跳到主内容
@wquguru
精选88Rohan Paul论文研究

Google DeepMind论文:AI模型无法自主生成真正随机数

New Google DeepMind paper finds that AI models cannot make truly random choices…

原文
发到 X
推荐理由

Agent开发必读,揭示了模型原生随机性的致命缺陷,直接指导如何构建可靠的随机决策链路。

New Google DeepMind paper finds that AI models cannot make truly random choices on their own, but larger models pick correctly when a computer program gives them a random number first.

Google DeepMind 的新论文发现,AI 模型无法自主做出真正的随机选择,但当计算机程序先提供一个随机数时,更大的模型能够正确选择。

Agents often need to pick actions at random with set odds, like checking servers in an unpredictable order. LLMs say they will pick uniformly, then keep landing on favorites like 7, 42, or answer C.

智能体通常需要以设定的概率随机选择动作,例如以不可预测的顺序检查服务器。大型语言模型(LLMs)声称会均匀选择,但实际却总是倾向于选择像 7、42 或答案 C 这样的偏好项。

They asked Gemini, Qwen3, and OLMo models to sample from simple distributions 1024 times each. They also tried other temperature and sampling settings, and gave models a random number to convert.

研究人员让 Gemini、Qwen3 和 OLMo 模型各自从简单分布中采样 1024 次。他们还尝试了其他温度和采样设置,并向模型提供了一个需要转换的随机数。

Every reported statistical test for matching the target odds failed, and changing temperature or top-p did not fix it. In a security game where truly random checks let attackers win 75% of the time, Gemini's picks let them win 96.6% to 99.9%.

所有用于匹配目标概率的统计测试均告失败,且调整温度或 top-p 参数也无法解决问题。在一个安全游戏中,如果真正的随机检查能让攻击者有 75% 的概率获胜,而 Gemini 的选择却让攻击者的胜率高达 96.6% 至 99.9%。

If your agent needs randomness, generate it in code and pass it in, instead of asking the model to be random.

如果你的智能体需要随机性,请在代码中生成随机数并将其传入,而不是要求模型自行产生随机性。

更进一步:量化金融体系

看懂新闻只是起点——沿量化金融路径,把它变成能交付的工程能力

进入量化体系 →

相似阅读

关联信息,但可能不是同一事件