ChatGPT Work supports SNU research report drafts

On a demonstration screen at OpenAI's Korea office in Samseong-dong, Seoul, an AI agent gathered materials and built up a document draft on behalf of a researcher. At a media demo held on August 20, OpenAI showcased a case in which ChatGPT Work was used not only to automate corporate tasks but also to draft university research reports and collections of materials.

At the heart of ChatGPT Work is OpenAI's Codex coding-agent technology. When users issue instructions in natural language, it uses websites and a built-in browser to conduct research and analysis, then combines the results into spreadsheets, slides and documents. More than 1,400 services and solutions, including Google Drive, Gmail, Salesforce and Databricks, can be connected as plugins. Scheduled tasks can also be used to update project information regularly.

Kilkon Ko, a professor at Seoul National University's Graduate School of Public Administration and director of its information systems office, introduced a research workflow that runs from collecting and structuring country-level statistics, legislation, policy documents and academic materials through data analysis and report drafting. While the agent contributes to drafting reports and helps analyze data, researchers verify sources and facts and focus on interpretation.

The scale is also growing. OpenAI said Codex Business had more than 3 million weekly active users as of April worldwide, and that combined weekly active users for the two services had surpassed 10 million since the launch of ChatGPT Work last month. In South Korea, Samsung Electronics, Samsung Display and Seoul National University are using the services.

So what changes in practice? Researchers and planners can hand off repetitive tasks such as finding materials and standardizing formats in a single workflow, giving them more time for fact-checking and judgment. At the same time, the more internal corporate data is connected, the more important access-rights management becomes. Daeyeol Shim, an engineer at OpenAI Korea, said AI's data-access scope can be limited through prompts, but people remain responsible for verifying the sources and facts in final reports.