LLM agents

5 atoms · 跨 3 天 · 首见 2026-05-09 · 最近 2026-06-19

三色: 🟦 fact 0 · 🟥 take 5 · stance ▲3/▼2/◆0

来源: X 5

时态: aging:3 · fresh:2

标签: 好观点:5 · 好思考:3

叙事 (narrative) (5)

• 2026-05-09 [老化中] · 重复重写自身经验为文字教训导致记忆变差
  when LLM agents repeatedly rewrite their own experiences into textual “lessons,” their memory can get worse, not better.
  tags: 好观点·好思考 · → daily
• 2026-05-09 [老化中] · 强制整合常将有用经验退化为错误或过度泛化的记忆
  forced consolidation often degrades useful experience into faulty or overgeneralized memories.
  tags: 好观点·好思考 · → daily
• 2026-05-09 [老化中] · 以例子作为记忆对象比提炼为可复用惯例效果好得多
  models seem much better at managing examples as memory objects than at distilling them into reusable routines.
  tags: 好观点·好思考 · → daily
• 2026-06-19 · invent tasks during free-form play, write code-as-policy, build reusable skill library without gradients or RL
  LLM agents invent tasks during free-form play, write code-as-policy, and build a reusable skill library without any gradients or RL.
  tags: 好观点 · → daily
• 2026-06-18 · deployment in daily life
  LLM agents should be everywhere: hospitals, supermarkets, schools, stores, city halls, every boring corner of daily life. Instead, outside our devices, they’re basically invisible
  tags: 好观点 · → daily
  value: direction=absent

时间轴 (近 20)

• 2026-06-19 · narrative · invent tasks during free-form play, write code-as-policy, bu · → daily
• 2026-06-18 · narrative · deployment in daily life · → daily
• 2026-05-09 · narrative · 重复重写自身经验为文字教训导致记忆变差 · → daily
• 2026-05-09 · narrative · 强制整合常将有用经验退化为错误或过度泛化的记忆 · → daily
• 2026-05-09 · narrative · 以例子作为记忆对象比提炼为可复用惯例效果好得多 · → daily

← 实体目录 · 系统日志