Everyone agrees that AI systems have become exponentially more capable over the last few years, but they might already be good enough to take over from human AI researchers.
Jerry Tworek, OpenAI’s former VP of Reinforcement Learning, recently shared a provocative theory about the future of AI development that challenges assumptions about human researchers’ continued centrality in the field. His comments suggest even those at the forefront of AI development are grappling with uncomfortable possibilities about the technology’s trajectory.
“ There is one version of the world where — which is not completely impossible, although slightly pessimistic on humans — that says coding agents are so good right now it’s getting them to the moment where they can automate AI research, and then have the models research better models,” Tworek said. “Because maybe we are at the last model that humans could have figured out,” he added.
He acknowledged the unsettling nature of this framing: “And it’s not a completely impossible framing. Maybe it makes sense with all those GPUs we have and all those really capable models and their tenacity. Maybe they should be researching future models.”
Yet Tworek ended on a note that balanced between resignation and hope: “But you know, maybe there are still a few things humans can do that we can put our heads to work.”
The implications of Tworek’s comments are striking, particularly given his former position overseeing reinforcement learning at one of the world’s leading AI labs. His scenario suggests we may be approaching an inflection point where AI systems become the primary architects of their own evolution—a form of recursive self-improvement that has long been theorized but never realized. Tworek’s scenario is the fast take-off that many experts anticipate — if AI systems become better than humans at doing AI research, they could end up doing the research much faster than humans, given how they can work all day without needing breaks, and hundreds of such AI instances can be sprung up nearly instantly to tackle problems in parallel. Some, like Elon Musk, say that we have already entered the singularity. And with researchers at prominent labs also saying that we might have made the last model made by humans, it appears that the singularity might indeed be closer than most people expect.