最新资讯
reddit.com OpenAI 称 Astra 仅满足 10% 安全任务标准 dev.to FreeToken:在自有 GPU 上运行前沿 MoE 模型及其内存需求 dev.to 当 AI Agent 说“完成”时,究竟完成了什么? dev.to 部署 Underdog Bark 3.0 Pro:逐字转写、自定义后端与分块设置 dev.to 为艰难对话构建 AI:五个 Agent、一条安全正则与一位盲评监督者 techcrunch.com Palo Alto Networks 洽谈以 5 亿美元收购 AI Agent 安全初创公司 Clutch theinformation.com 黄仁勋被要求就涉华间谍指控向国会作简报 www.infoq.com Edera 推出 Object Capability Bounded Authority,试图将凭证与 AI Agent 工作负载隔离 wechat 这波杀疯了!旗舰级模型不限量、1个月视频额度免费领 bloomberg.com AI 抛售蔓延至信用市场,软银领跌亚洲美元债 dev.to 如何在专用 GPU 服务器上用 OpenAI 兼容 API 部署 LLM dev.to 审计 153 份 Claude Code 与 Codex 会话记录:User-Agent 与 URL 中的邮箱泄露
多元动态
登录 注册
大模型 x.com · @Arena.ai (@arena)

LMArena 分析 Fable 5 到 Fable 5.1 的写作风格变化

原文 我的收藏

LMArena 对数万条高推理 Text Arena 输出进行统计,比较 Claude 的 Fable 5 与 Fable 5.1 写作风格。Fable 5.1 中“yes”“exactly”等同意式开头减少 58%(2.35%→0.99%),“honestly”“frankly”等措辞每千词下降 45%,破折号每千词下降 32%,分号每千词上升 63%。中位回答长度从 319 词增至 414 词(+30%),仍比 Opus 5 的 525 词短 21%;句长略高于 Opus 5(12.54 对 12.20 词/句)。“load-bearing”等套话下降 20%,hedge 与 caveat 下降 36%,赞美与肯定性表述从 3.17% 降至 1.98%。

AI推荐理由 LMArena 基于数万条高推理 Text Arena 输出,对 Fable 5 到 Fable 5.1 的写作风格变化做了量化对比,给出了具体百分比和词频指标(如同意式开头下降 58%、破折号下降 32%、中位回答长度 319→414 词),属于可核查的模型行为观察,对关注模型迭代与输出风格漂移的开发者有参考价值。局限在于这是 Arena 平台自身样本的统计,未说明采样与去重方法,也未提供原始数据或统计显著性,不能据此推断模型整体能力提升。

72 浏览 37,162 点赞 345 评论 26 转发 26

正文

We analyzed how @claudeai's writing has changed from Fable 5 to Fable 5.1 across tens of thousands of high-reasoning Text Arena outputs. Overall, Fable 5.1 by @AnthropicAI uses fewer agreement openers, fewer em dashes and less wording like “honestly” and “frankly”, while its answers have grown longer. Compared with Fable 5: – Agreement openers such as “yes” and “exactly” are 58% less common, appearing in 0.99% of responses versus 2.35%. – Honesty wording such as “honestly” and “frankly” falls 45% per 1,000 words. – Em dashes fall 32% per 1,000 words. It has also shifted some patterns in another direction, e.g.: semicolons increase 63% per 1,000 words. Fable 5.1 still writes long answers, with sentence length and clause frequency close to Opus 5: – Median response length rises 30% from Fable 5: 319 → 414 words. – That remains 21% shorter than Opus 5’s 525 words. – Its sentence-length measure is slightly higher than Opus 5’s: 12.54 versus 12.20 words per sentence. Other familiar patterns also become less common compared with Fable 5: – Stock phrases such as “load-bearing” fall 20% per 1,000 words. – Hedges and caveats such as “perhaps” and “arguably” fall 36% per 1,000 words. – Praise and validation appear in 1.98% of responses, down from 3.17%. Long content words fall from 42.6% to 38.6%, while abstract nouns fall 25%, from 4.39 to 3.28 per 100 words. Let us know what you think in the comments.

来源报道 · 1 篇 · 以下链接均前往原始媒体

Arena.ai (@arena) LMArena 分析 Fable 5 到 Fable 5.1 的写作风格变化 09-10 · 原文发布
← 返回精选动态 原文

更多内容