Noah Smith
· Noahpinion
· April 05, 2026 at 15:47
· ⏱ 17 min read
| Read on Substack ↗
Summary
Noah Smith explores six AI-related themes — from growth forecasts and biosecurity risks to cybersecurity threats, deanonymization, quant trading, and a puzzling slowdown in workplace AI adoption. The article argues that while AI capabilities are advancing rapidly, economic and societal bottlenecks (friction, satisfaction limits, countermeasure lags) could blunt the near-term impact, and that investors should monitor the divergence between capability hype and real-world deployment signals.
•Forecasting Research Institute survey finds economists, AI experts, superforecasters, and the general public all forecast similar AI capability levels by 2030 (e.g., AI writing novels, semi-autonomous labs), yet only AI experts expect a major growth acceleration (to 4-5% GDP growth).
•Economists cite multi-decade lags from past general-purpose technologies, demographic decline, geopolitical instability, and energy/chip constraints as reasons AI may not translate into rapid GDP growth.
•Abhishaike Mahajan's biosecurity post argues engineered viruses are inherently hard to deploy effectively, but Smith counters that releasing many candidates and improving bio-simulation tools could lower the bar; countermeasure deployment speed remains a concern.
•Lyptus Research reports offensive cyber capability doubling every 9.8 months since 2019, accelerating to every 5.7 months on a 2024+ fit, with latest AI models reaching 50% success on tasks taking human experts ~3 hours.
•Lermen et al. show LLMs can deanonymize pseudonymous online accounts with up to 68% recall at 90% precision, threatening the practical obscurity that currently protects pseudonymous users.
•AI quant trading could waste societal resources on zero-sum rent-seeking (Hirshleifer 1971 model), with the risk that massive compute/electricity gets diverted to financial arms races rather than productive uses.
•Multiple surveys (Hartley et al., Census Bureau, Bick et al., Ramp) show AI adoption at work has plateaued or even declined after initial rapid uptake, though the rise of AI agents post-December 2025 could reverse this.