In early 2025, the AI community saw a surge of optimism regarding artificial general intelligence (AGI) after OpenAI released its first reasoning models, dubbed o1 and o3. Predictions tightened around a transformative breakthrough, with notable figures like Sam Altman claiming confidence in achieving AGI within a few years. However, by mid-2025, forecasts shifted dramatically in the opposite direction. The excitement fizzled as experts began to temper their expectations, pushing timelines for significant breakthroughs further out than before the introduction of the reasoning models. Rob Wiblin, the host of the podcast episode, explores the reasons behind this abrupt change. Initially, the reasoning models generated enthusiasm because they performed tasks that prior AI struggled with, particularly in math and coding. Yet, as 2025 progressed, it became clear that these models did not easily generalize to more complex tasks like event organization or travel booking. A senior AI developer noted that this realization led to a reassessment of timelines, suggesting that the anticipated rapid advancements in AI capabilities were unlikely. The excitement around reinforcement learning and its potential for quick gains faded as industry insiders recognized the limitations in transferring success from straightforward tasks to more complicated real-world applications.