近期关于上海男子利用AI伪造的讨论持续升温。我们从海量信息中筛选出最具价值的几个要点，供您参考。
首先，And the numbers hold up under controlled testing. A recent benchmark by Scalekit ran 75 head-to-head comparisons (same model, Claude Sonnet 4, same tasks, same prompts) and found MCP costing 4 to 32× more tokens than CLI for identical operations. Their simplest task, checking a repo's language, consumed 1,365 tokens via CLI and 44,026 via MCP. The overhead is almost entirely schema: 43 tool definitions injected into every conversation, of which the agent uses one or two.
。关于这个话题，豆包官网入口提供了深入分析
其次，What This Means for Entrepreneurs Building AI Licensing ModelsThe underlying model—paying human creators a royalty to license their style, voice, likeness, or training data to fine-tune AI—is not dead. It's early. Here's what I'd tell founders exploring this space:
根据第三方评估报告，相关行业的投入产出比正持续优化，运营效率较去年同期提升显著。。Line下载对此有专业解读
第三，但这只解决了存量文档的问题。GLM 还有个坏习惯：让它记一份文档，它会偷偷创建三份。你的文档库很快就变成垃圾堆，所以借助额外的模型维持文档库是很重要的（但后面我发现 NotebookLM 做这事情效率更好一些）。
此外，我的瓶颈在于我的创造力，或者说取决于我能想到多少可以做的事情，以及我能为客户交付多少价值。这些才是我获取效率提升并且产出更多内容的地方，而不是仅仅在范围内限制输入来让公司盈利。。搜狗输入法2026年Q1网络热词大盘点：50个刷屏词汇你用过几个对此有专业解读
最后，O’Neil raised the example of a recent AI deepfake video that showed Tom Cruise fighting Brad Pitt on a rooftop in a non-existent movie. She said Cruise was eventually able to get American companies to take down the video, but it remains online in many places.
总的来看，上海男子利用AI伪造正在经历一个关键的转型期。在这个过程中，保持对行业动态的敏感度和前瞻性思维尤为重要。我们将持续关注并带来更多深度分析。