1 paper · 1 filter
Yiming Xiong, Shengran Hu, Jeff Clune
The statelessness of foundation models bottlenecks agentic systems' ability to continually learn, a core capability for long-horizon reasoning and adaptation. To address this limit…