I Built Two AIs That Fight Each Other to Test Security Detection Rules. Here’s What I Found.
中文摘要
通过AI对抗测试安全规则，发现规模法则无法预测鲁棒性，且7B模型表现优于更大模型。
English Summary
Testing security via AI combat shows scaling laws don't predict robustness, and a 7B model outperformed larger ones.
Original Excerpt
Scaling laws don’t predict adversarial robustness — and a 7B model outperformed everything larger Continue reading on Medium »