OpenAI消费者产品负责人Adam Fry近日在接受科技媒体Big Technology采访时披露了ChatGPT的一系列最新运营数据。根据他的介绍，ChatGPT每周活跃用户已经达到12亿，过去两个月新增了约2亿用户；其搜索类查询量目前约为Google的五分之一。此外，ChatGPT每周生成的图片达到15亿张，语音输入和语音对话的合计周活跃用户也已经达到2.5亿。
这些数据是在11月2日于纽约举行的一场OpenAI活动期间披露的。Adam Fry表示，ChatGPT的增长速度正在进一步加快。按照他的说法，ChatGPT在过去60天里，周活跃用户从10亿增长至12亿，相当于每个月增加约1亿用户。
其中最值得关注的数字之一，是ChatGPT搜索业务的增长。Fry透露，OpenAI对平台内部搜索类查询进行了统计，并与Google进行比较，发现ChatGPT目前处理的搜索类查询量已经达到Google的约20%。
他解释说，这里的搜索并不局限于传统意义上的关键词检索，而是指用户提出问题后，ChatGPT需要调用网络搜索获取最新信息的情况。每当用户提出一个需要联网搜索的问题，就会被计为一次搜索。
这种变化也反映了用户使用ChatGPT方式的转变。最初，用户主要将其用于写作辅助和编程，而这些任务并不属于传统搜索引擎的典型应用场景。但过去一年，越来越多用户开始通过ChatGPT查询人物、地点、产品以及其他需要实时信息的内容，金融和健康领域的问题也逐渐成为重要使用场景。
Fry认为，ChatGPT与传统搜索引擎最大的区别之一，在于能够保留用户的长期上下文。例如，用户询问巴黎埃菲尔铁塔的高度时，系统不仅可以回答具体数字，还能够结合用户此前提到的巴黎旅行计划，提供更符合实际需求的信息。这种持续理解用户背景和目标的能力，使ChatGPT逐渐从一次性问答工具转变为能够参与长期任务的助手。
语音功能同样取得了显著增长。目前，ChatGPT语音输入和语音对话的合计周活跃用户达到2.5亿，约占整体周活跃用户的五分之一。Fry表示，语音已经不再只是用户偶尔尝试的新奇功能，而是逐渐成为日常使用方式。有些用户只是通过语音快速提出一个问题，另一些用户则会进行持续时间较长的对话。
他将这一增长归功于新一代语音模型带来的交互体验改善，尤其是更自然的双向交流能力。系统能够在对话过程中作出适当回应，也能够处理用户打断和补充信息的情况。此外，后台语音模式让用户即使关闭应用或在家中走动，也能够继续进行对话。
在个人智能助手方面，Fry也谈到了OpenAI于9月29日推出的Dots。与传统聊天机器人不同，Dots被定位为能够持续运行、主动提供帮助的个人助手。例如，用户可以要求它在每天结束时提醒自己是否遗漏了重要事项，系统能够结合邮件和消息等信息，帮助用户回顾当天的待办任务。
不过，OpenAI目前对Dots的权限管理仍然相当谨慎。Fry表示，个人助手需要代表用户处理具有实际意义的事务，因此用户必须对它建立足够的信任。相比单纯回答问题，让AI代替用户执行操作涉及更高程度的责任和风险。
他还承认，Dots目前属于计算资源消耗较高的产品。OpenAI希望先确保其能够提供足够好的服务，再逐步优化运行成本。随着新一代模型不断提高效率，未来有望在维持服务质量的同时降低计算需求。
对于Dots的商业模式，Fry表示，OpenAI正在通过订阅、广告和额外使用额度等多种方式建立收入来源。他透露，ChatGPT此前公布的付费订阅用户已经超过5000万，此后这一数字又有了相当明显的增长。广告业务则在推出后的200天内实现了超过10亿美元的年度经常性收入（ARR）。
在产品战略方面，Fry还确认，OpenAI计划将ChatGPT目前的Chat和Work两种模式整合为一个统一界面。
他承认，要求用户在Chat和Work之间作出选择，已经成为产品使用过程中的一个困惑点。用户往往并不清楚什么时候应该使用普通聊天，什么时候又应该切换到能够执行复杂任务的Work模式。
OpenAI希望通过模型自身判断用户意图，自动区分简单问答与需要长时间执行的复杂任务，从而取消这种模式切换。Fry表示，这项调整正在推进，预计很快就会推出。
图像生成也是ChatGPT的重要增长领域。目前，用户每周通过ChatGPT生成的图片已经达到15亿张。Fry表示，最初用户主要将图像生成功能用于娱乐，例如制作复古风格的个人照片和家庭海报，但如今应用范围已经扩展到工作场景，包括制作餐厅菜单、宣传海报、生日海报以及各种可视化图表。
OpenAI还在持续降低图像生成的响应时间，希望通过提升速度进一步扩大使用场景。对于用户反映AI生成的餐厅菜单和标牌风格过于相似的问题，Fry表示认同，并承诺将相关意见反馈给产品团队。
他特别看好交互式界面的发展方向。Fry认为，过去几年AI聊天产品主要依靠文字进行交互，但未来将逐渐转向更加丰富的视觉体验。例如，用户通过语音询问越位规则时，系统应该能够立即生成示意图；如果用户仍然无法理解，还可以根据后续反馈重新调整图示。
他认为，这种能够随着对话不断更新的交互式内容，将在教育和知识学习领域发挥重要作用。
最后，谈到公众对人工智能的担忧时，Fry承认，AI正在改变工作的组织方式，这种变化对很多人而言确实令人不安。但他认为，随着更多人亲自使用AI，了解它在日常生活中能够提供的帮助，公众对这项技术的认识也会逐渐发生变化。
从这些数据和产品规划来看，OpenAI正在推动ChatGPT从传统聊天机器人向综合性个人助手转型。搜索、语音、图像生成和自动化任务逐渐融合，未来用户可能不再需要区分自己是在提问、搜索信息，还是要求AI执行一项复杂工作，而只需表达最终目标，由系统自行决定如何完成。
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Adam Fry, the consumer product manager of Open AI, recently released a series of up-to-date operating data on ChatGPT in an interview with Big Technology, a scientific and technical medium. According to his presentation, ChatGPT had reached 1.2 billion active users per week and had added about 200 million new users over the past two months; its search-type queries were currently about one fifth of Google's. In addition, ChatGPT produced 1.5 billion pictures per week and the combined weekly active users of voice input and voice conversations have reached 250 million.
These data were disclosed during an Open AI event held in New York on 2 November. Adam Fry stated that ChatGPT was growing at a further pace. According to him, ChatGPT has grown from 1 billion to 1.2 billion active users per week over the past 60 days, representing an increase of about 100 million users per month.
One of the most noteworthy figures is the growth of ChatGPT search operations. Fry disclosed that Open AI had conducted a statistical survey of the internal search-type queries of the Platform and compared them with Google, which found that ChatGPT was processing approximately 20 per cent of Google ' s search-type queries.
He explained that the search was not limited to the traditional keyword search, but rather referred to the fact that ChatGPT needed to call the network to search for up-to-date information after the user had asked questions. Whenever a user raises a question requiring an online search, it is counted as a search.
This change also reflects a shift in the way users use ChatGPT. Initially, users used it mainly for writing aids and programming, tasks that were not typical applications of traditional search engines. Over the past year, however, a growing number of users have started searching through ChatGPT for figures, locations, products and other content requiring real-time information, and issues in the financial and health areas have become important uses.
Fry believes that one of the greatest differences between ChatGPT and traditional search engines is the ability to retain long-term context users. For example, when a user asks about the height of the Eiffel Tower in Paris, the system is not only able to answer specific figures, but also provides more realistic information in conjunction with the travel plans in Paris that the user mentioned earlier. This continuous understanding of user background and objectives has led to the gradual transformation of ChatGPT from a one-time question-and-answer tool to an assistant who can participate in long-term assignments.
Voice functions have also increased significantly. Currently, the total number of active weekly users of ChatGPT voice input and voice conversations is 250 million, or about one fifth of the total weekly active users. Fry indicated that the voice was no longer just a novel feature that users occasionally tried to use, but was becoming a daily use. Some users simply quickly ask a question through voice, while others engage in longer-term dialogue.
He attributed this increase to the improved interactive experience brought about by the new generation of voice models, particularly the more natural two-way communication capability. The system is able to respond appropriately in the course of the dialogue and can also deal with cases where users interrupt and supplement information. In addition, the back-office voice mode allows users to continue dialogue even if they close their applications or move around at home.
In terms of personal intelligence assistants, Fry also spoke of the Dots launched by Open AI on September 29. Unlike traditional chat robots, Dots is positioned as personal helpers capable of running on a continuous basis. For example, users can ask them to remind themselves at the end of each day whether important matters have been omitted, and the system can help users to look back at the to-dos of the day by combining messages such as mail and messages.
However, Open AI is still managing Dots permission with considerable caution. Fry indicated that personal assistants were required to handle meaningful matters on behalf of users, who must therefore build sufficient trust in it. There is a higher degree of responsibility and risk involved in having AI perform on behalf of users than simply answering questions.
He also acknowledged that Dots were currently a product of higher resource consumption. Open AI hopes to ensure that it can provide adequate services before gradually optimizing operating costs. As a new generation of models improves efficiency, it is expected that in the future, while maintaining the quality of services, the need for computing will be reduced.
With respect to Dots ' business model, Fry indicated that Open AI was building revenue sources through subscriptions, advertising and additional use of quota. He disclosed that ChatGPT had previously published more than 50 million subscribers to paid and that the number had since increased considerably. The advertising business achieved more than $1 billion in annual recurrent income (ARR) within 200 days of its launch.
In terms of product strategies, Fry also confirmed that Open AI plans to integrate the current ChatGPT models into a single interface.
He acknowledged that requiring users to choose between Chat and Work had become a confusing point in product use. Users often do not know when ordinary chats should be used and when to switch to a work model capable of carrying out complex tasks.
Open AI wishes to eliminate this paradigm switch by judging user intent through the model itself and automatically distinguishing between simple questions and answers and complex tasks that require long periods of time. Fry indicated that this adjustment is progressing and is expected to be rolled out soon.
Image generation is also an important growth area for ChatGPT. Currently, 1.5 billion images are generated weekly by users via ChatGPT. Fry indicated that while the original users primarily used image generation functions for entertainment, such as the production of retropic-style personal photographs and family posters, their application has now been extended to work areas, including the production of restaurant menus, promotional posters, birthday posters and various visualization charts.
Open AI is continuing to reduce the response time for image generation and hopes to further expand its use by increasing speed. Fry agreed that users were too similar to the menus and signs generated by AI and promised to provide feedback to the product team.
In particular, he looked at the direction of the interactive interface. Fry believes that over the past few years AI chat products have been largely text-based for interaction, but that in the future they will gradually shift to more visual experience. For example, when a user asks through voice overstepping rules, the system should be able to generate a diagram immediately; if the user is still unable to understand it, the diagram could also be adjusted to follow-up feedback.
He believes that this interactive content, which can be updated as the dialogue continues, will play an important role in the field of education and knowledge learning.
Finally, when it comes to public concerns about artificial intelligence, Fry acknowledges that AI is changing the way it organizes its work, a change that is truly disturbing for many. However, he believes that as more people use AI themselves to understand the help it can provide in their daily lives, public awareness of the technology will gradually change.
From these data and product planning, Open AI is promoting the transformation of ChatGPT from a traditional chat robot to an integrated personal assistant. The search, voice, image generation and automation tasks are gradually integrated, and future users may no longer need to distinguish between asking questions, searching for information or asking AI to perform a complex task, but simply to express the ultimate goal and leave it to the system to decide how to do it.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.