Logan Kilpatrick, Google’s product manager for AI Studio and former OpenAI developer relations leader, has made a bold prediction that a direct path to artificial superintelligence (ASI) may be achievable without focusing on intermediate milestones. In a statement on X (formerly Twitter) on Monday, Kilpatrick suggested that this approach is “looking more and more probable by the month.”
The key to this potential breakthrough lies in scaling test-time compute—the computational resources used when an AI model performs tasks or answers questions in real-time. Kilpatrick cited the success of this approach as a “good indication” that reaching ASI directly might be feasible, representing a significant shift in AI development strategy.
This perspective marks a notable evolution in thinking about the path to advanced AI. While much of the industry conversation has centered on achieving artificial general intelligence (AGI)—where AI matches or surpasses human capabilities across a broad range of tasks—Kilpatrick believes AGI will arrive more quietly than expected. “It’s likely going to just look a lot like a product release” rather than a dramatic, singular moment, he explained.
The context for this shift is crucial: evidence suggests that pretraining AI models has plateaued, forcing companies and researchers to explore alternative improvement methods. Both Google and OpenAI have recently unveiled models with enhanced reasoning abilities that “think” through problems more like humans do, moving beyond simple pattern recognition.
Kilpatrick’s comments also reference Ilya Sutskever, OpenAI’s cofounder and former chief scientist, who left the company this year to launch Safe Superintelligence, a startup dedicated to pursuing “safe superintelligence in a straight shot, with one focus, one goal, and one product.” Kilpatrick suggested Sutskever may have identified “early signs” of test-time compute’s potential, which could explain his focused approach.
Remarkably, Kilpatrick admitted he previously believed Sutskever’s single-minded method would be a mistake but has since changed his view. When pressed about the advantages of a more iterative approach, he hedged: “I’m more bullish on iterating than I am straight shot, but the latter just might work.”
Kilpatrick’s opinion carries significant weight in the AI community, given his previous leadership role at OpenAI and his recent move to Google, which industry insiders viewed as a major win for the search giant. One source previously described him as a “secret weapon” for Google in the intensifying AI race.
Key Quotes
it’s likely going to just look a lot like a product release
Logan Kilpatrick, Google’s AI Studio product manager, describing how AGI will likely arrive—not as a dramatic singular moment but as a routine product launch, suggesting the transition to advanced AI may be more gradual and less obvious than many expect.
looking more and more probable by the month
Kilpatrick’s assessment of the feasibility of achieving artificial superintelligence directly without focusing on intermediate milestones, indicating his growing confidence in this approach based on recent developments in test-time compute scaling.
pursue safe superintelligence in a straight shot, with one focus, one goal, and one product
Ilya Sutskever’s mission statement for his new startup Safe Superintelligence after leaving OpenAI, representing the focused approach that Kilpatrick now believes may actually succeed despite his initial skepticism.
I’m more bullish on iterating than I am straight shot, but the latter just might work
Kilpatrick’s nuanced position when pressed about different approaches to superintelligence, showing he still favors gradual iteration but acknowledges the direct path could succeed—a hedge that reflects the uncertainty even experts face about AI’s future trajectory.
Our Take
Kilpatrick’s public reversal on the viability of a direct path to superintelligence is particularly noteworthy given his insider perspective at both OpenAI and Google. His willingness to reconsider Sutskever’s approach suggests concrete technical progress that insiders are witnessing but may not yet be fully public. The emphasis on test-time compute represents a fundamental shift from the “bigger models, more data” paradigm that dominated AI development for years. This could democratize advanced AI development somewhat, as reasoning improvements during inference may require less massive upfront training infrastructure. However, the casual discussion of superintelligence timelines—with a senior Google executive suggesting it’s increasingly probable—should raise both excitement and concern. The gap between technical capability and societal readiness for superintelligent systems remains vast, and the industry’s apparent acceleration toward ASI may be outpacing crucial safety and governance frameworks.
Why This Matters
This statement from a senior Google AI leader signals a potential paradigm shift in how the industry approaches superintelligence development. For years, the conventional wisdom held that AI would progress through clearly defined stages—from narrow AI to AGI to ASI—with each milestone requiring distinct breakthroughs. Kilpatrick’s suggestion that a direct path to ASI might be viable challenges this incremental model.
The implications are profound for AI companies, investors, and policymakers. If test-time compute scaling proves to be the key, it could accelerate timelines for advanced AI capabilities far beyond current expectations. This matters because superintelligence—AI that surpasses human intelligence across all domains—would represent a transformative technology with unprecedented societal impact.
The shift also reflects growing recognition that traditional pretraining methods are hitting limits, forcing innovation in AI development approaches. Companies investing heavily in reasoning-focused models and test-time compute may gain significant competitive advantages. For businesses and workers, this suggests the pace of AI-driven disruption could accelerate faster than anticipated, making adaptation and preparation even more urgent.
Recommended Reading
For those interested in learning more about artificial intelligence, machine learning, and effective AI communication, here are some excellent resources:
Recommended Reading
Related Stories
- Sam Altman’s Bold AI Predictions: AGI, Jobs, and the Future by 2025
- Artificial General Intelligence Could Arrive by 2024, According to AI Experts
- Google’s Gemini: A Potential Game-Changer in the AI Race
- The AI Hype Cycle: Reality Check and Future Expectations
- The Artificial Intelligence Race: Rivalry Bathing the World in Data