AI wasn't overhyped. Almost everyone underestimated it (benjamintodd.substack.com)

A recent analysis reveals that rather than being overhyped, AI advancements have significantly outpaced expectations over the past 18 months, with even the most skilled forecasters underestimating the speed of progress. This re-evaluation emerged as AI surpassed predictive benchmarks in areas such as mathematics and task automation. For instance, many experts predicted AI would reach certain proficiency milestones in 2026, but AI achieved these goals as early as 2025. While AI company leaders offered more optimistic forecasts, they too were often overly conservative, especially regarding tasks that were deemed too complex or far-fetched, like solving Millennium Prize problems. This accelerated progress is crucial for the AI/ML community as it challenges the historical skepticism surrounding AI capabilities. It underscores the pressing need for reassessing strategies related to AI governance and safety to accommodate rapid advancements. The findings suggest that consistent trends could lead us toward transformative breakthroughs sooner than anticipated, prompting a call from experts to prioritize measures that can mitigate risks associated with fast-paced AI development. As AI continues to evolve, the need for proactive approaches to manage its implications has never been more critical.