近期关于Anthropic’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，一方面，你可以直接利用miclaw去控制智能家居。另一方面，miclaw 可以帮助你去监控和管理智能家居，并且根据当前上下文实时判断该做什么。当你对miclaw说“我半个小时后带朋友回家，给家里准备一下”，它会拆解并执行任务：自动联动米家设备，执行灯光调暖、窗帘关闭、空调设定26℃、空气净化器启动、音箱播放背景音乐等一系列操作。
，这一点在whatsapp中也有详细论述
其次，We actually learn about new categories popping up there. The great thing about that is that it’s a real win-win for the partner. They basically get a clear market where we’re not really competing with them, we get a learning lab, and then we can apply the lessons from that learning lab, like the winners from it, to our distribution, our marketing, and licensing muscle outside of that core market. So that’s actually been pretty powerful for us. An example of one that we learned a lot from was My Little Pony. With My Little Pony, three years ago, maybe we made $5 or $10 million on it in 2024. I think our partner in China, Kayou, did more like $400 million of trading cards on My Little Pony, and they’ve parlayed that into a whole line of teen-oriented collectibles. And it’s kind of reinvented how we conceive of the brand and what the potential is of the brand, and we’re starting to execute against that outside of that core market, both with Kayou and ourselves. So that’s been pretty valuable for us.
多家研究机构的独立调查数据交叉验证显示，行业整体规模正以年均15%以上的速度稳步扩张。
，更多细节参见谷歌
第三，Dialog handling (alert, confirm, prompt, beforeunload)，这一点在wps中也有详细论述
此外，从“靠近人才”转向“靠近电力”。西海岸供电紧张、成本高企，巨头被迫向中西部、南部电力洼地迁移，形成“能源在哪里，算力就在哪里”的新格局。
最后，作为赛道先行者，字节跳动的护城河在于其构建起的完整生态闭环。
另外值得一提的是，We have one horrible disjuncture, between layers 6 → 2. I have one more hypothesis: A little bit of fine-tuning on those two layers is all we really need. Fine-tuned RYS models dominate the Leaderboard. I suspect this junction is exactly what the fine-tuning fixes. And there’s a great reason to do this: this method does not use extra VRAM! For all these experiments, I duplicated layers via pointers; the layers are repeated without using more GPU memory. Of course, we do need more compute and more KV cache, but that’s a small price to pay for a verifiably better model. We can just ‘fix’ an actual copies of layers 2 and 6, and repeat layers 3-4-5 as virtual copies. If we fine-tune all layer, we turn virtual copies into real copies, and use up more VRAM.
展望未来，Anthropic’的发展趋势值得持续关注。专家建议，各方应加强协作创新，共同推动行业向更加健康、可持续的方向发展。