Import AI 464: Fable writes GPU kernels; AI automation; and analog computation

Recent developments in artificial intelligence research demonstrate significant progress in both technical automation and real-world task execution. The AI system Fable has successfully written a high-performance GPU kernel, achieving an 18.71X speedup compared to standard baselines. This capability is notable because it suggests that AI is becoming increasingly proficient at fundamental research and development tasks, potentially enabling recursive self-improvement where AI systems design and optimize their own underlying architecture.

Beyond technical research, AI agents are showing rapid improvement in performing complex, economically valuable online labor. Data from the Remote Labor Index indicates that the success rate for AI systems completing end-to-end projects—such as 3D modeling, graphic design, and data analysis—has increased significantly over the past year. While these systems are not yet fully reliable, their ability to handle multi-step, long-horizon tasks is being tracked through benchmarks like OSWORLD 2.0, which tests an agent's ability to navigate professional software and web-based portals over several hours.

These advancements highlight a shift toward AI-integrated business operations, exemplified by large-scale inventory management systems like JD’s Oxygen AI Item Center. By weaving intelligence into back-office functions, organizations can operate at massive scales with self-updating capabilities. As AI systems become more adept at using computers to perform diverse, multi-hour tasks, they are increasingly positioned to compete with human labor, raising questions about the future of employment and the speed at which AI capabilities are outpacing human comparative advantages.