[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"post:12768":3},{"id":4,"product_id":5,"product":6,"user_id":10,"title":11,"content":12,"images":91,"create_at":92,"update_at":92,"status":93,"stats":94,"user":98,"emoji_reacts":106,"video_info":79,"vote":79,"user_remark":96,"user_interesting":105,"user_favorite":105,"topic_ids":110,"_figures":111,"_previewImage":112,"_rawTitle":113},12768,944,{"id":5,"name":7,"slug":8,"avatar_url":9},"千问","qian-wen","https:\u002F\u002Fwatcha.tos-cn-beijing.volces.com\u002Fprod\u002Fuser\u002Fuploads\u002F2_1763195049_8bd9224e-4986-4bf3-b319-29eacdaed17e.png",10018563,"千问 3.8 Max，一个勤奋的执行力选手",{"content":13,"type":90},[14,20,21,25,26,30,31,35,36,40,41,45,46,50,51,55,56,60,61,65,66,70,71],{"content":15,"type":19},[16],{"text":17,"type":18},"用了几天千问 3.8 Max，先说结论：它是个执行力选手，不是创意选手。","text","paragraph",{"type":19},{"content":22,"type":19},[23],{"text":24,"type":18},"我拿它跑了两轮，都是我日常在做的 AI 视频那套流程。",{"type":19},{"content":27,"type":19},[28],{"text":29,"type":18},"第一轮，我让它把一篇经典短篇小说改成视频脚本。没给原文，想试试纯靠它自己的知识储备产出如何。结果它一口气写了七千多字，自动切成十四个板块：改编核心、片长建议、人物处理、旁白文本、镜头风格参考、AI 生成注意事项、推荐片名……",{"type":19},{"content":32,"type":19},[33],{"text":34,"type":18},"态度是真勤奋。但我真正要的那点东西，大概只占它输出的七分之一，剩下六份得我自己动手删。同样的提示词丢给 Sonnet 5，产出精准得多，干净利落。",{"type":19},{"content":37,"type":19},[38],{"text":39,"type":18},"所以似乎问题不在能力，在它不太会判断 \"我到底想要啥\"。你话说得糙一点，它就开始努力地跑偏。",{"type":19},{"content":42,"type":19},[43],{"text":44,"type":18},"第二轮好很多。我把上一轮的稿子接进分镜 Skill，让它按格式拆镜头。它拆出 56 个镜头，总时长五分钟左右，格式该有的都有，颗粒度也够细，一稿过，我基本没改。这种有明确规则、有模板约束的活儿，它做得比我预期好不少。",{"type":19},{"content":47,"type":19},[48],{"text":49,"type":18},"代价是慢。这一轮它整整想了二十分钟。同平台同任务，Sonnet 5 和 Deepseek v4 pro 都是五到八分钟的事。而且我反复试了几次，只要挂了附件，速度拖得特别明显。",{"type":19},{"content":52,"type":19},[53],{"text":54,"type":18},"还有个绕不开的点：似乎有点贵。同类型人物，比 GM5.2 之类的国产模型贵出好几倍。",{"type":19},{"content":57,"type":19},[58],{"text":59,"type":18},"另外必须吐槽一下它的文风。不知道是不是我个人感受，感觉这模型有很重的排版口癖，特别爱换行、爱分段、爱加引用块，读起来像在看现代诗。跟 GPT 那个 \"不是…… 而是\" 的句式病、Claude 那个双破折号一个性质。本来三行能说完的事，它给你摊成半屏，扫读成本一下就上去了。",{"type":19},{"content":62,"type":19},[63],{"text":64,"type":18},"如果你团队已经有成熟的内容工作流，Skill 模板是现成的，只想在 \"落地执行\" 这段接 AI 提效，它挺合适。剧本转分镜、提示词细化这类活儿，交付质量稳定。但你要是指望它从零帮你想创意、直接做改编，大概率得来回澄清好几轮，不如换个模型。",{"type":19},{"content":67,"type":19},[68],{"text":69,"type":18},"我个人的定位：备用。单纯这几个个人工作场景，体感大概在 GPT-5.5 和 Sonnet 4.6 那个区间，能扛一部分执行类负载，但短期内还不会成为主力之一。",{"type":19},{"content":72,"type":19},[73,81,83,87,88],{"attrs":74,"type":80},{"id":75,"label":76,"mentionSuggestionChar":77,"objectType":78,"slug":79},44,"是骡子是马拉出来遛遛 ","#","topic",null,"mention",{"text":82,"type":18}," ",{"attrs":84,"type":80},{"id":85,"label":86,"mentionSuggestionChar":77,"objectType":78,"slug":79},28,"最近的观察",{"text":82,"type":18},{"type":89},"hardBreak","doc","https:\u002F\u002Fwatcha.tos-cn-beijing.volces.com\u002Fprod\u002Fuser\u002Fuploads\u002F10018563_1786163066_062a5606-870a-4f7b-a973-1911dda0b704.png;https:\u002F\u002Fwatcha.tos-cn-beijing.volces.com\u002Fprod\u002Fuser\u002Fuploads\u002F10018563_1786163082_1bac7a8b-72c1-49a4-be6a-7600cc0b2c97.png","2026-08-08T04:26:14.007Z",1,{"upvotes":95,"interests":96,"floors":97,"root_floors":93},12,0,2,{"id":10,"nickname":99,"username":100,"avatar_url":101,"cover_url":102,"bio":103,"is_gold_member":104,"is_official":105},"胡凯尔","hke","https:\u002F\u002Fwatcha.tos-cn-beijing.volces.com\u002Fprod\u002Fuser\u002Fprofile\u002Favatar\u002F10018563_1779507623_user_10018563.png","https:\u002F\u002Fwatcha.tos-cn-beijing.volces.com\u002Fprod\u002Fuser\u002Fuploads\u002F10018563_1768485484_e81a645b-bcd2-4a49-9fba-e08800f41941.png","你好呀 ; )\n🪐 AIGC 野生创作者 | 梦到什么说什么。",true,false,{"reacts":107},[108],{"emoji_code":109,"count":93,"reacted":105},"🤣",[75,85],[],"https:\u002F\u002Fwatcha.tos-cn-beijing.volces.com\u002Fprod\u002Fuser\u002Fuploads\u002F10018563_1786163066_062a5606-870a-4f7b-a973-1911dda0b704.png","千问3.8 Max，一个勤奋的执行力选手"]