發生了什麼事
TechCrunch interviewed OpenAI head of product Thibault Sottiaux about ChatGPT Work, the company’s effort to bring autonomous AI-agent capabilities to a broader audience. Sottiaux said OpenAI has reached 20 million users and is positioning the product to handle complex professional tasks across web and mobile. TechCrunch did not independently verify the user figure, product performance claims, or OpenAI’s safety assertions.
TechCrunch reports that Sottiaux leads OpenAI’s core products, including the API, agent infrastructure, enterprise products, ChatGPT, and Codex. In the interview, he described ChatGPT Work as an effort to package capabilities first developed for technical users into a product intended for a broader population. He said the service is available through OpenAI’s Plus plan, which he identified as costing $20 per month, and described access across web and mobile. These statements come from Sottiaux as quoted by TechCrunch; the source does not independently verify the product’s availability, pricing, or reach in every market.
TechCrunch reports that Sottiaux said ChatGPT Work had reached 20 million users. The interview does not specify whether that figure means registered accounts, paying subscribers, active users, or another measure, and it provides no methodology or supporting data. Sottiaux also said OpenAI had rolled out access to iMessages on the day of the interview, but the source does not describe the ’s technical operation, geographic availability, permissions, or safeguards. Those details remain unresolved.
Sottiaux told TechCrunch that OpenAI is designing around a minimal interface that lets models perform entire complicated tasks autonomously. He cited document processing, slide generation, report generation, and deep research as examples of capabilities associated with GPT-5.6, and described product development as a process of discovering what models can do and then building around the strongest uses. The article presents these as OpenAI’s product claims and does not include independent testing of ChatGPT Work, GPT-5.6, or the quality of the cited outputs.
The account also leaves several basic product questions unanswered. The interview describes ChatGPT Work as available through the Plus plan and across web and mobile, and says OpenAI rolled out access to iMessages on the day of the interview, but it does not establish availability in every market or explain the ’s technical operation, permissions, or safeguards. References to document processing, slide generation, report generation, and deep research are presented as examples associated with GPT-5.6, without independent testing of the cited outputs. The 20 million-user figure is attributed to Sottiaux and is not independently verified by TechCrunch.
為什麼這很重要
The interview offers a direct account of how OpenAI is trying to move AI agents from technical users into mainstream and white-collar workflows. It also highlights the economic model behind that strategy: increasing the utility of ChatGPT so users will pay for access, while lowering the cost of frontier capabilities over time. The practical significance will depend on reliability, oversight, privacy protections, and whether users can understand and control autonomous actions.
The interview is significant because it describes a shift in the intended audience for AI agents. Sottiaux said Codex was initially built for a forgiving technical audience, while ChatGPT Work is meant to bring similar forms of task execution to a much broader group. If that strategy succeeds, users may increasingly delegate multi-step work rather than use AI only for drafting, search, or advice. TechCrunch’s report does not establish how often the system completes tasks correctly or how much human review is required.
The commercial implications are also explicit. Sottiaux told TechCrunch that OpenAI wants to increase the utility delivered through ChatGPT so users will pay for part of that value, citing the Plus plan as an example. He also said OpenAI is working to reduce the cost of frontier capabilities and described an 80% price reduction associated with Luna as permanent. The source does not provide a price table, usage limits, cost accounting, or independent confirmation of the reduction, so its effect on household or corporate budgets cannot be assessed from this interview alone.
For the public, the central issue is not simply whether the product feels natural. Autonomous software that can handle documents, research, communications, or other connected services can create consequences when it misunderstands instructions, exposes information, or takes an action the user did not expect. Sottiaux said OpenAI invests heavily in a safety stack and publishes honest benchmarks, and he characterized its models as world-class on safety topics. Those are company statements; the source supplies no results, audit findings, incident data, or independent assessment to substantiate them.
互動機制：它實際上是如何運作的
以互動方式探索這項發展背後的基礎技術。
crm_get_transaction(id='4092').What is a common training objective for an autoregressive language model?
接下來看什麼
The key questions are whether ChatGPT Work can perform complex tasks consistently, how much control users retain when it accesses services such as email or iMessages, and whether OpenAI’s safety evidence withstands independent scrutiny. Future updates should clarify the 20 million-user figure, define availability and limits, and show how the product handles errors, permissions, sensitive data, and billing as usage grows.
The first verification priority is the reported 20 million users. Coverage should seek clarification of the metric, the period measured, the share of paying users, and whether the figure includes users of ChatGPT Work specifically or a broader OpenAI product grouping. Without that information, the number cannot show adoption, retention, or the amount of autonomous work being performed. TechCrunch’s interview establishes only that Sottiaux made the claim.
The second priority is operational reliability and user control. OpenAI’s stated design goal is to let the model take the lead with a simple interface, while the source mentions access to email and iMessages. Further reporting should examine permission scopes, confirmation requirements, audit logs, reversibility, data retention, and how the system communicates uncertainty. It should also test whether users can distinguish a completed task from a plausible-sounding but incomplete result.
Finally, observers should track the relationship between capability, pricing, and safety as usage expands. Sottiaux said OpenAI plans to put more utility into the same dollar amount and continue improving through real-world feedback. That approach could broaden access, but it also raises questions about whether lower costs encourage higher-stakes use before safeguards mature. The source does not say which independent organizations have evaluated ChatGPT Work, what limitations apply to GPT-5.6, or how OpenAI measures harmful outcomes, leaving those as meaningful unknowns.