Anthropic 推出 'Dreaming' 功能以輔助智能體反思
anthropic 在開發者大會上發布了名為 'dreaming' 的研究預覽功能。該功能通過分析智能體的活動日誌來識別模式並提升性能。它與 '記憶' 系統協同工作，在不同會話間精煉數據。公司將此與更廣泛的 ai 擬人化趨勢聯繫在一起。
Anthropic announced 'dreaming', a feature within its agent infrastructure that processes transcripts of completed tasks to extract insights for performance improvement. This works in conjunction with a 'memory' system to capture learnings and update agent knowledge between sessions. The naming continues an industry trend of anthropomorphizing AI capabilities, similar to OpenAI's 'reasoning' models and 'thinking' time.