OpenAI claims ChatGPT’s new default model hallucinates way less

OpenAI 声称 ChatGPT 的新默认模型“幻觉”大幅减少

The new model, GPT-5.5 Instant, will also use fewer ‘gratuitous’ emoji. 新模型 GPT-5.5 Instant 也将减少使用“无意义”的表情符号。

OpenAI’s newest default model for ChatGPT might not make stuff up as much. Hallucinations have been an ongoing problem for AI models, but OpenAI says its new GPT-5.5 Instant model has “significant improvements in factuality across the board.” OpenAI 为 ChatGPT 推出的最新默认模型可能不再那么容易“胡编乱造”了。“幻觉”一直是人工智能模型面临的长期问题，但 OpenAI 表示，其全新的 GPT-5.5 Instant 模型在“整体事实准确性方面有了显著提升”。

The company claims that, based on “internal evaluations,” GPT-5.5 Instant produced “52.5% fewer hallucinated claims” than its Instant model for GPT-5.3 “on high-stakes prompts covering areas like medicine, law, and finance.” GPT-5.5 Instant also “reduced inaccurate claims by 37.3% on especially challenging conversations users had flagged for factual errors.” (OpenAI has some information about how it evaluated the model in its GPT-5.5 Instant system card.) 该公司声称，基于“内部评估”，在涵盖医学、法律和金融等领域的高风险提示词测试中，GPT-5.5 Instant 产生的“幻觉声明”比 GPT-5.3 的 Instant 模型减少了 52.5%。此外，在用户标记为存在事实错误的特别具有挑战性的对话中，GPT-5.5 Instant 的“不准确声明减少了 37.3%”。（OpenAI 在其 GPT-5.5 Instant 系统卡中提供了有关其如何评估该模型的一些信息。）

OpenAI also claims that GPT-5.5 Instant is “more capable across everyday tasks,” like analyzing image uploads and knowing when to turn to the web for an answer. GPT-5.5 Instant has “tighter and more to-the-point” responses and will avoid using “gratuitous emojis.” OpenAI 还声称，GPT-5.5 Instant 在处理日常任务时“能力更强”，例如分析上传的图片以及判断何时需要联网搜索答案。GPT-5.5 Instant 的回复更加“精炼且切中要点”，并将避免使用“无意义的表情符号”。

With GPT-5.5 Instant, ChatGPT is now “more effective” at pulling in context from things like previous chats and your Gmail to give you more personalized responses, too. (This is a feature that Google is investing heavily in for Gemini as well.) And for all ChatGPT models, a new “memory sources” feature will let the chatbot show what context was used to inform personalized responses, and you can delete or correct information if you need. 借助 GPT-5.5 Instant，ChatGPT 现在能“更有效地”从之前的聊天记录和 Gmail 等内容中提取上下文，从而为你提供更个性化的回复。（这也是谷歌在 Gemini 上大力投入的一项功能。）此外，对于所有 ChatGPT 模型，一项新的“记忆来源”（memory sources）功能将允许聊天机器人展示其使用了哪些上下文来生成个性化回复，用户还可以根据需要删除或更正这些信息。

OpenAI will start rolling out GPT-5.5 Instant on Tuesday to “all ChatGPT users,” though GPT-5.3 Instant will be an option for three months until it’s “retired.” (In the past, users have mourned the loss of older models, so this gives people time to transition.) OpenAI 将于周二开始向“所有 ChatGPT 用户”推送 GPT-5.5 Instant，不过 GPT-5.3 Instant 将继续作为选项保留三个月，直到被“退役”。（过去，用户曾对旧模型的下线感到惋惜，因此这次调整给了人们过渡的时间。）

The enhanced personalization will roll out first to Plus and Pro users on the web and is “coming soon” to the mobile apps. OpenAI has “plans” to bring it “soon” to Free, Go, Business, and Enterprise users. The memory sources feature is rolling out to ChatGPT consumer plans now on the web “and soon on mobile.” 增强的个性化功能将首先向网页版的 Plus 和 Pro 用户推出，并“即将”登陆移动端应用。OpenAI “计划”在“不久后”将其提供给免费版、Go、商业版和企业版用户。“记忆来源”功能目前正向网页版的 ChatGPT 消费者计划用户推送，“移动端也将很快跟进”。