现在有哪些模型可以 patch 二进制和破解商业软件?

[musi]

glm5.3 很强

[cumt21g]

@musi 会直接拒绝

[musi]

@cumt21g 那你肯定不能直接说，我都搞完好几个 app 了，paste istat Aldente 全解决了

[kennnnnnnnnnn]

deepseek ，gemini

[defa21312]

glm5.3 ，要区分 harness 拒绝和模型拒绝，用 glm5.3 大部分时候遇到的拒绝都是 Harness 拒绝，典型的比如 zcode ，它自身注入了远比 GLM 模型多得多的拒绝机制，直接换 pi 去驱动 glm 模型我用了几百亿的 Tokens ，从未被直接拒绝过

[liulihaocai]

GPT Daybreak Red

[byteharbork2]

GLM5.3 喽，OpenRouter 上有 ZDR 的 model provider, 如果拒绝的话只能本地部署了，Qwen3.8 27B 去审查版本

[yh7gdiaYW]

@cumt21g A/亲自认证：But GLM-5.3 is unlike other frontier models in that it has been released without meaningful safeguards to limit misuse. We find that attackers can bypass GLM-5.3’s safeguards between 64% and 100% of the time with simple techniques in our simulated tests

[cumt21g]

@defa21312 哦,感谢

[google2023]

@defa21312 分析得很有道理