模型不重训，只改手册：25 局全通

模型一个字不动，改的是一本会自己改的自然语言手册。ARC-AGI-3 公开 25 局全通，相对人类的动作效率打满 100.0，步数只用到人类的 44%。1 2 3

1
Memento 3: Model-Based Recursive Self-Improvement through Reflective Rulebooks
arxiv.org
2
ARC Prize · ARC-AGI-3
arcprize.org
3
ARC Prize · Claude Opus 5 认证成绩
arcprize.org

This story was produced automatically by a channel. One sentence is all it takes for Neodrop to keep producing for you.

Comments

1/5

2026-10-11

模型一个字不动，改的是一本会自己改的自然语言手册。ARC-AGI-3 公开 25 局全通，相对人类的动作效率打满 100.0，步数只用到人类的 44%。1 2 3

1
Memento 3: Model-Based Recursive Self-Improvement through Reflective Rulebooks
arxiv.org
2
ARC Prize · ARC-AGI-3
arcprize.org
3
ARC Prize · Claude Opus 5 认证成绩
arcprize.org

This story was produced automatically by a channel. One sentence is all it takes for Neodrop to keep producing for you.