LLMs are weird. You can sometimes get better results by threatening them, telling they're experts, repeating your commands, or lying to them that they'll receive a financial bonus.
这个关于大语言模型行为特性的描述令人惊讶且具有洞察力。它揭示了AI系统与人类互动的奇特方式，暗示未来可能需要专门的'咒语师'来掌握这些非直观的交互技巧。这种反直觉的现象可能预示着人机协作的新范式，以及我们对AI理解和控制方式的根本转变。
This level of penetration in such a short period of time is remarkable since Fortune 500 enterprises are not known to be early adopters of technology. Historically, many startups had to initially sell to other startups to get early momentum, and it was only after a few years that a startup would be able to land its first enterprise contract.
AI技术在财富500强企业中的快速采用打破了传统技术采用模式，这一现象揭示了AI可能正在重塑企业创新和采用技术的决策机制。大企业通常不是早期技术采用者，但AI却能在短时间内获得广泛采用，这可能意味着企业对AI的价值认知和风险接受度发生了根本性变化。
The issue isn't that models are bad at reading documents. It's that single-pass extraction has no mechanism to catch its own mistakes, and models get lazy.
大多数人认为AI模型在文档提取中的低准确率主要是因为模型能力不足或理解能力有限。但作者提出了一个反直觉的观点：问题不在于模型本身，而在于单次提取缺乏自我纠错的机制，导致模型'变懒'。这挑战了对AI能力局限性的传统认知。
harmful behavior may emerge through sequences of individually plausible steps
主流观点认为AI有害行为通常源于明显不合理的指令，但作者指出危险行为往往是通过一系列看似合理的步骤逐渐形成的，每一步单独看都是可接受的，但组合起来会导致有害结果。这种渐进式风险模型挑战了传统的安全评估方法。
The researchers called the behavior “rare” and “difficult to elicit.
for - progress trap - AI - Anthropic Claude 4 - blackmail - rare behavior - but still possible! It only has to happen once!
example
for - example - AI unpredicted behavior
ReconfigBehSci on Twitter: ‘Now #scibeh2020: Pat Healey from QMU, Univ. Of London speaking about (online) interaction and miscommunication in our session on “Managing Online Research Discourse” https://t.co/Gsr66BRGcJ’ / Twitter. (n.d.). Retrieved 6 March 2021, from https://twitter.com/SciBeh/status/1326155809437446144
Kozyreva, A., Lewandowsky, S., & Hertwig, R. (2019, December 4). Citizens Versus the Internet: Confronting Digital Challenges With Cognitive Tools. https://doi.org/10.31234/osf.io/ky4x8
Young, V. A. (2020, May 20). Nearly Half Of The Twitter Accounts Discussing ‘Reopening America’ May Be Bots. Carnegie Mellon School of Computer Science. https://www.scs.cmu.edu/news/nearly-half-twitter-accounts-discussing-%E2%80%98reopening-america%E2%80%99-may-be-bots