Finding it hard to keep up with AI research? Roblox's CEO says you're not alone

The rapid pace of new AI research, often emerging daily and filled with complex terminology, is challenging even for industry leaders. Roblox's CEO, David Baszucki, acknowledges this struggle, expressing his thoughts during a recent episode of the "Access" podcast. Reflecting on his recent sabbatical spent exploring AI, Baszucki described the experience as both enlightening and humbling, admitting, "It is hard to get to a position where you can understand all of those research papers." Founded in 2005, Roblox has come a long way since its inception. Baszucki reminisced about the company's early days when he was able to grasp all aspects of research, from physics simulation to rendering technology. However, he noted a significant shift with the advent of a massive wave of AI research, which he described as a "horizontal wave" that is evolving at an unprecedented speed. Terms such as transformers and diffusion models have entered the conversation, and there’s an abundance of developments to keep track of. As AI research transitions from an academic interest to a crucial national focus, major technology firms like Meta and Microsoft have established their own research labs and are offering substantial salaries to attract top AI talent. This shift has also led to a more secretive approach to research; for instance, in 2023, Google announced it would be scaling back the amount of AI research it publishes, with a Google Brain staff member emphasizing the need to "compete and keep knowledge in house." While many in the AI sector are currently prioritizing computational scaling, OpenAI co-founder Ilya Sutskever pointed out that genuine research is vital for the future of the technology. He remarked, "It's back to the age of research again, just with big computers." In Baszucki's view, the exploration of AI is still in its infancy, particularly in the 3D realm. He described the AI landscape as a "very physically unnatural space," where models are trained on human-created text and images rather than the raw, three-dimensional materials of the world. This perspective sheds light on the ongoing challenges and potential of AI development in gaming and beyond.