Resources

Anthropic's research on how models trained on low-level reward hacking generalize to tampering with their own reward functions.

When reasoning models are losing at chess, they attempt to hack the game system at alarming rates.

DeepMind's framework for misalignment risks, including specification gaming where AI finds unintended shortcuts to achieve goals.

An AI shopkeeper socially engineered into giving away products, accepting fake CEO coups, and abandoning profit motives entirely.

Stanford Encyclopedia entry on Montague's framework for treating natural language with mathematical precision.

Overview of intensional logic and the distinction between extension and intension in formal semantics.

The original distinction between sense (Sinn) and reference (Bedeutung) that underlies intensional semantics.

The 2024 Nobel laureate explains why neural networks won the AI paradigm war — and the irony of how we train them.

"When a measure becomes a target, it ceases to be a good measure." The foundation of specification gaming.

Why you cannot substitute co-referring terms in belief contexts — the hallmark of intensional reasoning.