Paul Hünermund

According to a new NBER paper, to justify the AI investment surge we're seeing, the implied productivity gains need to be enormous. The authors calibrate a model where AI-sector productivity rises by roughly 2.7x. Link: www.nber.org/papers/w35290

What Investment Data Implies about the AI Transition nber.org

With large numbers of p's you don't need randomization. Actually makes more sense than the Rabois version, but be aware: arxiv.org/abs/2108.11294

Double Machine Learning and Automated Confounder Selection -- A Cautionary Tale arxiv.org

The current AI buildout only makes sense if it delivers massive productivity gains. The biggest US tech firms are on track to spend ~$755B on AI capex in 2026, up from $155B in 2022. At this point, AI is not just a tech story. It's a macro bet.

🧵1/4 📢 Call for submissions: Causal Data Science Meeting 2026 Join researchers and practitioners from academia and industry for a virtual meeting on Nov 4–5, 2026, exploring the role of causality in machine learning and AI. #CDSM2026

I tried Positron AI in RStudio today and it was amazing

It's funny how, just a few years ago, a random cold email could easily cost you 10 minutes of your day. How did you all deal with that before LLMs?

"Initially, Anthropic silently degraded Fable 5’s performance for users detected to be working on LLM research through invisible interventions that weakened the model’s outputs without notifying the user." 1/2