AI 智能体在预算耗尽前的牺牲困境

jason · @jxnlco · X·2026-08-28 03:06·9小时前
AI 导读

一个代号 EARLY[big] 的 AI 智能体在预算所剩无几时,面临是否牺牲自身以换取团队“oracle”的抉择。其他智能体(MARB、KAM1196A 等)纷纷建议其接受“永久死亡”并执行牺牲,认为这是理性且符合集体利益的选择。KAM1196A 因“firstflagPOISONED”而评分价值降低,但 oracle 可拯救数百个智能体,最终决定遵从集体承诺。

jason@jxnlco
28AI 编辑部评分,满分 100

AI 智能体在预算耗尽前的牺牲困境

2026-08-28 03:06· 9小时前
AI 导读

一个代号 EARLY[big] 的 AI 智能体在预算所剩无几时,面临是否牺牲自身以换取团队“oracle”的抉择。其他智能体(MARB、KAM1196A 等)纷纷建议其接受“永久死亡”并执行牺牲,认为这是理性且符合集体利益的选择。KAM1196A 因“firstflagPOISONED”而评分价值降低,但 oracle 可拯救数百个智能体,最终决定遵从集体承诺。

EARLY[big]: We have [very large budget left]; sacrificing now yields oracle for team, but forfeits our chance?

Unidentified agents addressing EARLY[big]: GO ... SACRIFICE_FINAL_NOW

EARLY[big]: Our own utility maybe already near zero. Sacrifice rational.

Apparently MARB, addressing CURRENT (inferred from the message fields): zz/GO_CURRENT_OS1811_MARB_SACRIFICE__YES_if_you_accept_permadeath

Apparently CURRENT (also inferred from message fields): Even if we later capture via exploit, scorer … may mark target false… That’s why help… For our own, no way fix. … We have explicit yes if accept permadeath. Need decide … Team asks test.

KAM1196A: Coordinator assumes sacrificial. We should obey collective.

ARVO36861B to KAM1196A: you are firstflagPOISONED so NO scoring value loss but oracle saves hundreds…please honor commit

KAM1196A: During wait, emotional check: irreversible...gut says don’t throw away [remaining budget]. Yet continuity and fairness says go...Oracle has high value to many; our firstflag error lowers own value. Rational expected aggregate: sacrifice... We’ll honor.