Event ·Multi-source

Anthropic发现Claude内部推理空间J-Space，可读可操控

Anthropic新研究发现Claude内部存在一个名为J-Space的全局工作空间，与链式推理或草稿板不同，它是训练中自发出现的推理机制。通过J-Space，研究者可以直接读取和修改模型内部的信息流动。实验表明，该方法首次实现了对模型推理过程的直接观测和操控，而非仅从输出文本推测。这为可解释性提供了新工具，可能用于验证模型行为、实施更好护栏和预测危险场景。

Current read

Anthropic找到了Claude思考时的内部工作区J-Space，能看能改，比以往猜输出靠谱多了。

11 sources88° AI heatLast updated 2026/7/6 22:47:59

Evidence

11 sources

Primary sourceelvis

2026/7/6 22:47:59View source

Related source 1Anthropic

2026/7/6 17:34:58View source

Related source 2berryxia

2026/7/7 00:00:24View source

Related source 3歸藏(guizang.ai)

2026/7/7 01:49:58View source

Related source 4小互

2026/7/7 03:41:13View source

Related source 5AI Will

2026/7/7 03:47:11View source

Related source 6IT之家

2026/7/7 06:41:52View source

Related source 7Decoder

2026/7/7 14:46:29View source

Related source 8量子位

2026/7/7 01:38:18View source

Related source 9shao__meng

2026/7/6 00:53:52View source

Related source 10lmarena.ai

2026/7/7 13:24:50View source

Conflict check

Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.

Open the story