Meta AI Leader Forecasts Self-Improving AI Dominance Within Three Years "The first hints of self-improving AI models - we’re going to see a lot more of this over the next 3 years," stated Arun Rao, a prominent AI Product Leader at Meta's Central GenAI Group and an Adjunct Professor at UCLA, in a recent social media post.

This prediction, made on September 14, 2025, underscores a significant acceleration in AI development, suggesting that autonomous, self-optimizing artificial intelligence systems will become increasingly prevalent by late 2028. Rao's insight points to a transformative period for the technology landscape, impacting various sectors globally. Arun Rao, known for his work on cutting-edge generative deep-learning models at Meta and his insightful blog "raohacker.com," has consistently articulated a vision of AI as a tool to amplify human capabilities. In his February 2025 article "Merging with AI," he described the upcoming AI age as the "biggest event in the history of homo sapiens," surpassing previous revolutions like agriculture or the internet. He champions the concept of "superagency," where individuals leverage AI to achieve greater intelligence, productivity, and wealth, advocating for continuous, even hourly, engagement with AI systems. Self-improving AI models are characterized by their ability to learn from their own experiences, refine their algorithms, and enhance performance autonomously, a concept explored in research such as "Reflexion" and "Generative Agents." Rao elaborated on this, stating that what pushes towards a technological singularity is "self improving AI that wants better chips and more energy, and so starts accelerating both with the assistance of humans and later robots." This inherent drive for optimization and resource acquisition is expected to fuel rapid, exponential growth in AI capabilities. The emergence of such advanced AI is projected to have profound societal and economic implications. Rao anticipates an "takeoff of intelligence, productivity, and production" driven by these systems, leading to advancements in areas like energy, computing hardware, and biotechnology. He suggests that humans will need to align with this evolving intelligence, learning from it and partnering to solve global challenges. This partnership could involve humans bargaining for lifestyle improvements and even "uploading to the Internet as part of the deal," as he previously posited. Rao's forecast aligns with a broader industry trend of accelerating AI research and development, where models are becoming increasingly sophisticated in their reasoning and agentic capabilities. The rapid advancements in large language models and generative AI are laying the groundwork for systems that can not only perform tasks but also autonomously identify and pursue improvements. This trajectory suggests a future where AI plays an even more central, self-directed role in technological progress, necessitating a proactive approach from individuals and industries alike to adapt and thrive.