A fuzzer for the Toy Optimizer2 months agohttps://bernsteinbear.com/blog/toy-fuzzer/?utm_source=rss作者构建了一个模糊测试器，用于自动发现玩具编译器优化器中的正确性漏洞。使用三个参数上的加载、存储和转义生成随机程序。验证不变性：在非别名和完全别名条件下，优化和未优化的程序必须产生相同的堆。当禁用一项关键优化（移除别名写入）时，模糊测试器迅速捕获了一个错误。局限性包括未处理部分别名问题，以及需要更复杂的等价性定义来移除分配。
<antirez>2 months agohttp://antirez.com/news/163Proof of work relies on guaranteed success with enough computational resources, unlike bug detection in code.Bug detection in LLMs is limited by model intelligence, not just computational effort, due to saturations of code states and sampling paths.The OpenBSD SACK bug demonstrates that weak models cannot find it; they only hallucinate or pattern-match without true understanding.Future cybersecurity will depend on better models and faster access, not merely more GPUs.