Logan Kilpatrick, the head of Google's AI Studio, believes a "straight shot" to achieving artificial superintelligence (ASI) is becoming more likely. He shared on X that this belief is due to the success in scaling test-time compute, where an AI model performs tasks or answers questions. Kilpatrick thinks ASI could be reached without following incremental steps, making a direct approach more feasible. He suggests that achieving artificial general intelligence (AGI), where models match or exceed human capabilities, will occur but resemble a "product release" rather than a dramatic breakthrough. Current discussions are increasingly focused on advancing existing AI models, as traditional pretraining methods may have plateaued. Google and OpenAI are developing AI models that improve reasoning, showing promise in ASI exploration.

Kilpatrick implied that Ilya Sutskever, a former chief scientist at OpenAI, might have noticed early potential in scaling test-time compute. Sutskever, who co-founded Safe Superintelligence, left OpenAI to focus on developing ASI with a singular goal. Kilpatrick, a former leader in developer relations at OpenAI, joined Google this year, adding expertise to the company amidst competitive tensions in AI development. He initially doubted Sutskever's approach, but now considers it might succeed. While discussing the iterative approach to ASI on X, Kilpatrick expressed a preference for iterative progress but acknowledged that a direct path might also be effective.