Mechanism Capital Partner: The scale of real AI data will expand by 100 times by 2026

PANews January 2 News, Mechanism Capital partner Andrew Kang posted on X platform that by 2025, breakthroughs in the robotics field will have addressed longstanding challenges in model architecture and training, and significant progress will have been made in data collection technology, data quality understanding, and data formulation. This will give AI companies the confidence to finally start investing in large-scale data collection. Companies like Figure, Dyna, and PI are utilizing innovative reinforcement learning (RL) techniques to achieve over 99% success rates in various practical applications. In addition, advances in memory technology have broken the “memory wall.” NVIDIA’s ReMEmber uses memory-based navigation, Titans and MIRAS achieve memory during testing, and improved virtual positioning models (VLM) mean that virtual positioning arrays (VLA) have better spatial understanding and can significantly enhance throughput in data annotation and processing workflows. The market in 2025 will initially experience the capabilities of zero-shot mapping, visual sensitivity, and general physical reasoning brought by data scale, with the physical AI data scale expected to increase 100 times in 2026.