Entity · Technology
Generative AI
Artificial intelligence systems that generate text, images, code or other content from input.
- Number of articles
- 6 articles
- First published
- First published
- Latest update
- Latest update
Latest coverage
为什么 AI 总在长任务里跑偏：多步任务的失败点在哪
长任务失败往往出在重复执行的稳定性：模型需要把同一件简单事连续做几十次，错误会在过程中累积。本文用 ICLR 2026、NeurIPS 2025 的一手实验拆出三个具体失败点：错误的自我强化、多轮目标漂移、验收环节缺位，并给出按步数预算切分任务、回滚清除错误历史、用 pass^k 验收等六个可落地做法。
上下文越长，答案越不准：资料该喂到什么程度
模型的可用上下文远短于标称上下文，退化也会在窗口用满之前逐步出现。五份独立测量揭示了长上下文失效的机制；成本、延迟和资料筛选共同决定一次调用该放多少内容。
AI 幻觉：哪些输出必须核验
幻觉会持续出现，背后有训练与评测规则的激励。文章结合一手研究，解释其成因和可检测程度，并按“错误能否被发现、能否撤回”整理核验清单。
Why the Same Question Produces Different Answers
Different answers to the same question are not always just randomness. This article separates generation, conversation context and version changes, and shows which differences need checking.
Before You Delegate to AI, Decide What Must Stay Human
Starting with an agentic system’s ability to act in the external world, this article explains how consequences, permissions, data sensitivity and recoverability define delegation boundaries—and how approval, responsibility red lines, least privilege, traceable actions and rehearsed recovery constrain AI action.
How AI Actually Improves Productivity at Work: By Cutting Waiting, Switching and Rework
Drawing on research in writing, customer support, workplace collaboration and software development, this article explains when AI speeds direct tasks, when it shifts time into verification and rework, and how teams should measure net efficiency.