Roundup #89: It isn’t X, it’s Y

* Americans are more concerned about AI losing human control than job loss.
* A game theory model suggests leading AI companies might slow development to allow alignment research time.
* China refines raw materials; other nations mine and ship ore to China for processing.
* Mass deportations have not resulted in increased native-born unemployment or prime-age employment rates.
* Rent has stabilized or fallen since 2022, relative to the post-pandemic period.
* Rent appears more affordable relative to income compared to the late 2010s.
* Job wages increased faster in AI-exposed jobs than in less-exposed jobs.
* Unemployment rates among recent college graduates did not spike in summer 2026 according to one analysis.
* Digital media jobs are identified as a category potentially facing displacement by AI.

The discussion around AI safety and development reveals a tension between the competitive drive for advancement and existential risk mitigation. The game theory approach introduces a rational incentive structure where slow progress becomes an option under high-stakes scenarios, but its real-world applicability is limited if safety research itself is not progressing at the necessary pace. This highlights a potential epistemic gap: the model’s success depends on assumptions about the trajectory of safety science that remain unverified. Simultaneously, the economic dynamics show that while AI exposure boosts wages in some areas, the threat to creative and fulfillment-based employment warrants separate attention.
The analysis of industrial shifts points toward entrenched global supply chain control, where refinement capacity dictates power rather than raw material extraction. The slowdown in Chinese investment, linked to constrained banking lending, suggests that financial stability is a more immediate bottleneck for industrial growth than pure production capacity. This dynamic echoes historical patterns where systemic constraints—like the relationship between finance and growth—can decouple stated economic success from actual prosperity.
The perspective on risk management further distinguishes between tangible, bounded risks like cyber threats and abstract, high-impact biological risks. The argument that defense against AI-driven cyberattacks is plausible because perfecting code is theoretically achievable contrasts sharply with the infeasibility of managing biological systems, specifically the self-limiting evolutionary mechanisms that would prevent an extinction virus from emerging. This raises a critical distinction between technical mastery (code) and systemic complexity (biology), suggesting that our capacity to control risk depends fundamentally on understanding the nature of the system being managed.