Caixin reported on November 13 that Xu Tiantian, head of finance and ecosystem at Moonshot AI, anticipates artificial intelligence agents will be capable within three to four months of autonomously executing tasks lasting several days or even weeks—tasks that are verifiable. She noted that currently, such tasks typically last between four and eight hours.
Speaking at the 17th Caixin Summit in Beijing on the 12th, Xu stated that AI capabilities continue to evolve at an accelerating pace, with no sign of approaching a ceiling. She summarized the advancement as the cumulative effect of three overlapping trends: continued growth in pre-training factors including computing power, energy efficiency, and model parameters; improvements in post-training and inference capabilities enabling large models to generate chain-of-thought reasoning; and increasing complexity in agent architectures and their control frameworks, which collectively expand performance boundaries. She also observed that enterprise workflows and institutional design are progressing at a pace far slower than the rate of model iteration.
Original source: toutiao.com/article/1881696574693384/
Disclaimer: The views expressed in this article are those of the author alone.