花 200 美元叫 Operator 做網路研究：成果很弱
案例摘要
評測者花 200 美元買 ChatGPT Operator，用一週時間丟白話瀏覽器任務：分析設計推文趨勢、彙整 AI 新聞、列 AI 建站工具價格、摘要文章、截圖進共用 Google Doc。結果弱：推文重複又過期、只摘要了一篇 Axios、抄了某 Bing 清單的前十條；多篇文章摘要時在分頁間鬼打牆，截圖功能根本沒法用。
實際結果
成果弱：重複、過期、鬼打牆；截圖失敗。
台灣／大陸場景對照
🇹🇼 台灣
誠實的付費踩雷報告：瀏覽器 agent 做開放式研究目前仍不可靠；付費前先看這類評測，別為 demo 付錢。
🇨🇳 大陸
同理：通用 agent 做深度研究前，先用小任務試水溫；研究類任務建議採「agent 搜集、人工綜合」。
英文原文
Run web research jobs and capture screenshots into a doc
The reviewer paid $200 for ChatGPT Operator and spent a week giving it plain-language browser tasks. These included analyzing trending web-design tweets, summarizing AI news, listing AI website builders with pricing, summarizing a set of linked articles, and screenshotting sites into a shared Google Doc. Operator browsed on its own. The reviewer re-prompted it, and had to take over at logins and CAPTCHAs.
Results were weak. Operator returned duplicate, stale tweets, summarized one Axios article, and copied the first ten items from a single Bing listicle. It looped between tabs on multi-article summaries and could not take screenshots at all.