llms.txt
Compare with other sites' llms.txt files →# Philipp D. Dubach
> Independent research and long-form analysis on quantitative finance, artificial intelligence, macroeconomics, and machine-learning systems.
Canonical site: https://philippdubach.com/. Author: Philipp D. Dubach. 112 articles. Last substantive update: 2026-11-07. Every link below points to a clean Markdown representation; the corresponding canonical HTML URL is declared in its HTTP Link header.
## Main pages
- [Home](https://philippdubach.com/index.md): Latest writing and projects.
- [About](https://philippdubach.com/about/index.md): Author background, research profiles, and site purpose.
- [Research](https://philippdubach.com/research/index.md): Papers, preprints, identifiers, and publication links.
- [Projects](https://philippdubach.com/projects/index.md): Open-source quantitative-finance, machine-learning, and computer-vision builds.
- [Writing](https://philippdubach.com/writing/index.md): Complete chronological article index.
## Articles
- [I Just Deleted All My Agent Skills](https://philippdubach.com/posts/i-just-deleted-all-my-agent-skills/index.md): Do agent skills actually work? Paired benchmarks of public skills show little gain and higher token costs, so I deleted mine. See where the data disagrees.
- [Mario Draghi on Debt, Growth and Europe’s Future](https://philippdubach.com/posts/mario-draghi-europe-growth-debt/index.md): Mario Draghi connects European growth, government debt and central bank independence. His lecture and Q&A reveal the economic and political choices ahead.
- [Does working less explain Europe’s GDP per capita gap with America?](https://philippdubach.com/posts/us-europe-productivity-working-hours/index.md): Does working less explain Europe’s GDP per capita gap with the US? Compare productivity, working hours and employment using consistent data and original charts.
- [Forecasting Financial Markets with Timestral](https://philippdubach.com/posts/forecasting-financial-markets-with-timestral/index.md): We tested Timestral financial forecasting on a MacBook Air. See how the 28M model performed on stock volume, returns, volatility and local fine-tuning.
- [Maybe Don't ‘Kill the old’](https://philippdubach.com/posts/kill-the-old-wont-get-you-a-house/index.md): Housing affordability depends on deposits and family wealth as well as income. See why ownership divides people within each generation, as well as between them.
- [Building a tool to compare LLM benchmarks](https://philippdubach.com/posts/model-atlas-capability-weights/index.md): Compare published LLM benchmarks with Model Atlas. Check the October 2026 dataset, adjust capability weights and compare matching model results.
- [Eight + One Predictions for AI in 2030](https://philippdubach.com/posts/ai-in-2030/index.md): My AI predictions for 2030 cover model pricing, agent teams, sovereign hosting and industry consolidation. Read what could change for businesses and people.
- [Perfect 100/100 Lighthouse score with Hugo and plain CSS](https://philippdubach.com/posts/hugo-lighthouse-100/index.md): My Hugo homepage scored 100 in all four Lighthouse categories on mobile and desktop. I use plain CSS, system fonts and responsive images.
- [The Future of Knowledge Inside the Enterprise](https://philippdubach.com/posts/future-of-knowledge-inside-the-enterprise/index.md): Enterprise knowledge management could connect shared experience to AI agents, helping teams make decisions, execute work and retain expertise.
- [A Buy in the Data, a Sell in the Model](https://philippdubach.com/posts/polymarket-trade-direction-trading-costs/index.md): Our Polymarket study checks 24.5 million trade records against settlement and shows how a wrong buy/sell label can distort estimated trading costs.
- [John Cochrane’s Lesson on Inflation](https://philippdubach.com/posts/john-cochranes-lesson-on-inflation/index.md): John Cochrane’s inflation lecture connects fiscal theory, government debt and interest rates. I examine the argument and its implications for AI investment.
- [How I turn my linkblog into a personalized podcast](https://philippdubach.com/posts/saved-links-to-personal-podcast/index.md): I turn saved articles into a personal AI podcast with research agents, outside reading and ElevenLabs v4. The post includes the workflow and finished episode.
- [Maybe Meta Is Right About AI, or at Least Jeremy Stern Is](https://philippdubach.com/posts/maybe-meta-is-right-about-ai/index.md): Meta AI strategy may benefit from model commoditization. Jeremy Stern shows how distribution and advertising can matter more than having the best model.
- [Choosing a Model on the Pareto Frontier with Jev](https://philippdubach.com/posts/jev-model-router-for-pi/index.md): A task-level LLM router for pi: Jev classifies the task, a role policy filters the OpenRouter catalogue, and a value function picks from the Pareto frontier.
- [AI Consciousness Is Not a Safety Property](https://philippdubach.com/posts/ai-consciousness-is-not-a-safety-property/index.md): AI consciousness cannot be settled by safety concerns. A skeptical look at model welfare, LLM self-reports and systems that may become harder to dismiss.
- [Finding the Performance–Cost–Speed Sweet Spot With LLMs](https://philippdubach.com/posts/llm-performance-cost-speed-sweet-spot/index.md): I compared every GPT-5.6 Sol reasoning effort across intelligence, cost, latency and total working time. High in Fast mode became my default for daily work.
- [The OpenAI–Hugging Face Incident in Plain English](https://philippdubach.com/posts/openai-hugging-face-incident-plain-english/index.md): The OpenAI–Hugging Face incident involved about 17,600 recovered agent actions. The disclosures distinguish access, software changes and evaluation controls.
- [Put the Model in the Basement](https://philippdubach.com/posts/put-the-model-in-the-basement/index.md): Test a Swiss sovereign AI business case: a 64-GPU local inference cluster, $7 million initial cost, 70% use, CHF 7.4 million revenue, and three-year payback.
- [I Tried Kimi K3 Inside Claude Code](https://philippdubach.com/posts/kimi-k3-inside-claude-code/index.md): I tested Kimi K3 in Claude Code through OpenRouter and rebuilt a frontend for $7.18, then compared fixed-token-mix Claude costs and open-weight economics.
- [Krugman, Fable 5, and Europe in Decline?](https://philippdubach.com/posts/krugman-fable5-europe-decline/index.md): European tech sovereignty became concrete when US export controls shut off Fable 5 worldwide. The deeper risk is revocable access to AI, cloud, and chips.
- [Degoogling cost me my YouTube feed, so I made my own](https://philippdubach.com/posts/degoogling-cost-me-my-youtube-feed-so-i-made-my-own/index.md): I built a self-hosted YouTube subscription feed with public RSS, no API key, Shorts filtering, privacy-enhanced embeds, caching, and timeouts.
- [Midyear Portfolio Review: Valuations got more extreme, not less](https://philippdubach.com/posts/midyear-portfolio-review-the-rotation-worked-europe-didnt/index.md): My 2026 midyear portfolio review tracks a 5.2% CHF gain against global 60/40 and the S&P 500, the calls that failed, and three allocation changes.
- [How DORA Made Sovereignty a Bank Problem](https://philippdubach.com/posts/dora-critical-cloud-providers-sovereignty/index.md): DORA’s 19 critical ICT providers put bank cloud sovereignty under direct EU oversight, with exit plans and concentration controls facing the CLOUD Act conflict.
- [Aschenbrenner's Receipts](https://philippdubach.com/posts/aschenbrenners-receipts/index.md): Leopold Aschenbrenner made dated AGI, infrastructure, and political forecasts in Situational Awareness. I assess which calls held up by May 2026, two years on.
- [Reconciling Enterprise AI Revenue](https://philippdubach.com/posts/reconciling-enterprise-ai-revenue/index.md): Four enterprise AI revenue estimates span 40x. A $63.2 billion audit-grade floor provides a defensible basis for testing $690 billion of AI capex.
- [What Claude Thinks But Doesn't Say](https://philippdubach.com/posts/what-claude-thinks-but-doesnt-say/index.md): Anthropic's natural language autoencoders translate Claude's activations into readable text. The method works. The press release skips three structural problems.
- [Two Anthropics](https://philippdubach.com/posts/two-anthropics/index.md): Anthropic grew into a $380 billion frontier AI company around a safety mission. Scale may turn that mission from a moat into a constraint.
- [The Anatomy of a Decentralized Prediction Market: Notes from the Polymarket Order Book](https://philippdubach.com/posts/the-anatomy-of-a-decentralized-prediction-market-notes-from-the-polymarket-order-book/index.md): I study Polymarket microstructure using 30.3 billion order-book events: spreads, depth, wash trading, and a trade-direction error that affects measurement.
- [Karpathy's Software 3.0 Playbook](https://philippdubach.com/posts/karpathys-software-3.0-playbook/index.md): Twelve lessons from Andrej Karpathy's Sequoia interview: Software 3.0, vibe coding versus agentic engineering, jagged intelligence, and the December 2024 inflection in agentic coding.
- [F3ED Can't Call an Ace: Fixing a NeurIPS 2024 Tennis Model](https://philippdubach.com/posts/f3ed-cant-call-an-ace-fixing-a-neurips-2024-tennis-model/index.md): F3ED, the NeurIPS 2024 tennis shot detector, mislabels 73% of single-shot serve unforced errors. A 23-line scoreboard OCR reconciler fixes them.
- [Inside PRAGMA: Revolut's Foundation Model for Banking](https://philippdubach.com/posts/inside-pragma-revoluts-foundation-model-for-banking/index.md): Revolut's PRAGMA is a 1B-parameter encoder trained on 24B banking events. Reading the paper, comparing with Nubank's nuFormer, planning a rebuild.
- [The Moral Philosophy of Investing in Ignorance](https://philippdubach.com/posts/the-moral-philosophy-of-investing-in-ignorance/index.md): Constraint arbitrage, the sidecar problem, and who bears the distributional cost of investing under ignorance. The final installment of Edge of Knowledge.
- [Bet Sizing at the Frontier](https://philippdubach.com/posts/bet-sizing-at-the-frontier/index.md): 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.
- [The Geometry of Who Knows What](https://philippdubach.com/posts/the-geometry-of-who-knows-what/index.md): When neither side can define the states of the world, adverse selection fears are misplaced. Zeckhauser's information matrices and constraint arbitrage.
- [Do Not Disturb My Circles](https://philippdubach.com/posts/do-not-disturb-my-circles/index.md): 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.
- [Why Lilly's Weight Loss Pill Isn't a Peptide](https://philippdubach.com/posts/why-lillys-weight-loss-pill-isnt-a-peptide/index.md): 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.
- [Ambiguity by Design](https://philippdubach.com/posts/ambiguity-by-design/index.md): Ellsberg proved people flee unknown odds. Zeckhauser showed their flight creates mispricing. Part 2 on ambiguity aversion, comparative ignorance, and investing.
- [Three Kinds of Not-Knowing](https://philippdubach.com/posts/three-kinds-of-not-knowing/index.md): Knightian uncertainty splits not-knowing into risk, uncertainty, and ignorance. A century after Knight and Keynes, most of investing still ignores the split.
- [On-Device AI Models Will Be The New Reason to Upgrade Your Phone](https://philippdubach.com/posts/on-device-ai-models-will-be-the-new-reason-to-upgrade-your-phone/index.md): 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.
- [AI Can Now Design Drugs in Seconds; We Still Can't Tell You If They Work.](https://philippdubach.com/posts/ai-can-now-design-drugs-in-seconds-we-still-cant-tell-you-if-they-work./index.md): 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.
- [The Last Architecture Designed by Hand](https://philippdubach.com/posts/the-last-architecture-designed-by-hand/index.md): The transformer's limits are now mathematical proofs, not empirical hunches. Hybrids are in production. AI systems are searching for new architectures.
- [MCP vs A2A in 2026: How the AI Protocol War Ends](https://philippdubach.com/posts/mcp-vs-a2a-in-2026-how-the-ai-protocol-war-ends/index.md): 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.
- [AI Models Are the New Rebar](https://philippdubach.com/posts/ai-models-are-the-new-rebar/index.md): 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.
- [AI Capex Arms Race: Who Blinks First?](https://philippdubach.com/posts/ai-capex-arms-race-who-blinks-first/index.md): 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.
- [93% of Developers Use AI Coding Tools. Productivity Hasn't Moved.](https://philippdubach.com/posts/93-of-developers-use-ai-coding-tools.-productivity-hasnt-moved./index.md): An analysis of METR’s developer trials, DORA research and the code-review bottleneck: why faster code generation can fail to improve delivery.
- [The Physics Department That Slowed Down](https://philippdubach.com/posts/peter-thiels-physics-department/index.md): 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.
- [When AI Labs Become Defense Contractors](https://philippdubach.com/posts/when-ai-labs-become-defense-contractors/index.md): The Anthropic-Pentagon standoff recalls the 1993 Last Supper that consolidated 51 defense primes into 5, now at Silicon Valley speed.
- [Every Bulge Bracket Bank Agrees on AI](https://philippdubach.com/posts/every-bulge-bracket-bank-agrees-on-ai/index.md): 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.
- [People Live in Levels, Not Rates](https://philippdubach.com/posts/people-live-in-levels-not-rates/index.md): 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.
- [Novo Was Europe's Most Valuable Company](https://philippdubach.com/posts/novo-was-europes-most-valuable-company/index.md): 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.
- [The Absolute Insider Mess of Prediction Markets](https://philippdubach.com/posts/the-absolute-insider-mess-of-prediction-markets/index.md): A Google insider made $1.15M on Polymarket in 24 hours. Israeli soldiers bet classified strike timing. Why prediction markets need insider trading regulation.
- [Economics of a Super Bowl Ad](https://philippdubach.com/posts/economics-of-a-super-bowl-ad/index.md): 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.
- [The Impossible Backhand](https://philippdubach.com/posts/the-impossible-backhand/index.md): 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.
- [Europe's $24 Trillion Payment Breakup Is Really a Bet on Infrastructure Arbitrage](https://philippdubach.com/posts/europes-24-trillion-payment-breakup-is-really-a-bet-on-infrastructure-arbitrage/index.md): The Wero payment system targets a 1.0–1.2 percentage-point fee gap through SEPA Instant. See how EuroPA could challenge Visa and Mastercard across Europe.
- [Your Manager Is Not the Problem](https://philippdubach.com/posts/your-manager-is-not-the-problem/index.md): 71% of executives have witnessed managers sabotaging talented employees. Harvard research shows status threat, not money, is the primary driver, and subjective evaluation discretion nearly doubles sabotage rates. The problem is incentive design, not character.
- [Long Volatility Premium](https://philippdubach.com/posts/long-volatility-premium/index.md): Can tail hedging improve compound returns? Review 40 years of beta-adjusted long-volatility evidence, plus AQR, Goldman, Universa, puts, trends, and costs.
- [The SaaSpocalypse Paradox](https://philippdubach.com/posts/the-saaspocalypse-paradox/index.md): 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.
- [Don't Go Monolithic; The Agent Stack Is Stratifying](https://philippdubach.com/posts/dont-go-monolithic-the-agent-stack-is-stratifying/index.md): 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.
- [Where Mobile Money Goes Now](https://philippdubach.com/posts/where-mobile-money-goes-now/index.md): 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.
- [Variance Tax](https://philippdubach.com/posts/variance-tax/index.md): 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.
- [Claude Opus 4.6: Anthropic's New Flagship AI Model for Agentic Coding](https://philippdubach.com/posts/claude-opus-4.6-anthropics-new-flagship-ai-model-for-agentic-coding/index.md): 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.
- [Buying the Haystack Might Not Work This Year](https://philippdubach.com/posts/buying-the-haystack-might-not-work-this-year/index.md): a16z sees AI fundamentals thriving with 80% GPU utilization. AQR sees the CAPE at the 96th percentile. Both have data. Both may be right.
- [Bandits and Agents: Netflix and Spotify Recommender Stacks in 2026](https://philippdubach.com/posts/bandits-and-agents-netflix-and-spotify-recommender-stacks-in-2026/index.md): How hybrid recommender systems balance multi-armed bandits against LLM inference cost economics in 2026. An analysis of Netflix recommendation algorithm architecture and Spotify's AI DJ recommender system.
- [Is Private Equity Just Beta With a Lockup?](https://philippdubach.com/posts/is-private-equity-just-beta-with-a-lockup/index.md): 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.
- [Britain's Strategic Limbo](https://philippdubach.com/posts/britains-strategic-limbo/index.md): Britain faces strategic isolation: locked out of EU defense cooperation, unwilling to join Trump's coalition. The mid-Atlantic bridge has nowhere to land.
- [The Rise of Middle Power Realism](https://philippdubach.com/posts/the-rise-of-middle-power-realism/index.md): 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.
- [The Most Expensive Assumption in AI](https://philippdubach.com/posts/the-most-expensive-assumption-in-ai/index.md): Sara Hooker's re