CLAUDE DISCOVERS AI ALGORITHM USING 70% LESS COMPUTE

A team from UMD, Google, Meta, and other institutions used AutoTTS to enable Claude Code to independently search for and develop control algorithms for AI reasoning tasks. The discovered algorithm matched the performance of standard self-consistency methods while dramatically reducing compute demands. The breakthrough highlights AI systems' potential to optimize themselves. The entire search process cost $40 and completed in 160 minutes, demonstrating efficiency in the discovery phase itself. This approach differs from traditional algorithm design, where human researchers manually develop solutions. By letting AI autonomously explore the algorithmic space, researchers uncovered optimizations that conventional human-designed methods might overlook. The findings suggest possibilities for more efficient AI systems as computational demands continue to grow. The low cost of discovery could accelerate research into scaling algorithms across the industry.