Primary sourceSimon Willison’s Weblog
View sourceEvent ·Multi-source
Anthropic用Claude Mythos发现HAWK和弱AES的数学缺陷
Anthropic研究人员使用Claude Mythos在60小时内发现了HAWK和弱版AES(AES-128 r7)的数学弱点,但两者对当前系统无实际影响。研究花费约10万美元API费用,关键干预是反复用自然语言提示模型“找点值得发表的东西”。研究人员公开了带拼写错误的prompt,例如“no again the goal is that we have highly inteligent model as good top researcher”。实验表明,即使强大模型也需要明确引导才能坚持解决困难问题。
Current read
Anthropic花了10万刀让Claude跑60小时,逼它找出HAWK和弱AES的数学漏洞,还公开了怎么给模型打鸡血的prompt,很有参考价值。
11 sources78° AI heatLast updated 2026/7/28 22:45:37
Evidence
11 sourcesRelated source 1向阳乔木
View sourceRelated source 2Decoder
View sourceRelated source 3Lenny Rachitsky
View sourceRelated source 4IT之家
View sourceRelated source 5Anthropic
View sourceRelated source 6AI Breakfast
View sourceRelated source 7Amjad Masad
View sourceRelated source 8Y Combinator
View sourceRelated source 9elvis
View sourceRelated source 10SuperTechFans
View sourceConflict 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.