https://txtmix.com/categories/%E8%AE%BA%E6%96%87%E8%A7%A3%E8%AF%BB/Recent content in 论文解读 on Text MatrixHugozh-cnTue, 21 Jul 2026 20:06:14 +0800https://txtmix.com/posts/tech/arxiv-2606-18208-looped-world-models-adaptive-depth-scaling/Wed, 01 Jul 2026 16:32:00 +0800https://txtmix.com/posts/tech/arxiv-2606-18208-looped-world-models-adaptive-depth-scaling/<h2 id="译序为什么-loopwm-值得完整读">译序：为什么 LoopWM 值得完整读</h2> <blockquote> <p>&ldquo;Current world models face a fundamental tension: faithful long-horizon simulation demands deep computation, but deeper models are expensive to deploy and prone to compounding errors.&rdquo;</p> <p>—— Looped World Models, arXiv 2606.18208, 2026-06-17</p></blockquote> <p>如果你做过世界模型（World Model, WM），你一定被两件事反复折磨过：</p>https://txtmix.com/posts/tech/arxiv-2606-26978-code-execution-cost-effectiveness-llm-program-repair/Tue, 30 Jun 2026 20:56:00 +0800https://txtmix.com/posts/tech/arxiv-2606-26978-code-execution-cost-effectiveness-llm-program-repair/<h2 id="译序为什么这篇-issta-2026-论文值得完整读">译序：为什么这篇 ISSTA 2026 论文值得完整读</h2> <p>论文标题就一句话：<strong>To Run or Not to Run</strong>——LLM 编程 Agent 是否应该默认执行代码？</p> <p>这是 2026 年所有 AI 编程助手（Claude Code / Codex / OpenCode / Cursor / OpenHands / SWE-agent）都默认开启的能力。你问任何 AI 工程师，TA 都会说&quot;执行测试反馈是 agent 修复 bug 的关键&quot;。</p>