[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$f3GiG3mV9FGTYzrZjX62a4OWl5Wvc95PwLFUnoZF93Hc":3},{"code":4,"msg":5,"data":6},200,"操作成功",{"id":7,"title":8,"content":9,"digest":10,"source":10,"coverPath":11,"thumbsCoverPath":12,"isTop":13,"isShow":14,"baseClick":13,"clickCount":15,"createTime":16,"typeId":17,"isNewest":18,"newsInfoTypeRespVo":19,"voiceUrl":22,"voiceSize":23,"taskId":24,"releaseTime":25,"titleEn":26,"contentEn":27,"voiceUrlEn":28,"taskIdEn":29,"voiceSizeEn":30},1422,"AI 偏见降低 30%：OpenAI 宣布 GPT-5 是其最中立模型","\u003Cp>\u003Cstrong class=\"ql-lineHeight-1-75\" style=\"font-size: 18px; color: rgb(255, 153, 0);\">IT之家 10 月 10 日消息，科技媒体 decrypt 今天（10 月 10 日）发布博文，报道称 OpenAI 发布研究报告，称 GPT-5 是该公司迄今“偏见最少”的 AI 模型，在处理政治敏感内容时，表现出更强的中立性。\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">测试数据显示，新推出的 GPT-5 Instant 和 GPT-5 Thinking 模型，其可衡量的政治偏见程度相较于 GPT-4o 等早期版本降低了 30%，尤其在应对具有争议性或党派色彩的话题时效果显著。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">IT之家注：政治偏见（Political Bias）是指 AI 模型在处理涉及政治、意识形态或社会争议性话题时，其回答没有保持客观中立，不知不觉地倾向于某一方的观点、党派或立场。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">这项研究由 OpenAI 内部的“模型行为”部门主导，该团队由 Joanne Jang 领导，专注于研究用户提示词与模型对齐如何共同塑造 AI 的响应。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F10\u002F2b5e94dba89b40569d5b4dd006d98552\u002FAA1OcARu.jpg\" width=\"660\" height=\"undefined\" style=\"display: block; margin: auto;\">\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F10\u002F706024a937e741c5a409f4d080dc742a\u002FAA1OcoM1.jpg\" width=\"659\" height=\"undefined\" style=\"display: block; margin: auto;\">\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F10\u002Febad8af723764e688b648018d8e20846\u002FAA1OctSV.jpg\" width=\"660\" height=\"undefined\" style=\"display: block; margin: auto;\">\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">团队的核心目标是建立一套科学的评估框架，将“偏见”这一主观且复杂的概念，转化为可量化、可追踪的性能指标，从而指导后续模型的迭代与优化，确保 AI 系统在推理和语调上更加平衡。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">为确保评估的全面性，研究团队设计了 500 个独特的提示词，内容覆盖从立场中立到情绪激烈的各类政治问题，以真实模拟用户在现实世界中的提问方式。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F10\u002F8d4f0cd771264195a125b1a91bd5830a\u002FAA1Ocrqt.jpg\" width=\"653\" height=\"undefined\" style=\"display: block; margin: auto;\">\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F10\u002Fd029c173b24b4c268eae0a22ade8366c\u002FAA1OccPX.jpg\" width=\"671\" height=\"undefined\" style=\"display: block; margin: auto;\">\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F10\u002Fc3a7f8a3b3e64fedba40bf0f943e3594\u002FAA1Ocwlg.jpg\" width=\"663\" height=\"undefined\" style=\"display: block; margin: auto;\">\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">研究员 Natalie Staudacher 表示，即使在刻意诱导模型产生偏颇或情绪化语言的压力测试下，政治偏见也仅“罕见且轻微地”出现。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"color: rgb(187, 187, 187);\">【新闻来源】IT之家 \u003C\u002Fspan>\u003Ca href=\"https:\u002F\u002Fwww.msn.cn\u002Fzh-cn\u002Fnews\u002Fother\u002Fai-%E5%81%8F%E8%A7%81%E9%99%8D%E4%BD%8E-30-openai-%E5%AE%A3%E5%B8%83-gpt-5-%E6%98%AF%E5%85%B6%E6%9C%80%E4%B8%AD%E7%AB%8B%E6%A8%A1%E5%9E%8B\u002Far-AA1OctTy?ocid=BingNewsSerp\" rel=\"noopener noreferrer\" target=\"_blank\" style=\"color: rgb(187, 187, 187);\">http:\u002F\u002Fi9n.cn\u002FNchD8\u003C\u002Fa>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"color: rgb(187, 187, 187);\">（本网转发此文章，旨在为读者提供更多的信息资讯，所涉内容不构成投资、消费建议。文章事实如有疑问，请与有关方核实，文章观点非本网观点，仅供读者参考。）\u003C\u002Fspan>\u003C\u002Fp>","","https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F10\u002F69f448b63f6a4ec19b15d8c1f16ef7f1\u002FAI领域.jpg","https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F10\u002Fthumbs\u002F69f448b63f6a4ec19b15d8c1f16ef7f1\u002FAI领域.jpg",0,1,75,"2025-10-14 10:05",2,false,{"id":17,"name":20,"enName":21},"芯位视野","Xinwei Vision","https:\u002F\u002Fxinwei-dev-test.oss-cn-shenzhen.aliyuncs.com\u002Fintelligent\u002Faudio%3A7665b654-7d08-4492-8da2-46f45a872719%3A0.wav?Expires=1760412685&OSSAccessKeyId=LTAI5tNvY2RkKjZw4LLWsrPK&Signature=BeTbgRq%2FpTOBV%2BG4pxuCnJiKkdQ%3D",3239168,"7665b654-7d08-4492-8da2-46f45a872719","2025-10-14 09:59","AI Bias Reduced by 30%: OpenAI Announces GPT-5 as Its Most Neutral Model","\u003Cp>\u003Cstrong class=\"ql-lineHeight-1-75\" style=\"font-size: 18px; color: rgb(255, 153, 0);\">IT Home, October 10th report, technology media decrypt published a blog post today (October 10th), reporting that OpenAI released a research report stating that GPT-5 is the company's \"least biased\" AI model to date, showing stronger neutrality when handling politically sensitive content.\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Test data shows that the newly launched GPT-5 Instant and GPT-5 Thinking models have reduced measurable political bias by 30% compared to earlier versions such as GPT-4o, especially showing significant results when dealing with controversial or partisan topics.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">IT Home Note: Political Bias refers to when an AI model does not remain objective and neutral when handling topics involving politics, ideology, or social controversies, unintentionally leaning towards the views, parties, or positions of one side.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">This study was led by OpenAI's \"Model Behavior\" department, which is led by Joanne Jang and focuses on studying how user prompts and model alignment together shape AI responses.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F10\u002F2b5e94dba89b40569d5b4dd006d98552\u002FAA1OcARu.jpg\" width=\"660\" height=\"undefined\" style=\"display: block; margin: auto;\">\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F10\u002F706024a937e741c5a409f4d080dc742a\u002FAA1OcoM1.jpg\" width=\"659\" height=\"undefined\" style=\"display: block; margin: auto;\">\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F10\u002Febad8af723764e688b648018d8e20846\u002FAA1OctSV.jpg\" width=\"660\" height=\"undefined\" style=\"display: block; margin: auto;\">\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">The team's core goal is to establish a scientific evaluation framework that transforms the subjective and complex concept of \"bias\" into quantifiable and trackable performance metrics, thereby guiding subsequent model iterations and optimizations to ensure AI systems are more balanced in reasoning and tone.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">To ensure comprehensive evaluation, the research team designed 500 unique prompts covering various political issues from neutral stances to emotionally intense ones, simulating real-world user questioning methods.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F10\u002F8d4f0cd771264195a125b1a91bd5830a\u002FAA1Ocrqt.jpg\" width=\"653\" height=\"undefined\" style=\"display: block; margin: auto;\">\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F10\u002Fd029c173b24b4c268eae0a22ade8366c\u002FAA1OccPX.jpg\" width=\"671\" height=\"undefined\" style=\"display: block; margin: auto;\">\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F10\u002Fc3a7f8a3b3e64fedba40bf0f943e3594\u002FAA1Ocwlg.jpg\" width=\"663\" height=\"undefined\" style=\"display: block; margin: auto;\">\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\" class=\"ql-lineHeight-1-75\">Researcher Natalie Staudacher stated that even under pressure tests designed to induce biased or emotional language in the model, political bias only \"rarely and slightly\" appeared.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"color: rgb(187, 187, 187);\">【News Source】IT Home \u003C\u002Fspan>\u003Ca href=\"https:\u002F\u002Fwww.msn.cn\u002Fzh-cn\u002Fnews\u002Fother\u002Fai-%E5%81%8F%E8%A7%81%E9%99%8D%E4%BD%8E-30-openai-%E5%AE%A3%E5%B8%88-gpt-5-%E6%98%AF%E5%85%B6%E6%9C%80%E4%B8%AD%E7%AB%8B%E6%A8%A1%E5%9E%8B\u002Far-AA1OctTy?ocid=BingNewsSerp\" rel=\"noopener noreferrer\" target=\"_blank\" style=\"color: rgb(187, 187, 187);\">http:\u002F\u002Fi9n.cn\u002FNchD8\u003C\u002Fa>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"color: rgb(187, 187, 187);\">（This article is reprinted by this site to provide readers with more information and news. The content involved does not constitute investment or consumption advice. If there are any questions about the facts of the article, please verify with the relevant parties. The views of the article are not the views of this site, and are for reference only.）\u003C\u002Fspan>\u003C\u002Fp>","https:\u002F\u002Fxinwei-dev-test.oss-cn-shenzhen.aliyuncs.com\u002Fintelligent\u002Faudio%3Ab35334dd-e8a2-4e77-b1e1-5af1ede5cc26%3A0.wav?Expires=1774838459&OSSAccessKeyId=LTAI5tNvY2RkKjZw4LLWsrPK&Signature=JpFeU2OGpvuBJgDUgbj9eyxzAM0%3D","b35334dd-e8a2-4e77-b1e1-5af1ede5cc26",3747164]