Models performing better on PhD-level physics than high school physics reflects differences in how questions are formulated (abstractions vs. visual aids) rather than task difficulty, suggesting performance is heavily dependent on task presentation rather than underlying capability.

Models can complete longer and longer tasks, with the doubling time for task length reaching approximately seven months in recent studies, suggesting exponential improvement in sustained reasoning capability.

AI development will likely be compute-constrained by approximately 2030, creating a bottleneck where either the global economy cannot sustain continued scaling or a new paradigm of training is required.

There is a pattern where AI safety researchers have historically been skeptical that current AI systems pose near-term risks, but recent models' capabilities (superhuman performance on math and coding) have surprised them, providing some evidence against worst-case scenarios about alignment but not strongly disproving concerns.

When communicating about policy and technical AI work, one should aim for intellectual honesty about timelines and risks; the 'slow world' compartmentalization is defensible in personal life but not in professional/policy contexts where truth-telling is primary

Long-term projects that take years to pay off (PhDs, research programs, papers) may be difficult to justify if timelines are genuinely short (2-5 years), creating a problem: we need long-term projects but individuals are psychologically averse to them under short-timeline belief