The debate pitting AI safety against AI opportunity presents a false choice. Historical parallels, like the railroad industry, show that safety regulations (e.g., standardized tracks, air brakes) were essential for enabling greater speed, reliability, and economic potential. Trustworthy AI will unlock greater opportunity.

The primary danger in AI safety is not a lack of theoretical solutions but the tendency for developers to implement defenses on a "just-in-time" basis. This leads to cutting corners and implementation errors, analogous to how strong cryptography is often defeated by sloppy code, not broken algorithms.

Unlike previous technological revolutions that unfolded over centuries, allowing for societal adaptation, the current AI transition is happening too fast. This speed prevents the development of adequate mitigations, understanding, and defenses. The common-sense intuition that "we are going too fast" is the correct and most important take.

The focus on AI risk overlooks the opportunity cost of "pacing." Delaying frontier model development, while potentially safer, also pushes back the timeline for major societal benefits like AI-driven scientific discoveries and disease cures, creating a difficult societal trade-off.

Instead of only slowing down risky AI, a key strategy is to accelerate beneficial technologies like decision-making tools. This 'differential technology development' aims to equip humanity with better cognitive tools before the most dangerous AI capabilities emerge, improving our odds of a safe transition.

A fundamental tension within OpenAI's board was the catch-22 of safety. While some advocated for slowing down, others argued that being too cautious would allow a less scrupulous competitor to achieve AGI first, creating an even greater safety risk for humanity. This paradox fueled internal conflict and justified a rapid development pace.

The default assumption is that slowing innovation is inherently bad. With a technology as potent as AI, a deliberate slowdown is a feature, providing critical time to understand the systems, manage disruptions, and build governance structures before irreversible consequences occur. A true halt is not the alternative.

The current approach to recursive self-improvement involves AIs gradually assisting human researchers, a stark contrast to Eliezer Yudkowsky's vision of a solo AI rapidly rewriting its own code. This modern, human-in-the-loop model is slower and offers more opportunities for safety checks and oversight.

Techniques created to make AI safer and more aligned with human intent, such as Reinforcement Learning from Human Feedback (RLHF), have turned out to be the very methods that significantly enhance model performance and usability. Safety work is capability work.