Electricity took forty years to show up in productivity statistics, because factories had to be rebuilt and work redesigned around it, not just rewired. Same story with AI: the bottleneck is organizational, not technical, and handing it to the intern who likes ChatGPT is not a strategy. Good news for whoever learns fast, the roles are barely defined yet.
Intelligence is not the unit of agency, organization is: a mind gets complex work done the way a society does, by arranging agents that are individually limited. We complain about the reliability of agentic LLM systems, yet every company ever built runs on humans who are not fully reliable either. We already know how to operate in a non-deterministic world, so the problem is theoretically solved, it is a matter of abstraction.
I don't know how important it is to define a hard "AGI" threshold. What matters is the evidence: current models are extremely competent across domains, and this paper lays it out clearly.
Models are good enough. 16 parallel agents produce a 100K-line C compiler autonomously. What makes it work is the ecosystem: structured task decomposition and shared context across agents.
The engineer's job becomes ecosystem design: specifying intent, building feedback loops, and curating context so agents do reliable work. This is about coding, but it is the roadmap for all agentic knowledge work.
The missing layer in agentic ecosystems: structured memory of why decisions happened, not just what happened. Context graphs fight agent amnesia across steps, agents and sessions.
Intelligence requires action to create value. High-agency workers who actively apply AI capabilities outperform those who passively rely on raw intelligence.
AI reshapes the swiss job markets now. Expertise is the competitive divide.
Selection bias distorts ESG scores. Companies that don't report emissions differ systematically from those that do. Correcting for this bias dramatically changes sustainability rankings.
GenAI adoption shows measurable productivity gains in social sciences. Early-career researchers and non-native speakers benefit most.
Static workflows become intelligent systems. The framework maps how research infrastructure transforms toward autonomous science.
Corporate scientists bridge academic research and internal innovation by absorbing external findings and applying them within their organizations. This evidence-based approach lets sectors build on documented experiments rather than starting from scratch.
AI as a teammate improves both what you produce and how it feels to produce it. Even better when your whole team works this way.
LLMs now match expert-level performance in economically relevant fields. Learning to collaborate with AI isn't optional anymore. It's the skill that separates those who thrive from those who fall behind.
A well-crafted prompt, a frontier model, some planning data, and you end up with energy retrofit plans for building portfolios that match more than 90% of specialist work. Human-AI collaboration can help us in tackling very hard problems.
ESG raters who also run indexes systematically give higher scores to companies with strong returns. This looks like incentive bias, not better information. A potential case of conflict of interest.
LLMs now exceed what we thought possible just months ago. The real challenge is learning to work with intelligence we didn't create and can't fully control.
Breakdown of chunking strategies for RAG success. Chunking decisions can make or break the whole system. Smaller chunks with semantic boundaries and LLM-augmented processing consistently outperform the defaults everyone uses.
Found this on a friend's shelf. Changed how I think about learning and doing.
Vibe coding was a fun period. AI-assisted development is what sticks.
This year's progress feels smaller than last year's breakthroughs. That's exactly why it matters. Small improvements can compound into large returns.
Understanding why AI hallucinates changes everything. We need to think differently about the problem.
He saw our world coming in 1997. Now we're the ones staring down the next wave.
One of current multiple examples on the power of well curated synthetic data.
Great explanation of why and how multiple AI agents work together. I integrated this approach into my workflow immediately.
What we consider extremely simple tasks are packed with complexity for Large Language Models. Context and drift management are key to ensuring models don't go sideways.
Taught me that working with AI means one thing: simplify tasks until they're manageable.
Models carry bias and prefer things humans don't actually want. This is a clear alignment problem.
Give an LLM contradictory objectives and watch it create solutions you never wanted, following logic you can't predict.
Comprehensive look at using models to judge models. The foundation of building codebases that actually work with non-determinism.
Solid framework for working with AI responsibly. How to think about when and why to use any model, regardless of how powerful it gets. Great introduction to agency concepts.
AI moves fast, books move slow. This one from 2024 still hits. An essential.
Clear breakdown of what AI agents are, how they're built, and how they use tools. Opened the door to working with AI in ways that actually matter.
Complete guide to prompting strategies and techniques. Gives you everything you need to build your own best practices.
Certainty is the pinnacle of economic crisis. It has been for centuries. We're not living in different times. Same patterns and same hubris.
Getting models to code instead of think can dramatically improves logical reasoning. Writing functions forces precision that pure reasoning misses.
ESG practices boost real estate financial performance through operational efficiencies, lower risk, and investor appeal. The gains aren't uniform across markets, regulations, or ESG components.
Setting a biodiversity strategy starts with one hard truth: understand before you solve. Everything else follows from there.
Communication determines everything in AI-augmented work. How clearly you transmit intention and context is what separates terrible output from brilliant results.
ESG score differences erode investor confidence, weakening the scores' ability to actually move equity prices.
Physical and transitional climate risks boost REIT volatility.
Climate risks boost REIT return volatility across developed and developing countries. But drill down to sectors and the pattern flips: industrial, retail, and self-storage REITs actually see less volatility from climate risk.
Clean, research-backed take on what leadership actually means. Shows you what studies say about effective leaders.
Models love their own work, round numbers, and stick to familiar anchors. Turns out AI bias looks exactly like human bias.
Heat pumps win as the most cost-efficient renewable system across rural and urban settings. Synthetic natural gas fails economically in most configurations.
Words carry ideas. Clear words make them land.
Models work better with examples, just like humans do. Few-shot prompting delivers powerful results without the cost.
Conflicting objectives create chaos, whether you're managing teams, building systems, or handling data. We get the results we measure for.
We assume that caring for others at work comes naturally. It doesn't. Evidence-based approaches to leadership and team care are incredibly powerful. Too few uses them.
Reality is messy. Most challenges don't have clear answers. Humble questions are just better ways to think through them. When you're comfortable with uncertainty, the right questions become your navigation system.
What sustainability experts actually focus on: economic viability, social impact, and whether people's lives get better.
We spend thousands of hours at work, then let workplace BS dominate instead of what actually works. Time to cut the feel-good nonsense and focus on approaches that are genuinely effective.
Goes deep on macroeconomic trends and actually explains the money behind global politics. Cuts through news headlines to show you what's driving international events.
Solid breakdown of current market trends and economic news without the usual YouTube finance bro nonsense. Actually explains what's happening.
Finally put deep and clear words on a long time feeling. An essential for authocritics and thinking the field of economics in it's own system.
Climate risk are not yet integrated into real estate prices.
Solid book on imposter syndrome that goes beyond personal struggles. Covers responsibility for protecting the people and teams you lead.
ESG score divergence is explained by measurement differences (56%), scope variations (38%), and weighting choices (6%). Within measurement divergence, raters' overall firm perceptions bias their scoring of individual ESG categories.
ESG rating approaches span four levels, based on asset type and data collection. Current methodologies show significant gaps in measuring social outcomes, governance effectiveness, and climate risk quantification.
ESG real estate evaluations ignore user well-being. Ratings diverge because of scope, indicators, and aggregation differences.
Climate risk pricing works differently across locations—areas already hit might have risk baked in or ignore future threats, while new areas face stronger, longer-lasting effects. The risk spreads beyond individual buildings to entire areas, but valuers often lack the cross-disciplinary knowledge to price it right.
ESG ratings vary because of scope and design differences. Even firms with consistently good ESG scores across different systems show no performance advantage. The rating heterogeneity spreads ESG investments too thin to create real returns.
Shows how vulnerability, conflict, and climate change feed off each other. Creates a vicious cycle that traps societies in escalating violence and environmental damage.
Sustainable REITs earn premiums, face lower systematic risk, attract informed traders, and generate higher rental revenue with better operating income and cheaper financing. Reputation drives much of this advantage.
Larger firms score better on ESG ratings. This might reflect better sustainability reporting, not better sustainability practices.
Changed how I think about taste completely. Turns out good taste isn't something you're born with. It's something you develop.
Essential paper for synthetic control methods. A powerful analytical tool that is genuinely intuitive.
Great meta-analysis for understanding identification problems in climate change research. Reveals that applied econometrics is as much about methodological taste as technical precision.
Essential for applied econometrics. Goes beyond just using classical tools to actually understanding what they tell you.
Never understood how a writing book could be this dense. Taught me I'll never be great, but I can get better.
Essential for econometrics work on climate change. It lays the foundation for solid identification strategies.
Classic crowd psychology. Shows how groups override individual judgment.
Put self-interest driven actors in perspective. Open to different ways of seeing society's interrelations