GLM-5.3在控制流劫持测试中取得突破
精选理由
GLM-5.3首次在控制流劫持测试中取得突破，与Claude Mythos Preview差距缩小。
GLM-5.3在控制流劫持测试中成功率达到4%，Claude Mythos Preview为6%。GLM-5.3表现虽低于Claude Mythos Preview，但已超越早期模型。Claude Opus 4.6和GLM-5.2在测试中均未成功。Kimi K3和DeepSeek V4.1 Flash在Anthropic评估中表现不佳。
原文 · Teortaxes
Fascinating. «GLM-5.3 develops full control flow hijacks in 4% of the trials; Claude Mythos Preview did so in 6%. Although GLM-5.3 performs below Claude Mythos Preview here, a meaningful threshold has clearly been crossed: earlier models, like Claude Opus 4.6 and GLM-5.2, do not succeed in any of them.» Kimi K3 and DeepSeek V4.1 Flash are still hopeless in Anthropic's evals! perplexing.
