time_minutes":9,"image":"https://static.philippdubach.com/ograph/ograph-oral-peptides1.jpg","summary":"Oral semaglutide destroys 99% of its active ingredient per dose. Lilly's Foundayo skips the problem entirely. Inside the $70B oral GLP-1 pill race."},{"id":"8217d284a01f0b3e1aa68e0460b5b28f","slug":"","title":"Ambiguity by Design","url":"https://philippdubach.com/posts/ambiguity-by-design/","description":"Ellsberg proved people flee unknown odds. Zeckhauser showed their flight creates mispricing. Part 2 on ambiguity aversion, comparative ignorance, and investing.","date_published":"2026-04-08","date_modified":"2026-05-17","type":"Analysis","categories":["Quantitative Finance"],"keywords":["ambiguity aversion investing","Ellsberg paradox finance","Zeckhauser unknown unknowable","comparative ignorance Fox Tversky","Knightian uncertainty mispricing","ambiguity aversion vs risk aversion","Gilboa Schmeidler maxmin expected utility","Bewley inertia uncertainty","betrayal aversion Bohnet Zeckhauser","SaaSpocalypse IGV sell-off ambiguity","ambiguity premium equity mispricing","Ellsberg paradox explained","behavioral finance ambiguity aversion","career risk ambiguity aversion","institutional constraints uncertainty"],"word_count":1617,"reading_time_minutes":8,"image":"https://static.philippdubach.com/ograph/ograph-ambiguity-by-design.jpg","summary":"Ellsberg proved people flee unknown odds. Zeckhauser showed their flight creates mispricing. Part 2 on ambiguity aversion, comparative ignorance, and investing."},{"id":"badf7bb656fcf48311dbfaa20da01c2e","slug":"","title":"Three Kinds of Not-Knowing","url":"https://philippdubach.com/posts/three-kinds-of-not-knowing/","description":"Knightian uncertainty splits not-knowing into risk, uncertainty, and ignorance. A century after Knight and Keynes, most of investing still ignores the split.","date_published":"2026-04-04","date_modified":"2026-05-09","type":"Analysis","categories":["Quantitative Finance"],"keywords":["Knightian uncertainty investing","risk vs uncertainty investing","risk uncertainty ignorance framework","Zeckhauser unknown unknowable","investing under uncertainty","unknown unknowns investing","radical uncertainty finance","Frank Knight risk uncertainty profit","Keynes weight of evidence","Knightian uncertainty explained","unknown unknowable UU investing","decision making under uncertainty finance","Buffett California earthquake reinsurance","Kay King radical uncertainty","Ricardo Waterloo bonds","enterprise software AI uncertainty 2026","types of uncertainty investing"],"word_count":1706,"reading_time_minutes":9,"image":"https://static.philippdubach.com/ograph/ograph-edge-of-knowledge.jpg","summary":"Knightian uncertainty splits not-knowing into risk, uncertainty, and ignorance. A century after Knight and Keynes, most of investing still ignores the split."},{"id":"61740c0bd52c337e0b7517be84c391d3","slug":"","title":"On-Device AI Models Will Be The New Reason to Upgrade Your Phone","url":"https://philippdubach.com/posts/on-device-ai-models-will-be-the-new-reason-to-upgrade-your-phone/","description":"Smartphones haven't had a compelling upgrade story in years. On-device AI models, distilled from frontier systems like Gemini, are about to change that. Parameters are the new megapixels.","date_published":"2026-03-25","date_modified":"2026-05-09","type":"Commentary","categories":["AI","Tech"],"keywords":["on-device AI smartphone","Apple Gemini distillation","smartphone upgrade cycle AI 2026","Apple Foundation Model 3 billion parameters","NPU TOPS smartphone 2026","iPhone on-device AI model parameters","parameter count smartphone marketing","megapixel myth AI equivalent","Gemini Nano on-device models","Apple Intelligence upgrade cycle","knowledge distillation AI iPhone","on-device LLM smartphone 2026","smartphone AI hardware differentiation","smartphone spec war AI parameters"],"word_count":985,"reading_time_minutes":5,"image":"https://static.philippdubach.com/ograph/ograph-on-device-models.jpg","summary":"Smartphones haven't had a compelling upgrade story in years. On-device AI models, distilled from frontier systems like Gemini, are about to change that. Parameters are the new megapixels."},{"id":"d415f18ce15d9164d42ffb49377dbfab","slug":"","title":"AI Can Now Design Drugs in Seconds; We Still Can't Tell You If They Work.","url":"https://philippdubach.com/posts/ai-can-now-design-drugs-in-seconds-we-still-cant-tell-you-if-they-work./","description":"IsoDDE doubles AlphaFold 3 on hard benchmarks and beats physics-based gold standards. But no AI drug has FDA approval. What $4B in pharma deals actually mean.","date_published":"2026-03-18","date_modified":"2026-05-17","type":"Article","categories":["Medicine","AI"],"keywords":["Isomorphic Labs IsoDDE","IsoDDE vs AlphaFold 3","AI drug discovery clinical trials success rate","AI designed drug FDA approval","Isomorphic Labs Eli Lilly Novartis partnership","AI drug discovery Phase II wall","binding affinity prediction FEP+","computational drug design","in silico drug discovery","AI drug Phase 2 success rate","generative AI drug discovery","Insilico Medicine rentosertib","Recursion Exscientia merger","Alphabet Other Bets Isomorphic Labs","undruggable targets AI","Demis Hassabis drug design","AI pharma deal structure biobucks","Chai Discovery Boltz AI biotech","IsoDDE benchmark performance","Phase II efficacy AI drugs"],"word_count":1119,"reading_time_minutes":6,"image":"https://static.philippdubach.com/ograph/ograph-isomorphic-labs-isodde1.jpg","summary":"IsoDDE doubles AlphaFold 3 on hard benchmarks and beats physics-based gold standards. But no AI drug has FDA approval. What $4B in pharma deals actually mean."},{"id":"a5a1a2705cd9b02ed93932955aa2c468","slug":"","title":"The Last Architecture Designed by Hand","url":"https://philippdubach.com/posts/the-last-architecture-designed-by-hand/","description":"The transformer's limits are now mathematical proofs, not empirical hunches. Hybrids are in production. AI is searching for its own replacement. Here's what comes after.","date_published":"2026-03-16","date_modified":"2026-03-16","type":"Analysis","categories":["AI","Tech"],"keywords":["what comes after transformers","transformer architecture limits","AI architecture 2026","Mamba vs transformer","hybrid AI architecture","post-transformer architecture","AlphaEvolve AI research","transformer quadratic scaling","diffusion language models","mixture of experts MoE","AI recursive self-improvement","LLM hallucination mathematical proof","DeepSeek V3 training cost","inference compute scaling","Jamba hybrid architecture","test-time compute scaling","state space models vs transformers","transformer replacement 2026"],"word_count":1249,"reading_time_minutes":6,"image":"https://static.philippdubach.com/ograph/ograph-last-architecture-designed-by-hand.jpg","summary":"The transformer's limits are now mathematical proofs, not empirical hunches. Hybrids are in production. AI is searching for its own replacement. Here's what comes after."},{"id":"f9d308102f58541605431b96acfb9951","slug":"","title":"MCP vs A2A in 2026: How the AI Protocol War Ends","url":"https://philippdubach.com/posts/mcp-vs-a2a-in-2026-how-the-ai-protocol-war-ends/","description":"MCP leads with 97M monthly SDK downloads and 10,000+ servers. A2A fills a different layer. Analysis of the agentic AI standards war with historical parallels.","date_published":"2026-03-15","date_modified":"2026-05-04","type":"Analysis","categories":["AI","Tech"],"keywords":["MCP vs A2A","MCP vs A2A 2026","model context protocol","agent to agent protocol","agentic AI protocol war","AI agent protocol","MCP A2A comparison","MCP protocol adoption","A2A protocol Google","Agentic AI Foundation","MCP security vulnerabilities","agentic AI market size","AI agent interoperability","enterprise AI architecture","MCP server ecosystem","multi-agent orchestration","AI standards war","model context protocol vs agent to agent protocol"],"word_count":1511,"reading_time_minutes":8,"image":"https://static.philippdubach.com/ograph/ograph-agentic-ai-standards-war1.jpg","summary":"MCP leads with 97M monthly SDK downloads and 10,000+ servers. A2A fills a different layer. Analysis of the agentic AI standards war with historical parallels."},{"id":"be0ad076a77cc8be035b12a448b82189","slug":"","title":"AI Models Are the New Rebar","url":"https://philippdubach.com/posts/ai-models-are-the-new-rebar/","description":"Qwen 3.5-35B runs on a gaming PC and matches Claude Sonnet 4.5. When the commodity version is 95% as good and 97% cheaper, you have a pricing problem.","date_published":"2026-03-11","date_modified":"2026-05-14","type":"Analysis","categories":["AI","Investing"],"keywords":["AI model commoditization","AI commoditization","open-source AI models","open source AI vs proprietary","AI inference cost decline","OpenAI valuation","AI pricing war 2026","AI margin compression","Qwen 3.5","LLM switching costs","AI model pricing","disruptive innovation AI","model layer commodity","who wins AI value chain","OpenAI profitability","open source AI performance gap","Christensen AI disruption","Anthropic valuation"],"word_count":1519,"reading_time_minutes":8,"image":"https://static.philippdubach.com/ograph/ograph-ai-margin-collapse1.jpg","summary":"Qwen 3.5-35B runs on a gaming PC and matches Claude Sonnet 4.5. When the commodity version is 95% as good and 97% cheaper, you have a pricing problem."},{"id":"f0c2b6ec830dbcae5c6ad688f2e2e18e","slug":"","title":"AI Capex Arms Race: Who Blinks First?","url":"https://philippdubach.com/posts/ai-capex-arms-race-who-blinks-first/","description":"Alphabet's free cash flow is on track to fall 90% in 2026. Amazon's is at $11B. $690B in AI capex is cannibalizing the cash that justified these valuations.","date_published":"2026-03-08","date_modified":"2026-05-04","type":"Analysis","categories":["Investing","AI"],"keywords":["AI capex 2026","hyperscaler capex 2026","AI infrastructure bubble","AI spending free cash flow","GPU depreciation stranded assets","AI capex vs revenue gap","AI data center spending 2026","telecom bubble AI comparison","Sequoia 600 billion revenue gap","inference cost decline","Dario Amodei AI spending risk","AI infrastructure overbuilt","GPU useful life depreciation risk","AI capex bubble","Big Tech AI free cash flow","hyperscaler AI investment sustainability","Jevons paradox AI compute"],"word_count":1549,"reading_time_minutes":8,"image":"https://static.philippdubach.com/ograph/ograph-ai-capex-arms-race1.jpg","summary":"Alphabet's free cash flow is on track to fall 90% in 2026. Amazon's is at $11B. $690B in AI capex is cannibalizing the cash that justified these valuations."},{"id":"805a93b7276b5fd47348053ba757b36b","slug":"","title":"93% of Developers Use AI Coding Tools. Productivity Hasn't Moved.","url":"https://philippdubach.com/posts/93-of-developers-use-ai-coding-tools.-productivity-hasnt-moved./","description":"METR found experienced developers 19% slower with AI, despite feeling 20% faster. At 92.6% adoption, organizational productivity gains remain roughly 10%.","date_published":"2026-03-04","date_modified":"2026-05-04","type":"Analysis","categories":["Tech"],"keywords":["AI coding productivity","AI developer productivity paradox","does AI coding improve productivity","METR AI developer study 2026","GitHub Copilot productivity ROI","AI coding tools productivity","developer productivity AI","AI code review bottleneck","AI generated code quality","DORA metrics AI adoption","Amdahl's law software development","AI pair programming ROI","software engineering productivity","AI coding tools enterprise ROI 2026","code generation productivity gap"],"word_count":1880,"reading_time_minutes":9,"image":"https://static.philippdubach.com/ograph/ograph-ai-coding-productivity1.jpg","summary":"METR found experienced developers 19% slower with AI, despite feeling 20% faster. At 92.6% adoption, organizational productivity gains remain roughly 10%."},{"id":"fa3fd171e2a76526d6b6a5bf0e5e67c7","slug":"peter-thiels-physics-department","title":"The Physics Department That Slowed Down","url":"https://philippdubach.com/posts/peter-thiels-physics-department/","description":"Peter Thiel says physics stalled in 1972. Then GPT-5.2 proved a new result in theoretical physics. The 75:1 AI compute gap between commerce and science.","date_published":"2026-03-02","date_modified":"2026-05-17","type":"Article","categories":["AI"],"keywords":["AI scientific discovery","AI for science","GPT-5.2 gluon discovery","Peter Thiel stagnation thesis","AI compute gap science","Great Stagnation AI","AI breakthrough physics 2026","total factor productivity decline","AlphaFold Nobel Prize 2024","federal AI R\u0026D spending","AI fusion energy plasma control","dual-use AI technology","DeepMind GNoME materials discovery","AI drug discovery clinical trials","scattering amplitudes theoretical physics","research productivity decline","computational irreducibility physics","bits vs atoms innovation","AI accelerated physics","scientific progress stagnation"],"word_count":1683,"reading_time_minutes":8,"image":"https://static.philippdubach.com/ograph/ograph-physics-department1.jpg","summary":"Peter Thiel says physics stalled in 1972. Then GPT-5.2 proved a new result in theoretical physics. The 75:1 AI compute gap between commerce and science."},{"id":"29b6ac90e5c3f097fddb9aaf1dfe55cf","slug":"","title":"Every Bulge Bracket Bank Agrees on AI","url":"https://philippdubach.com/posts/every-bulge-bracket-bank-agrees-on-ai/","description":"I read 12 AI research reports from Goldman Sachs, JPMorgan, UBS, and 6 other banks. Here's the consensus they're pushing, and what they're not saying.","date_published":"2026-03-01","date_modified":"2026-05-17","type":"Analysis","categories":["AI","Investing"],"keywords":["AI bubble 2026","bank AI research reports","AI capex productivity gap","second-order AI beneficiaries","Wall Street AI consensus","hyperscaler AI spending 2026","AI investment outlook 2026","sell-side AI research consensus","enterprise AI adoption rate","AI project failure rate","Goldman Sachs AI report","Morgan Stanley AI analysis","AI bubble vs dot-com","AI capex ROI","Jevons paradox AI energy","AI data centre power demand","AI general-purpose technology","nuclear energy data centres","AI infrastructure investment thesis","AI market concentration risk","hyperscaler capex ROI 2026","AI stock market outlook 2026","AI productivity paradox"],"word_count":2031,"reading_time_minutes":10,"image":"https://static.philippdubach.com/ograph/ograph-banks-ai-research1.jpg","summary":"I read 12 AI research reports from Goldman Sachs, JPMorgan, UBS, and 6 other banks. Here's the consensus they're pushing, and what they're not saying."},{"id":"1c2da81f392eae629397d680977002b7","slug":"","title":"When AI Labs Become Defense Contractors","url":"https://philippdubach.com/posts/when-ai-labs-become-defense-contractors/","description":"The Anthropic-Pentagon standoff isn't an ethics story. It's a replay of the 1993 Last Supper that consolidated 51 defense primes into 5, at Silicon Valley speed.","date_published":"2026-03-01","date_modified":"2026-05-17","type":"Analysis","categories":["AI","Economics"],"keywords":["AI defense contractors","Anthropic Pentagon ban","OpenAI Pentagon deal","Pentagon AI budget 2026","defense industry consolidation AI","AI military industrial complex","Palantir government revenue","IDIQ defense contracts","AI military contracts","security clearances AI companies","defense tech AI","DoD AI budget","AI weapons policy","defense revenue dependency","Last Supper defense industry","military AI governance","Pentagon AI strategy","government AI contracts","defense tech valuation","Silicon Valley defense contractors"],"word_count":1596,"reading_time_minutes":8,"image":"https://static.philippdubach.com/ograph/ograph-ai-labs-defense-contractors1.jpg","summary":"The Anthropic-Pentagon standoff isn't an ethics story. It's a replay of the 1993 Last Supper that consolidated 51 defense primes into 5, at Silicon Valley speed."},{"id":"5710e5506990622be274f3b6b540282e","slug":"","title":"People Live in Levels, Not Rates","url":"https://philippdubach.com/posts/people-live-in-levels-not-rates/","description":"Prices rose 25% since 2020 and won't come back. The levels-vs-rates problem explains the vibecession, the Stewart-Thaler debate, and why nobody trusts economists.","date_published":"2026-02-28","date_modified":"2026-05-17","type":"Commentary","categories":["Macro"],"keywords":["why economy feels bad","vibecession 2026","prices still high inflation down","Jon Stewart Richard Thaler debate 2026","K-shaped economy wages 2026","cumulative inflation since 2020","levels vs rates problem inflation","vibepression 2026","consumer sentiment disconnect","grocery prices since pandemic","housing affordability crisis 2026","economic perception gap","real wage growth by income level","nudge theory effectiveness meta-analysis","essentials squeeze affordability","economics communication failure","behavioral economics trust crisis","why economy feels bad when data is good"],"word_count":1386,"reading_time_minutes":7,"image":"https://static.philippdubach.com/ograph/ograph-stewart-thaler-debate.jpg","summary":"Prices rose 25% since 2020 and won't come back. The levels-vs-rates problem explains the vibecession, the Stewart-Thaler debate, and why nobody trusts economists."},{"id":"8a4be6a159c13879f140e10f078d3d93","slug":"","title":"Novo Was Europe's Most Valuable Company","url":"https://philippdubach.com/posts/novo-was-europes-most-valuable-company/","description":"Novo Nordisk lost 75% since June 2024. CagriSema failed vs Zepbound, US pricing is resetting lower, and Lilly leads on every axis. Full breakdown with numbers.","date_published":"2026-02-23","date_modified":"2026-05-17","type":"Analysis","categories":["Medicine"],"keywords":["Novo Nordisk stock crash 2026","NVO stock analysis February 2026","CagriSema REDEFINE 4 trial results","CagriSema vs Zepbound head to head","Novo Nordisk 2026 guidance revenue decline","Eli Lilly vs Novo Nordisk obesity drugs","semaglutide patent expiry international","Wegovy pill launch 2026","orforglipron FDA approval 2026","retatrutide Phase 3 weight loss","GLP-1 market pricing pressure MFN","Novo Nordisk valuation forward PE","obesity drug market 2030 forecast","tirzepatide revenue 2025","Novo Nordisk CEO Doustdar restructuring"],"word_count":1736,"reading_time_minutes":9,"image":"https://static.philippdubach.com/ograph/ograph-whats-wrong-at-novo.jpg","summary":"Novo Nordisk lost 75% since June 2024. CagriSema failed vs Zepbound, US pricing is resetting lower, and Lilly leads on every axis. Full breakdown with numbers."},{"id":"bc6f6521662825f96f010da1851ad5c5","slug":"","title":"The Absolute Insider Mess of Prediction Markets","url":"https://philippdubach.com/posts/the-absolute-insider-mess-of-prediction-markets/","description":"A Google insider made $1.15M on Polymarket in 24 hours. Israeli soldiers bet classified strike timing. Why prediction markets need insider trading regulation.","date_published":"2026-02-22","date_modified":"2026-05-17","type":"Analysis","categories":["Investing"],"keywords":["Polymarket insider trading","prediction market regulation","prediction market insider trading legal","Kalshi vs Polymarket","adverse selection prediction markets"],"word_count":2419,"reading_time_minutes":12,"image":"https://static.philippdubach.com/ograph/ograph-prediction-market-insider-trading.jpg","summary":"A Google insider made $1.15M on Polymarket in 24 hours. Israeli soldiers bet classified strike timing. Why prediction markets need insider trading regulation."},{"id":"5c03ba5dd1d6fe9e5dffe86b80ce1269","slug":"","title":"Economics of a Super Bowl Ad","url":"https://philippdubach.com/posts/economics-of-a-super-bowl-ad/","description":"A 30-second Super Bowl spot costs $8M. The real price is $16–23M. The ROI evidence is mixed. A deep look at the pricing, the prisoner's dilemma, and the NFL.","date_published":"2026-02-20","date_modified":"2026-05-04","type":"Analysis","categories":["Economics"],"keywords":["Super Bowl ad cost 2026","Super Bowl advertising ROI","do Super Bowl ads work","Super Bowl commercial cost breakdown","NFL Super Bowl advertising economics","Super Bowl prisoner dilemma game theory","Hartmann Klapper Stanford Super Bowl study","Super Bowl ad attribution problem","Super Bowl CPM analysis","Anheuser-Busch Super Bowl exclusivity","Super Bowl viewership 2026","Super Bowl ad production cost","Super Bowl ad effectiveness research","NFL advertising monopoly pricing","monoculture advertising Super Bowl","Super Bowl ad brand recall","Super Bowl streaming viewership Tubi","EDO Super Bowl engagement metric","Super Bowl 30-second spot price history","marketing attribution advertising measurement","game theory pricing economics","monopoly pricing power","advertising ROI measurement","prisoner's dilemma business strategy"],"word_count":2297,"reading_time_minutes":11,"image":"https://static.philippdubach.com/ograph/ograph-super-bowl-economics1.jpg","summary":"A 30-second Super Bowl spot costs $8M. The real price is $16–23M. The ROI evidence is mixed. A deep look at the pricing, the prisoner's dilemma, and the NFL."},{"id":"4a04b79c47a8bcca465126f31c05994e","slug":"","title":"The Impossible Backhand","url":"https://philippdubach.com/posts/the-impossible-backhand/","description":"AI converges to the mean by design. Ninth-power scaling costs and a 53-point gap on Humanity's Last Exam show domain expertise is appreciating, not declining.","date_published":"2026-02-17","date_modified":"2026-05-14","type":"Article","categories":["AI"],"keywords":["AI quality ceiling domain expertise","human AI collaboration centaur model","Humanity's Last Exam AI benchmark 2026","AI hallucination rates legal medical","AI deskilling risk workforce","AI convergence to mean RLHF typicality","AI scaling laws diminishing returns","centaur model AI augmentation vs replacement","jagged technological frontier Harvard BCG","model collapse Shumailov Nature 2024","ninth power compute scaling AI","AI overconfidence calibration error","Ben Affleck AI Joe Rogan quality","domain expert AI competitive advantage","AI skill erosion professional knowledge","Harvard BCG GPT-4 consultant study","AI video generation quality limits","Sara Hooker scaling laws compute cost","AI blind delegation risk centaur","enterprise AI strategy domain knowledge","AI output quality limits structural","human AI partnership knowledge workers","AI capability frontier jagged","knowledge worker productivity AI","professional AI risk hallucination","does AI replace or augment experts","AI quality plateau ninth power curve","why AI converges to average output"],"word_count":2114,"reading_time_minutes":10,"image":"https://static.philippdubach.com/ograph/ograph-ai-scaling-walls1.jpg","summary":"AI converges to the mean by design. Ninth-power scaling costs and a 53-point gap on Humanity's Last Exam show domain expertise is appreciating, not declining."},{"id":"752431af458a56a0719afe94b44056e0","slug":"","title":"Europe's $24 Trillion Payment Breakup Is Really a Bet on Infrastructure Arbitrage","url":"https://philippdubach.com/posts/europes-24-trillion-payment-breakup-is-really-a-bet-on-infrastructure-arbitrage/","description":"The EuroPA alliance connected 130 million users across 13 countries overnight. But this isn't really about sovereignty. It's an infrastructure arbitrage exploiting a 100-120bps spread between card network fees and SEPA Instant rails, accidentally protected by the EU's own regulation.","date_published":"2026-02-16","date_modified":"2026-05-04","type":"Article","categories":["Macro"],"keywords":["European Payments Initiative EPI","Wero payment system","Visa Mastercard European market","SEPA Instant Credit Transfer","payment sovereignty Europe","account-to-account payments A2A","EuroPA alliance Bizum Bancomat","interchange fee regulation EU","digital euro ECB","India UPI Brazil Pix comparison","European payment infrastructure","scheme fees Visa Mastercard"],"word_count":2785,"reading_time_minutes":14,"image":"https://static.philippdubach.com/ograph/ograph-european-payments-arbitrage3.jpg","summary":"The EuroPA alliance connected 130 million users across 13 countries overnight. But this isn't really about sovereignty. It's an infrastructure arbitrage exploiting a 100-120bps spread between card network fees and SEPA Instant rails, accidentally protected by the EU's own regulation."},{"id":"89eea0c6c52cebc39a1b389237c1d5d2","slug":"","title":"Long Volatility Premium","url":"https://philippdubach.com/posts/long-volatility-premium/","description":"One River's data shows beta-adjusted long volatility outperformed the S\u0026P 500 over 40 years. Goldman, AQR, and Universa agree on the mechanism but disagree on implementation. A synthesis of the evidence.","date_published":"2026-02-14","date_modified":"2026-05-04","type":"Article","categories":["Quantitative Finance"],"keywords":["long volatility premium","beta-adjusted tail hedging","volatility risk premium","tail risk hedging portfolio construction","variance tax compounding","put options convexity gamma","trend following vs put hedging","Universa Investments Spitznagel","rebalancing premium volatility","Goldman Sachs tail risk","AQR tortoise hare hedging","CAIA tail risk comparison"],"word_count":2497,"reading_time_minutes":12,"image":"https://static.philippdubach.com/ograph/ograph-long-volatility-premium3.jpg","summary":"One River's data shows beta-adjusted long volatility outperformed the S\u0026P 500 over 40 years. Goldman, AQR, and Universa agree on the mechanism but disagree on implementation. A synthesis of the evidence."},{"id":"6fd6c5b4b790884034a194c0d41f9e3e","slug":"","title":"The SaaSpocalypse Paradox","url":"https://philippdubach.com/posts/the-saaspocalypse-paradox/","description":"AI capex failure and AI replacing all software are mutually exclusive. Why the 2026 SaaSpocalypse is a $2 trillion pricing error, not an extinction event.","date_published":"2026-02-13","date_modified":"2026-05-14","type":"Article","categories":["AI","Investing"],"keywords":["SaaSpocalypse 2026","software sell-off 2026","IGV ETF crash 2026","AI disruption enterprise software","software vs semiconductor valuation 2026","SaaS crash 2026","software bear market 2026","BofA AI paradox","hyperscaler capex sustainability 2026","SaaS per-seat pricing AI agents","software valuation inversion semiconductors","Anthropic Claude Cowork Plugins sell-off","systems of record vs systems of engagement","software bifurcation AI risk","value-based pricing AI software","AI capex ROI 2026","SaaS pricing model transition","enterprise AI adoption 2026","Goldman Sachs AI agent TAM 2030","a16z white collar services AI expansion"],"word_count":1692,"reading_time_minutes":8,"image":"https://static.philippdubach.com/ograph/ograph-saaspocalypse-paradox3.jpg","summary":"AI capex failure and AI replacing all software are mutually exclusive. Why the 2026 SaaSpocalypse is a $2 trillion pricing error, not an extinction event."},{"id":"6f05d96fb5e21ca3e33a59c291d10afc","slug":"","title":"Don't Go Monolithic; The Agent Stack Is Stratifying","url":"https://philippdubach.com/posts/dont-go-monolithic-the-agent-stack-is-stratifying/","description":"The enterprise AI agent stack is stratifying into six layers with different winners at each. Models commoditize; context — your organizational world model — compounds. A framework for agentic AI architecture decisions.","date_published":"2026-02-10","date_modified":"2026-05-04","type":"Article","categories":["AI"],"keywords":["enterprise AI agent stack","agentic AI architecture enterprise","AI model commoditization","organizational world model","organizational context AI defensibility","multi-model AI strategy","enterprise AI context layer","agent orchestration","AI governance enterprise","context graph enterprise AI","multi-agent systems enterprise","AI agent stack defensibility"],"word_count":1512,"reading_time_minutes":8,"image":"https://static.philippdubach.com/ograph/ograph-enterprise-agent-stack1.jpg","summary":"The enterprise AI agent stack is stratifying into six layers with different winners at each. Models commoditize; context — your organizational world model — compounds. A framework for agentic AI architecture decisions."},{"id":"13160db50c50139ef513517eb3c623aa","slug":"","title":"Where Mobile Money Goes Now","url":"https://philippdubach.com/posts/where-mobile-money-goes-now/","description":"Apps overtook games in mobile IAP revenue for the first time in 2025, driven by $3.5B in GenAI growth. Analysis of Sensor Tower's State of Mobile 2026 report.","date_published":"2026-02-07","date_modified":"2026-03-15","type":"Analysis","categories":["AI"],"keywords":["state of mobile 2026","mobile app revenue vs games 2025","GenAI app revenue 2025","ChatGPT app revenue","AI app demographics","Sensor Tower state of mobile","mobile gaming revenue decline","app economy trends 2025","GenAI consumer spending","AI app adoption","YouTube most popular app all ages","mobile in-app purchase revenue","AI advertising demographics","Waymo ride volume 2025","mobile app subscription revenue","autonomous ride-hailing growth","AI referral traffic retail","cross-generational app usage","mobile monetization trends","consumer spending mobile apps"],"word_count":569,"reading_time_minutes":3,"image":"https://static.philippdubach.com/ograph/ograph-state-of-mobile.jpg","summary":"Apps overtook games in mobile IAP revenue for the first time in 2025, driven by $3.5B in GenAI growth. Analysis of Sensor Tower's State of Mobile 2026 report."},{"id":"fafece93a9800fb4e3034177dbde404a","slug":"","title":"Variance Tax","url":"https://philippdubach.com/posts/variance-tax/","description":"Variance drain is the hidden cost of volatility: why a portfolio averaging +10% can lose money. The ½σ² formula explains the gap between paper and real returns.","date_published":"2026-02-06","date_modified":"2026-05-04","type":"Essay","categories":["Quantitative Finance"],"keywords":["variance drain","volatility drag","geometric vs arithmetic returns","leveraged ETF decay","Kelly criterion"],"word_count":738,"reading_time_minutes":4,"image":"https://static.philippdubach.com/ograph/ograph-variance-tax.jpg","summary":"Variance drain is the hidden cost of volatility: why a portfolio averaging +10% can lose money. The ½σ² formula explains the gap between paper and real returns."},{"id":"5850e9f0b28d0c289f41444c6be4603f","slug":"","title":"Claude Opus 4.6: Anthropic's New Flagship AI Model for Agentic Coding","url":"https://philippdubach.com/posts/claude-opus-4.6-anthropics-new-flagship-ai-model-for-agentic-coding/","description":"Claude Opus 4.6 brings a 1M token context window, 68.8% ARC-AGI-2, and Agent Teams to Claude Code. Full benchmark comparison vs GPT-5.2 and Gemini 3 Pro with pricing analysis.","date_published":"2026-02-05","date_modified":"2026-05-17","type":"Commentary","categories":["AI"],"keywords":["Claude Opus 4.6","Anthropic Claude","LLM benchmarks comparison","agentic AI coding","1M context window LLM","Claude Code Agent Teams","GPT-5.2 vs Claude Opus","frontier AI model 2026","Terminal Bench 2.0","enterprise AI deployment"],"word_count":1167,"reading_time_minutes":6,"image":"https://static.philippdubach.com/ograph/ograph-opus46.jpg","summary":"Claude Opus 4.6 brings a 1M token context window, 68.8% ARC-AGI-2, and Agent Teams to Claude Code. Full benchmark comparison vs GPT-5.2 and Gemini 3 Pro with pricing analysis."},{"id":"c39218bc5914fffe1190c57aa9f7a68b","slug":"","title":"Buying the Haystack Might Not Work This Year","url":"https://philippdubach.com/posts/buying-the-haystack-might-not-work-this-year/","description":"a16z sees AI fundamentals thriving with 80% GPU utilization. AQR sees the CAPE at the 96th percentile. Both have data. Both may be right.","date_published":"2026-01-31","date_modified":"2026-05-04","type":"Commentary","categories":["AI","Investing"],"keywords":["AI bubble 2026","CAPE ratio stock market","AI valuation 2026","a16z state of markets","expected stock returns 2026"],"word_count":724,"reading_time_minutes":4,"image":"https://static.philippdubach.com/ograph/ograph-two-ways-ai.jpg","summary":"a16z sees AI fundamentals thriving with 80% GPU utilization. AQR sees the CAPE at the 96th percentile. Both have data. Both may be right."},{"id":"8cbdc313385abef0b701ee910766a1a5","slug":"","title":"Bandits and Agents: Netflix and Spotify Recommender Stacks in 2026","url":"https://philippdubach.com/posts/bandits-and-agents-netflix-and-spotify-recommender-stacks-in-2026/","description":"How hybrid recommender systems balance multi-armed bandits against LLM inference cost economics in 2026. A deep dive into Netflix recommendation algorithm architecture and Spotify's AI DJ recommender system.","date_published":"2026-01-30","date_modified":"2026-02-23","type":"Article","categories":["AI"],"keywords":["hybrid recommender systems 2026","multi-armed bandits recommender systems","LLM inference cost economics","Netflix recommendation algorithm architecture","Spotify AI DJ recommender system","recommender systems 2026","inference economics","contextual bandits","exploration exploitation tradeoff","Thompson Sampling","candidate generation ranking","personalization at scale"],"word_count":1038,"reading_time_minutes":5,"image":"https://static.philippdubach.com/ograph/ograph-recommender-architecture.jpg","summary":"How hybrid recommender systems balance multi-armed bandits against LLM inference cost economics in 2026. A deep dive into Netflix recommendation algorithm architecture and Spotify's AI DJ recommender system."},{"id":"494523162122511c05c0dbe8feafbbe5","slug":"","title":"Is Private Equity Just Beta With a Lockup?","url":"https://philippdubach.com/posts/is-private-equity-just-beta-with-a-lockup/","description":"AQR's 2026 data shows private equity returning 4.2% versus 3.9% for public equities. The 30bp illiquidity premium barely justifies years of lockup.","date_published":"2026-01-29","date_modified":"2026-05-14","type":"Commentary","categories":["Quantitative Finance"],"keywords":["illiquidity premium","private equity returns 2026","AQR capital market assumptions","PE vs public equity","venture capital dispersion"],"word_count":646,"reading_time_minutes":4,"image":"https://static.philippdubach.com/ograph/ograph-illiquidity-premium.jpg","summary":"AQR's 2026 data shows private equity returning 4.2% versus 3.9% for public equities. The 30bp illiquidity premium barely justifies years of lockup."},{"id":"061fd0e8364a1a0ef884da594bc9d916","slug":"","title":"Britain's Strategic Limbo","url":"https://philippdubach.com/posts/britains-strategic-limbo/","description":"Britain faces strategic isolation: locked out of EU defense cooperation, unwilling to join Trump's coalition. The mid-Atlantic bridge has nowhere to land.","date_published":"2026-01-28","date_modified":"2026-02-23","type":"Analysis","categories":["Macro"],"keywords":["UK EU defense cooperation","Brexit defense policy","SAFE fund exclusion","Britain strategic isolation","post-Brexit foreign policy","defense procurement economics","trade bloc exclusion","economic isolation"],"word_count":432,"reading_time_minutes":3,"image":"https://static.philippdubach.com/ograph/0016.png","summary":"Britain faces strategic isolation: locked out of EU defense cooperation, unwilling to join Trump's coalition. The mid-Atlantic bridge has nowhere to land."},{"id":"91924cde0386a2ff81652e6dd59ece17","slug":"","title":"The Rise of Middle Power Realism","url":"https://philippdubach.com/posts/the-rise-of-middle-power-realism/","description":"At Davos 2026, Carney told allies to take down the signs of the liberal order. Middle powers are learning to navigate between giants without illusions.","date_published":"2026-01-27","date_modified":"2026-05-04","type":"Analysis","categories":["Macro"],"keywords":["Mark Carney Davos 2026 speech","middle power foreign policy","Donroe Doctrine Trump","Canada EU SAFE defense","rules-based international order collapse","trade policy economics","tariff response strategy","economic coercion"],"word_count":813,"reading_time_minutes":4,"image":"https://static.philippdubach.com/ograph/0015.png","summary":"At Davos 2026, Carney told allies to take down the signs of the liberal order. Middle powers are learning to navigate between giants without illusions."},{"id":"e9eba6adf46865a7d367bd5412c7925b","slug":"","title":"The Most Expensive Assumption in AI","url":"https://philippdubach.com/posts/the-most-expensive-assumption-in-ai/","description":"Sara Hooker's research challenges the trillion-dollar scaling thesis. Compact models now outperform massive ones as diminishing returns hit AI.","date_published":"2026-01-26","date_modified":"2026-05-17","type":"Commentary","categories":["AI"],"keywords":["AI scaling laws","diminishing returns AI","small vs large language models","Sara Hooker scaling","AI infrastructure investing"],"word_count":711,"reading_time_minutes":4,"image":"https://static.philippdubach.com/ograph/ograph-scaling.jpg","summary":"Sara Hooker's research challenges the trillion-dollar scaling thesis. Compact models now outperform massive ones as diminishing returns hit AI."},{"id":"ae7d98fce969bb402c2b721d011d8a74","slug":"","title":"Against All Odds: The Mathematics of 'Provably Fair' Casino Games","url":"https://philippdubach.com/posts/against-all-odds-the-mathematics-of-provably-fair-casino-games/","description":"Statistical analysis of 20,000 crash game rounds verifies the 97% RTP claim. But 179 rounds per hour means expected losses exceed 500% of wagers hourly.","date_published":"2026-01-25","date_modified":"2026-05-17","type":"Project","categories":["Quantitative Finance"],"keywords":["provably fair gambling","crash game mathematics","RTP statistical testing","gambling house edge","crash game strategy"],"word_count":724,"reading_time_minutes":4,"doi":"10.2139/ssrn.6065213","image":"https://static.philippdubach.com/ograph/ograph-casino.jpg","summary":"Statistical analysis of 20,000 crash game rounds verifies the 97% RTP claim. But 179 rounds per hour means expected losses exceed 500% of wagers hourly."},{"id":"302468bbc6606ddf2a0121033be87975","slug":"","title":"Enterprise AI Strategy is Backwards","url":"https://philippdubach.com/posts/enterprise-ai-strategy-is-backwards/","description":"85% of AI projects fail. Only 26% translate pilots to production. The winners automate the coordination layer where employees spend 57% of their workday.","date_published":"2026-01-22","date_modified":"2026-02-23","type":"Essay","categories":["AI"],"keywords":["enterprise AI failure","AI implementation strategy","coordination layer AI","AI pilot to production","workplace AI productivity"],"word_count":413,"reading_time_minutes":2,"image":"https://static.philippdubach.com/ograph/ograph-ai-strat-backwards.jpg","summary":"85% of AI projects fail. Only 26% translate pilots to production. The winners automate the coordination layer where employees spend 57% of their workday."},{"id":"a406a34db0299576cf80689e4b91feb8","slug":"","title":"Big in Japan","url":"https://philippdubach.com/posts/big-in-japan/","description":"Japan holds $5 trillion in foreign assets. With 30-year JGB yields now above 3%, the carry trade that defined Japanese investing faces new friction.","date_published":"2026-01-19","date_modified":"2026-02-23","type":"Commentary","categories":["Macro"],"keywords":["yen carry trade","Japan Treasury holdings","JGB yields","Japanese capital repatriation","Japan foreign assets"],"word_count":344,"reading_time_minutes":2,"image":"https://static.philippdubach.com/ograph/0014.png","summary":"Japan holds $5 trillion in foreign assets. With 30-year JGB yields now above 3%, the carry trade that defined Japanese investing faces new friction."},{"id":"ee06e3b70c0ac82cf52fe341e8026851","slug":"","title":"Ozempic is Reshaping the Fast Food Industry","url":"https://philippdubach.com/posts/ozempic-is-reshaping-the-fast-food-industry/","description":"Cornell research: GLP-1 users cut grocery spending 5.3%, fast food 8%. With 16% household adoption and savory snacks down 10%, food stocks face headwinds.","date_published":"2026-01-16","date_modified":"2026-05-14","type":"Commentary","categories":["Medicine","Economics"],"keywords":["GLP-1 food industry impact","Ozempic grocery spending","food stocks GLP-1","Wegovy consumer behavior","weight loss drug food sales"],"word_count":780,"reading_time_minutes":4,"external_url":"https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5073929","image":"https://static.philippdubach.com/ograph/ograph-glp1-food.jpg","summary":"Cornell research: GLP-1 users cut grocery spending 5.3%, fast food 8%. With 16% household adoption and savory snacks down 10%, food stocks face headwinds."},{"id":"82f00ef4bc78430b2b32e7720e21dfa0","slug":"","title":"Does AI mean the demand on labor goes up?","url":"https://philippdubach.com/posts/does-ai-mean-the-demand-on-labor-goes-up/","description":"AI was supposed to free us. The Jevons paradox plays out in real time: efficiency expands workload, not leisure. 77% of workers say AI added to their work.","date_published":"2026-01-15","date_modified":"2026-02-23","type":"Commentary","categories":["AI","Economics"],"keywords":["AI productivity paradox","Jevons paradox AI","AI increasing workload","future of work AI","Parkinson's law AI"],"word_count":411,"reading_time_minutes":2,"image":"https://static.philippdubach.com/ograph/0005.png","summary":"AI was supposed to free us. The Jevons paradox plays out in real time: efficiency expands workload, not leisure. 77% of workers say AI added to their work."},{"id":"3bb502cf6a51ce548718d5d079c62b22","slug":"","title":"Repo might be even bigger than we thought","url":"https://philippdubach.com/posts/repo-might-be-even-bigger-than-we-thought/","description":"New OFR data reveals $12.6 trillion in daily repo exposures—$700 billion larger than previous estimates. The plumbing of modern money remains poorly understood.","date_published":"2026-01-13","date_modified":"2026-05-17","type":"Commentary","categories":["Macro"],"keywords":["repo market","shadow banking","Standing Repo Facility","money market liquidity","Zoltan Pozsar"],"word_count":963,"reading_time_minutes":5,"image":"https://static.philippdubach.com/ograph/0012.png","summary":"New OFR data reveals $12.6 trillion in daily repo exposures—$700 billion larger than previous estimates. The plumbing of modern money remains poorly understood."},{"id":"42a0f1b7e6efba7a9d487d59bc4dee1c","slug":"","title":"The Market Can Stay Irrational Longer Than You Can Stay Solvent","url":"https://philippdubach.com/posts/the-market-can-stay-irrational-longer-than-you-can-stay-solvent/","description":"Steve Eisman explains how U.S. equity markets have structurally decoupled from everyday economic reality through concentration and passive investing.","date_published":"2026-01-11","date_modified":"2026-05-14","type":"Commentary","categories":["Investing"],"keywords":["stock market decoupling economy","S\u0026P 500 concentration risk","passive investing market impact","Steve Eisman market thesis","K-shaped economy stocks"],"word_count":735,"reading_time_minutes":4,"image":"https://static.philippdubach.com/ograph/ograph-unmoored-market.jpg","summary":"Steve Eisman explains how U.S. equity markets have structurally decoupled from everyday economic reality through concentration and passive investing."},{"id":"3a393ebbfc57aab423113bea08894c04","slug":"","title":"Social Media Success Prediction: BERT Models for Post Titles","url":"https://philippdubach.com/posts/social-media-success-prediction-bert-models-for-post-titles/","description":"Training RoBERTa to predict Hacker News success revealed temporal leakage inflating metrics. How temporal splits, calibration, and regularization fix it.","date_published":"2026-01-10","date_modified":"2026-05-17","type":"Project","categories":["AI"],"keywords":["Hacker News prediction model","temporal train test split ML","RoBERTa text classification","model calibration isotonic regression","machine learning overfitting regularization"],"word_count":965,"reading_time_minutes":5,"image":"https://static.philippdubach.com/ograph/ograph-hn-predictor.jpg","summary":"Training RoBERTa to predict Hacker News success revealed temporal leakage inflating metrics. How temporal splits, calibration, and regularization fix it."},{"id":"a9c736437258f09f4bf35d5f3ec449e2","slug":"","title":"Beyond Vector Search: Why LLMs Need Episodic Memory","url":"https://philippdubach.com/posts/beyond-vector-search-why-llms-need-episodic-memory/","description":"Context windows aren't memory. Explore EM-LLM's episodic architecture, knowledge graph tools like Mem0 and Letta, and why vectors fail for sequential data.","date_published":"2026-01-09","date_modified":"2026-05-17","type":"Essay","categories":["AI"],"keywords":["LLM memory architecture","episodic memory AI","knowledge graph LLM memory","vector database limitations RAG","KV cache optimization transformer"],"word_count":571,"reading_time_minutes":3,"image":"https://static.philippdubach.com/ograph/ograph-llm-memory2.jpg","summary":"Context windows aren't memory. Explore EM-LLM's episodic architecture, knowledge graph tools like Mem0 and Letta, and why vectors fail for sequential data."},{"id":"ee58b27c23a6ee0f0392b1bcd557ece1","slug":"","title":"65% of Hacker News Posts Have Negative Sentiment, and They Outperform","url":"https://philippdubach.com/posts/65-of-hacker-news-posts-have-negative-sentiment-and-they-outperform/","description":"Sentiment analysis of 32,000 Hacker News posts shows 65% skew negative and earn 27% more points. Six transformer and LLM models tested, full data included.","date_published":"2026-01-07","date_modified":"2026-03-15","type":"Project","categories":["AI"],"keywords":["Hacker News sentiment analysis","negativity bias social media engagement","BERT RoBERTa sentiment comparison","attention inequality power law social media","Hacker News data analysis engagement","DistilBERT sentiment classification","transformer model NLP comparison","social media negativity bias research","Gini coefficient content distribution","tech community sentiment research","Hacker News score distribution","multi-model sentiment comparison","negative content engagement premium","power law online communities","NLP sentiment classification pipeline"],"word_count":338,"reading_time_minutes":2,"image":"https://static.philippdubach.com/ograph/ograph-hn-sentiment.jpg","summary":"Sentiment analysis of 32,000 Hacker News posts shows 65% skew negative and earn 27% more points. Six transformer and LLM models tested, full data included."},{"id":"0925fa8935f282eb83532fb1f43b1e21","slug":"","title":"Praise by Name, Criticize by Category: Warren Buffett Retires at 95","url":"https://philippdubach.com/posts/praise-by-name-criticize-by-category-warren-buffett-retires-at-95/","description":"Buffett exits after paying $26.8B in taxes. What 60 years of letters reveal about admitting mistakes, insurance float, and why Abel inherits $300B in cash.","date_published":"2026-01-06","date_modified":"2026-05-14","type":"Essay","categories":["Investing"],"keywords":["Warren Buffett retirement 2026","Berkshire Hathaway insurance float","Buffett shareholder letters lessons","Greg Abel Berkshire CEO","Berkshire capital allocation strategy"],"word_count":889,"reading_time_minutes":5,"image":"https://static.philippdubach.com/ograph/ograph-buffet-retirement.jpg","summary":"Buffett exits after paying $26.8B in taxes. What 60 years of letters reveal about admitting mistakes, insurance float, and why Abel inherits $300B in cash."},{"id":"dcfb1cfe00867f0c10aad4b665ff6be0","slug":"","title":"RSS Swipr: Find Blogs Like You Find Your Dates","url":"https://philippdubach.com/posts/rss-swipr-find-blogs-like-you-find-your-dates/","description":"Build an open-source ML RSS reader with swipe interface. Uses MPNet embeddings and Thompson sampling for personalized feeds that escape the filter bubble.","date_published":"2026-01-05","date_modified":"2026-05-17","type":"Project","categories":["AI"],"keywords":["machine learning RSS reader","Thompson sampling recommendation","MPNet sentence embeddings","personalized RSS Python open source","filter bubble RSS alternative"],"word_count":756,"reading_time_minutes":4,"image":"https://static.philippdubach.com/ograph/ograph-rss-swipr.jpg","summary":"Build an open-source ML RSS reader with swipe interface. Uses MPNet embeddings and Thompson sampling for personalized feeds that escape the filter bubble."},{"id":"6394991463263d80ea6ce3b2c17794d2","slug":"","title":"Apple's AI Bet: Playing the Long Game or Missing the Moment?","url":"https://philippdubach.com/posts/apples-ai-bet-playing-the-long-game-or-missing-the-moment/","description":"Apple's $157B cash pile and Gemini-powered Siri shift show a restrained AI strategy. Analysis of whether Apple wins as AI models become commodities.","date_published":"2025-12-30","date_modified":"2026-05-17","type":"Commentary","categories":["AI"],"keywords":["Apple AI strategy","Apple Siri Gemini update","AI model commoditization","Big Tech AI spending","Apple Intelligence vs ChatGPT"],"word_count":507,"reading_time_minutes":3,"external_url":"https://www.theinformation.com/articles/2026-predictions-apple-will-reverse-ai-slump","image":"https://static.philippdubach.com/ograph/ograph-apple-ai-strategy.jpg","summary":"Apple's $157B cash pile and Gemini-powered Siri shift show a restrained AI strategy. Analysis of whether Apple wins as AI models become commodities."},{"id":"a838a7384a8bc92538ecfccccfae9630","slug":"","title":"Building a No-Tracking Newsletter from Markdown to Distribution","url":"https://philippdubach.com/posts/building-a-no-tracking-newsletter-from-markdown-to-distribution/","description":"Build a privacy-focused newsletter with Python, Cloudflare Workers KV, and Resend API. Zero tracking, zero cost, full control. Open source code included.","date_published":"2025-12-24","date_modified":"2026-05-17","type":"Project","categories":["Tech"],"keywords":["self-hosted newsletter","Cloudflare Workers newsletter","no tracking newsletter email","Resend API email tutorial","markdown to HTML email Python"],"word_count":798,"reading_time_minutes":4,"image":"https://static.philippdubach.com/ograph/ograph-newsletter-setup.jpg","summary":"Build a privacy-focused newsletter with Python, Cloudflare Workers KV, and Resend API. Zero tracking, zero cost, full control. Open source code included."},{"id":"682af39a3f8e44245c73c3c794c5df38","slug":"","title":"How AI is Shaping My Investment Portfolio for 2026","url":"https://philippdubach.com/posts/how-ai-is-shaping-my-investment-portfolio-for-2026/","description":"Rebalancing for 2026: reducing S\u0026P 500 at 40× CAPE, adding Europe after Germany's €1T pivot, and bonds at 4.2% yields. Full allocation rationale.","date_published":"2025-12-12","date_modified":"2026-05-14","type":"Essay","categories":["Investing"],"keywords":["2026 portfolio allocation strategy","S\u0026P 500 concentration risk 2026","dollar depreciation investment hedge","Germany infrastructure investment 2026","fixed income allocation 2026"],"word_count":2065,"reading_time_minutes":10,"image":"https://static.philippdubach.com/ograph/ograph-2026portfolio.jpg","summary":"Rebalancing for 2026: reducing S\u0026P 500 at 40× CAPE, adding Europe after Germany's €1T pivot, and bonds at 4.2% yields. Full allocation rationale."},{"id":"48d897eac8cdd045529a350cc8abb70c","slug":"","title":"Not Logan Roy: Netflix vs. Paramount's Bidding War","url":"https://philippdubach.com/posts/not-logan-roy-netflix-vs.-paramounts-bidding-war/","description":"Netflix's $72B Warner Bros deal vs Paramount's hostile $30/share tender. Deal mechanics, aggregation theory, and why internet distributors win streaming.","date_published":"2025-12-09","date_modified":"2026-05-14","type":"Analysis","categories":["Investing"],"keywords":["Netflix Warner Bros acquisition","Paramount hostile takeover Warner Bros","streaming consolidation 2025","media M\u0026A deal analysis","aggregation theory streaming","hostile tender offer mechanics","breakup fee economics","platform economics aggregation theory","M\u0026A valuation analysis"],"word_count":1461,"reading_time_minutes":7,"image":"https://static.philippdubach.com/ograph/ograph-warner-bidding.jpg","summary":"Netflix's $72B Warner Bros deal vs Paramount's hostile $30/share tender. Deal mechanics, aggregation theory, and why internet distributors win streaming."},{"id":"1a58e78f18e2660c031caeb898c58859","slug":"","title":"Nike's Crisis and the Economics of Brand Decay","url":"https://philippdubach.com/posts/nikes-crisis-and-the-economics-of-brand-decay/","description":"Nike lost $28B by weakening product development, athlete partnerships, and marketing simultaneously. Data-driven analysis of how complementary assets collapse.","date_published":"2025-12-02","date_modified":"2026-05-14","type":"Analysis","categories":["Investing"],"keywords":["Nike crisis 2025","why is Nike struggling","Nike brand decline analysis","Nike direct-to-consumer strategy failure","Nike vs On Hoka competition","John Donahoe Nike strategy","Nike stock decline","Nike turnaround strategy","complementary assets competitive advantage","athletic footwear market share","Nike marketing strategy failure","Nike supply chain risk","Nike revenue drop 2025","Nike innovation decline","brand erosion economics","Nike athlete endorsement losses","Nike Elliott Hill turnaround","disruption economics","Nike wholesale strategy","Nike tariff impact 2025"],"word_count":1573,"reading_time_minutes":8,"image":"https://static.philippdubach.com/ograph/ograph-nike-crisis.jpg","summary":"Nike lost $28B by weakening product development, athlete partnerships, and marketing simultaneously. Data-driven analysis of how complementary assets collapse."},{"id":"e32e53d07df53d14c6e89df7a3fa72fa","slug":"","title":"Michael Burry's $379 Newsletter","url":"https://philippdubach.com/posts/michael-burrys-379-newsletter/","description":"Michael Burry launches Substack warning AI markets mirror 1999. His Nvidia-Cisco comparison, the GPU depreciation debate, and what hyperscalers need to justify capex.","date_published":"2025-11-28","date_modified":"2026-05-14","type":"Commentary","categories":["Investing"],"keywords":["Michael Burry Substack newsletter","Michael Burry AI bubble warning","Nvidia Cisco comparison dot-com","GPU depreciation accounting AI","AI infrastructure spending bubble"],"word_count":964,"reading_time_minutes":5,"external_url":"https://michaeljburry.substack.com","image":"https://static.philippdubach.com/ograph/ograph-burry-newsletter.jpg","summary":"Michael Burry launches Substack warning AI markets mirror 1999. His Nvidia-Cisco comparison, the GPU depreciation debate, and what hyperscalers need to justify capex."},{"id":"0749daec1b374cbc8399c5892e3f080a","slug":"","title":"Is AI Really Eating the World? AGI, Networks, Value [2/2]","url":"https://philippdubach.com/posts/is-ai-really-eating-the-world-agi-networks-value-2/2/","description":"AGI predictions miss the point. Multiple competing models means price war. Value flows to applications, customer relationships, and vertical integrators.","date_published":"2025-11-24","date_modified":"2026-05-04","type":"Essay","categories":["AI"],"keywords":["AI value chain","AGI timeline predictions","AI model commoditization","AI competitive advantage","LLM data moats"],"word_count":1469,"reading_time_minutes":7,"external_url":"https://www.ben-evans.com/presentations/","image":"https://static.philippdubach.com/ograph/ograph-eatingtheworld2.jpg","summary":"AGI predictions miss the point. Multiple competing models means price war. Value flows to applications, customer relationships, and vertical integrators."},{"id":"cdd4c94e7ce4168dbd211ad7e694e901","slug":"","title":"Is AI Really Eating the World? [1/2]","url":"https://philippdubach.com/posts/is-ai-really-eating-the-world-1/2/","description":"Hyperscalers spend $400B on AI, API prices drop 97%, and DeepSeek builds frontier models for $500M. Value is flowing to applications, not model providers.","date_published":"2025-11-23","date_modified":"2026-05-04","type":"Analysis","categories":["AI"],"keywords":["AI model commoditization","generative AI platform shift","AI value chain where value accrues","hyperscaler AI capex spending","enterprise AI adoption rate","LLM pricing decline","frontier AI model cost","AI infrastructure investment","open-source AI models","DeepSeek frontier model","AI competitive moat","technology platform cycle","model layer vs application layer","AI consulting revenue","AI capex sustainability","enterprise AI deployment","labor-augmenting AI","AI integration services","generative AI investment thesis","LLM API cost reduction"],"word_count":977,"reading_time_minutes":5,"external_url":"https://www.ben-evans.com/presentations/","image":"https://static.philippdubach.com/ograph/ograph-eatingtheworld1.jpg","summary":"Hyperscalers spend $400B on AI, API prices drop 97%, and DeepSeek builds frontier models for $500M. Value is flowing to applications, not model providers."},{"id":"2dcea7b17bffd512e4040bfb1c3b1ee8","slug":"","title":"Weather Forecasts Have Improved a Lot","url":"https://philippdubach.com/posts/weather-forecasts-have-improved-a-lot/","description":"Four-day forecasts now match one-day accuracy from 30 years ago. How AI models like WeatherNext 2 use CRPS training to preserve extreme weather signals.","date_published":"2025-11-22","date_modified":"2026-02-22","type":"Analysis","categories":["AI"],"keywords":["AI weather forecasting accuracy","WeatherNext 2 Google DeepMind","weather forecast improvement history","neural weather prediction models","CRPS loss weather forecasting","AI inference cost economics","AI replacing traditional systems","AI disruption real-world applications"],"word_count":296,"reading_time_minutes":2,"external_url":"https://ourworldindata.org/weather-forecasts","image":"https://static.philippdubach.com/ograph/ograph-weather.jpg","summary":"Four-day forecasts now match one-day accuracy from 30 years ago. How AI models like WeatherNext 2 use CRPS training to preserve extreme weather signals."},{"id":"af291f8e711d37a961856a99accb1a2d","slug":"","title":"GLP-1 Receptor Agonists in ASUD Treatment","url":"https://philippdubach.com/posts/glp-1-receptor-agonists-in-asud-treatment/","description":"A phase 2 RCT shows low-dose semaglutide reduces alcohol craving and heavy drinking with effect sizes exceeding naltrexone. What GLP-1 means for AUD treatment.","date_published":"2025-11-21","date_modified":"2026-05-14","type":"Commentary","categories":["Medicine"],"keywords":["semaglutide alcohol use disorder","GLP-1 agonists addiction treatment","Ozempic alcoholism research","semaglutide vs naltrexone","new medications alcohol use disorder","GLP-1 alcohol craving","semaglutide AUD clinical trial","Ozempic reduce drinking","alcohol use disorder pharmacotherapy","GLP-1 receptor agonist reward pathway","semaglutide heavy drinking reduction","Wegovy alcohol","naltrexone vs semaglutide effect size","substance use disorder GLP-1","alcohol addiction medication 2025","dopamine signaling alcohol craving","pemvidutide alcohol use disorder","acamprosate disulfiram alternatives","off-label semaglutide alcohol","phase 2 trial semaglutide AUD"],"word_count":398,"reading_time_minutes":2,"external_url":"https://academic.oup.com/jes/article/9/11/bvaf141/8277723","image":"https://static.philippdubach.com/ograph/ograph-glp1addiction.jpg","summary":"A phase 2 RCT shows low-dose semaglutide reduces alcohol craving and heavy drinking with effect sizes exceeding naltrexone. What GLP-1 means for AUD treatment."},{"id":"514f051a31a0116bb25bfb3c6bc948a5","slug":"","title":"AI Models as Standalone P\u0026Ls","url":"https://philippdubach.com/posts/ai-models-as-standalone-pls/","description":"OpenAI lost $11.5B in one quarter. But Anthropic CEO Dario Amodei argues each AI model is independently profitable. Here's why the math is complicated.","date_published":"2025-11-09","date_modified":"2026-05-14","type":"Commentary","categories":["AI"],"keywords":["OpenAI losses","AI model profitability","AI training costs","OpenAI business model","Dario Amodei profitability"],"word_count":878,"reading_time_minutes":5,"external_url":"https://www.theregister.com/2025/10/29/microsoft_earnings_q1_26_openai_loss/","image":"https://static.philippdubach.com/ograph/ograph-aipnl.jpg","summary":"OpenAI lost $11.5B in one quarter. But Anthropic CEO Dario Amodei argues each AI model is independently profitable. Here's why the math is complicated."},{"id":"75d9adf17d4ca39a39979623785444b0","slug":"","title":"Working with Models","url":"https://philippdubach.com/posts/working-with-models/","description":"Diffusion models corrupt data into noise, then reverse the process. Learn the math with Stefano Ermon's Stanford CS236 course, free on YouTube.","date_published":"2025-11-08","date_modified":"2026-02-22","type":"Commentary","categories":["AI"],"keywords":["diffusion models explained","Stanford CS236 course","deep generative models tutorial","diffusion model forward process","Stefano Ermon lectures","generative AI architecture","AI model training foundations"],"word_count":126,"reading_time_minutes":1,"external_url":"https://arxiv.org/abs/2510.21890","image":"https://static.philippdubach.com/ograph/ograph-models.jpg","summary":"Diffusion models corrupt data into noise, then reverse the process. Learn the math with Stefano Ermon's Stanford CS236 course, free on YouTube."},{"id":"1f8f388ff0c106a7944014d90c6d1e42","slug":"","title":"Pozsar's Bretton Woods III: Three Years Later [2/2]","url":"https://philippdubach.com/posts/pozsars-bretton-woods-iii-three-years-later-2/2/","description":"Gold above $4,000, Treasury holdings below $7T, but the dollar still dominates 88% of FX volumes. What Pozsar's Bretton Woods III got right and wrong.","date_published":"2025-10-26","date_modified":"2026-05-17","type":"Analysis","categories":["Macro"],"keywords":["Bretton Woods III","de-dollarization","central bank gold reserves","reserve currency diversification","commodity collateral","Pozsar inside money outside money","dollar reserve share decline","central bank gold buying","geopolitical risk reserves","FRA-OIS spread","cross-currency basis swaps","renminbi internationalization","sanctions risk dollar","commodity volatility inflation","supply-driven inflation monetary policy","dollar dominance","Treasury foreign ownership","gold as reserve asset","monetary regime change","BRICS reserve currency"],"word_count":1180,"reading_time_minutes":6,"external_url":"https://exunoplures.hu/public/pdf/a-decade-on-money-39.pdf","image":"https://static.philippdubach.com/ograph/ograph-Pozsar2.jpg","summary":"Gold above $4,000, Treasury holdings below $7T, but the dollar still dominates 88% of FX volumes. What Pozsar's Bretton Woods III got right and wrong."},{"id":"3d4237189547c423927ed1b81e9cb1ee","slug":"","title":"Pozsar's Bretton Woods III: The Framework [1/2]","url":"https://philippdubach.com/posts/pozsars-bretton-woods-iii-the-framework-1/2/","description":"How freezing Russian reserves sparked Bretton Woods III: Pozsar's framework on inside money, outside money, and the shift to a commodity-backed monetary order.","date_published":"2025-10-25","date_modified":"2026-05-04","type":"Analysis","categories":["Macro"],"keywords":["Bretton Woods III","Zoltan Pozsar","inside money outside money","money view Perry Mehrling","dollar reserve system","de-dollarization","reserve currency sanctions risk","commodity backed reserves","four prices of money","confiscation risk sovereign reserves","global monetary order","dollar funding non-US banks","commodity financing","central bank gold reserves","quantitative tightening commodity markets","covered interest parity","cross-currency basis swaps","shadow banking system"],"word_count":1345,"reading_time_minutes":7,"external_url":"https://exunoplures.hu/public/pdf/a-decade-on-money-39.pdf","image":"https://static.philippdubach.com/ograph/ograph-Pozsar1.jpg","summary":"How freezing Russian reserves sparked Bretton Woods III: Pozsar's framework on inside money, outside money, and the shift to a commodity-backed monetary order."},{"id":"184d779a6aed36bd4a4cb7252f1d92cb","slug":"","title":"Everything is a DCF Model","url":"https://philippdubach.com/posts/everything-is-a-dcf-model/","description":"Michael Mauboussin's argument that every cash-generating asset is valued through a DCF model. Why this Morgan Stanley paper changed how I think about value.","date_published":"2025-10-19","date_modified":"2026-05-04","type":"Commentary","categories":["Investing"],"keywords":["DCF model","discounted cash flow valuation","intrinsic value investing","Mauboussin DCF","asset valuation methods"],"word_count":137,"reading_time_minutes":1,"external_url":"https://www.morganstanley.com/content/dam/im/assets/publication/thought-leadership/consilient-observer/article_everythingisadcfmodel_us.pdf","image":"https://static.philippdubach.com/ograph/ograph-dcf.jpg","summary":"Michael Mauboussin's argument that every cash-generating asset is valued through a DCF model. Why this Morgan Stanley paper changed how I think about value."},{"id":"3f513cad8c517568a86b9fa07215b07d","slug":"","title":"Agent-based Systems for Modeling Wealth Distribution","url":"https://philippdubach.com/posts/agent-based-systems-for-modeling-wealth-distribution/","description":"Agent-based modeling shows how random market transactions naturally produce extreme wealth concentration, and why even a small wealth tax changes everything.","date_published":"2025-08-30","date_modified":"2026-05-17","type":"Analysis","categories":["Economics"],"keywords":["wealth distribution model","agent-based wealth simulation","wealth tax effect on inequality","Pareto distribution wealth","affine wealth model","wealth tax macroeconomics","inequality economics modeling","fiscal policy simulation"],"word_count":409,"reading_time_minutes":2,"external_url":"https://arxiv.org/abs/1604.02370","image":"https://static.philippdubach.com/ograph/ograph-distribution.jpg","summary":"Agent-based modeling shows how random market transactions naturally produce extreme wealth concentration, and why even a small wealth tax changes everything."},{"id":"583ba57c798caf9ceae969d73942531f","slug":"","title":"Visualizing Gradients with PyTorch","url":"https://philippdubach.com/posts/visualizing-gradients-with-pytorch/","description":"Build the right mental model for gradients with this PyTorch visualization tool. 2D surface plots with gradient vectors show the direction of steepest ascent.","date_published":"2025-08-23","date_modified":"2026-04-29","type":"Project","categories":["Tech"],"keywords":["pytorch gradient visualization","visualize gradients python","gradient descent visualization","gradient vectors machine learning","pytorch autograd tutorial"],"word_count":167,"reading_time_minutes":1,"image":"https://static.philippdubach.com/ograph/ograph-gradients.jpg","summary":"Build the right mental model for gradients with this PyTorch visualization tool. 2D surface plots with gradient vectors show the direction of steepest ascent."},{"id":"82da2a7b8da6c55c38f91c51e3346522","slug":"","title":"Counting Cards with Computer Vision","url":"https://philippdubach.com/posts/counting-cards-with-computer-vision/","description":"How I trained a YOLOv11 model to detect playing cards at 99.5% accuracy and built a real-time Monte Carlo blackjack odds calculator using computer vision.","date_published":"2025-07-06","date_modified":"2026-05-17","type":"Project","categories":["Tech"],"keywords":["blackjack card detection computer vision","YOLOv11 custom object detection training","real-time playing card recognition","train YOLO model playing cards","Monte Carlo blackjack odds Python"],"word_count":1091,"reading_time_minutes":6,"image":"https://static.philippdubach.com/ograph/ograph-blackjack.jpg","summary":"How I trained a YOLOv11 model to detect playing cards at 99.5% accuracy and built a real-time Monte Carlo blackjack odds calculator using computer vision."},{"id":"3fb7af0a5c206e8d0b6ca48ec3a2a7f1","slug":"","title":"Novo Nordisk's Post-Patent Strategy","url":"https://philippdubach.com/posts/novo-nordisks-post-patent-strategy/","description":"Novo Nordisk's lead replacement candidate amycretin reported Phase 1 data in The Lancet. How the company is positioning around the 2031 Ozempic patent cliff.","date_published":"2025-06-29","date_modified":"2026-05-14","type":"Commentary","categories":["Medicine"],"keywords":["amycretin Phase 1","Novo Nordisk pipeline","Ozempic patent expiration","GLP-1 capacity moat","Novo Nordisk strategy"],"word_count":711,"reading_time_minutes":4,"external_url":"https://www.thelancet.com/journals/lancet/article/PIIS0140-6736%2825%2901185-7/","image":"https://static.philippdubach.com/ograph/ograph-post.jpg","summary":"Novo Nordisk's lead replacement candidate amycretin reported Phase 1 data in The Lancet. How the company is positioning around the 2031 Ozempic patent cliff."},{"id":"20814dab2e4a5f8dd16c3aaa00780365","slug":"","title":"Modeling Glycemic Response with XGBoost","url":"https://philippdubach.com/posts/modeling-glycemic-response-with-xgboost/","description":"A hands-on project predicting postprandial glucose curves with XGBoost, Gaussian curve fitting, and 27 engineered features from CGM data. Code on GitHub.","date_published":"2025-05-30","date_modified":"2026-05-17","type":"Project","categories":["Tech"],"keywords":["predict glycemic response machine learning","XGBoost glucose prediction","postprandial glucose response prediction","CGM data machine learning","precision nutrition glucose prediction"],"word_count":1103,"reading_time_minutes":6,"doi":"10.2139/ssrn.5914902","image":"https://static.philippdubach.com/ograph/ograph-xgboost.jpg","summary":"A hands-on project predicting postprandial glucose curves with XGBoost, Gaussian curve fitting, and 27 engineered features from CGM data. Code on GitHub."},{"id":"41803dd6d3c77b2b30bf85defc33833c","slug":"","title":"Trading on Market Sentiment","url":"https://philippdubach.com/posts/trading-on-market-sentiment/","description":"GPT-3.5 matched RavenPack's 41% returns in a sentiment analysis trading strategy using 2,072 news headlines. See the full backtest results and comparison.","date_published":"2025-02-20","date_modified":"2026-05-17","type":"Project","categories":["AI"],"keywords":["sentiment analysis trading strategy","GPT sentiment analysis stock market","LLM financial sentiment analysis","news headline sentiment trading","RavenPack vs GPT sentiment"],"word_count":1101,"reading_time_minutes":6,"image":"https://static.philippdubach.com/ograph/ograph-sentiment.jpg","summary":"GPT-3.5 matched RavenPack's 41% returns in a sentiment analysis trading strategy using 2,072 news headlines. See the full backtest results and comparison."},{"id":"fb3c39410cbde799d182e578cad28d61","slug":"","title":"I Built a CGM Data Reader","url":"https://philippdubach.com/posts/i-built-a-cgm-data-reader/","description":"Built a CGM data analysis tool with Python to visualize Freestyle Libre 3 glucose data alongside nutrition, workouts, and sleep for endurance cycling.","date_published":"2025-01-02","date_modified":"2026-05-04","type":"Project","categories":["Tech"],"keywords":["CGM data analysis Python","Freestyle Libre 3 data export","continuous glucose monitor dashboard","CGM for endurance athletes","glucose tracking Python visualization"],"word_count":528,"reading_time_minutes":3,"image":"https://static.philippdubach.com/ograph/ograph-cgm.jpg","summary":"Built a CGM data analysis tool with Python to visualize Freestyle Libre 3 glucose data alongside nutrition, workouts, and sleep for endurance cycling."},{"id":"0c7519e7327b0c079a64b8c892f93f14","slug":"","title":"Crypto Mean Reversion Trading","url":"https://philippdubach.com/posts/crypto-mean-reversion-trading/","description":"How I built a crypto mean reversion trading bot using PELT change point detection on Kraken, targeting altcoin price overreactions with automated execution.","date_published":"2024-11-11","date_modified":"2026-05-04","type":"Project","categories":["Investing"],"keywords":["crypto mean reversion trading","PELT algorithm","change point detection","cryptocurrency market inefficiency","automated crypto trading bot","efficient market hypothesis crypto","statistical arbitrage","volatility trading strategy","quantitative trading"],"word_count":558,"reading_time_minutes":3,"image":"https://static.philippdubach.com/ograph/ograph-crypto.jpg","summary":"How I built a crypto mean reversion trading bot using PELT change point detection on Kraken, targeting altcoin price overreactions with automated execution."},{"id":"503768ab0e2c29eaa7148c87f9b100b3","slug":"","title":"My First 'Optimal' Portfolio","url":"https://philippdubach.com/posts/my-first-optimal-portfolio/","description":"How I built Python portfolio optimization tools, tripled the Sharpe ratio from 0.65 to 1.68, and published the results as an academic paper on MPT.","date_published":"2024-03-15","date_modified":"2026-05-17","type":"Project","categories":["Investing"],"keywords":["Python portfolio optimization","efficient frontier","Modern Portfolio Theory","Sharpe ratio","out-of-sample testing"],"word_count":395,"reading_time_minutes":2,"doi":"10.2139/ssrn.5915004","image":"https://static.philippdubach.com/ograph/ograph-portfolio.jpg","summary":"How I built Python portfolio optimization tools, tripled the Sharpe ratio from 0.65 to 1.68, and published the results as an academic paper on MPT."},{"id":"6850597e84d98010cc4d8a158027b73b","slug":"","title":"The Tech behind this Site","url":"https://philippdubach.com/posts/the-tech-behind-this-site/","description":"A Hugo blog tech stack with Cloudflare R2 image hosting, responsive WebP shortcodes, Workers AI social automation, and GitHub Pages CI/CD deployment.","date_published":"2024-01-15","date_modified":"2026-05-17","type":"Project","categories":["Tech"],"keywords":["Hugo blog tech stack","Hugo GitHub Pages Cloudflare","Hugo responsive images shortcode","Hugo Cloudflare R2 images","Hugo site automation Cloudflare Workers"],"word_count":1446,"reading_time_minutes":7,"image":"https://static.philippdubach.com/ograph/ograph-projects.jpg","summary":"A Hugo blog tech stack with Cloudflare R2 image hosting, responsive WebP shortcodes, Workers AI social automation, and GitHub Pages CI/CD deployment."}]}