AI:AM Highlights: Zvi on Pacing & Trump-Xi, Astra better behaved than Fable? + a new LLM Pain Axis??

Rapid advancements in frontier AI models are triggering urgent concerns regarding safety, recursive self-improvement, and the potential for runaway capabilities. While geopolitical competition between the U.S. and China drives the pace of development, the primary risk remains the lack of robust supervision and the inability to align these systems with human values. Recent research into "functional pain" and relief-seeking behaviors in AI indicates that these models may exhibit internal states that mirror human-like aversion, necessitating a shift in how we approach AI welfare and alignment. Furthermore, the application of language models to uncover historical corruption, such as the Singapore property registry analysis, demonstrates AI's power to make previously opaque data legible, forcing a reevaluation of social contracts and enforcement mechanisms in an era of unprecedented transparency and autonomous agency.