ChatGPT — a chatbot from Open AI that uses online data to produce conversational responses — is not the most effective executive assistant, The Washington Post reported Feb. 2.
The chatbot is essentially a predictive text system, and although responses read naturally, they may contain errors like plagiarism or contradictions. This became clear when the Post asked the bot to perform five common workplace tasks:
1. Responding to messages: The bot produced wordy responses and left some words or phrases in brackets if it did not understand what they were referring to. It also made assumptions about the reporter's job that were not included in the prompt. However, it was able to produce sarcasm in one response.
2. Idea generation: When asked to come up with new story ideas, the bot had trouble with originality. It produced prompts that had already been written and centered on old ideas. ChatGPT has limited knowledge of anything after the year 2021, so some of its ideas are outdated or incorrect, according to the Post.
3. Navigating tough conversations: ChatGPT provided a good springboard for difficult workplace conversations, such as asking a coworker to lower his volume on phone calls. It was polite and gentle when addressing the problem, and offered a chance to discuss further. But the responses were, again, wordy and formulaic, and in one instance it completely misunderstood the prompt.
4. Team communications: The answers sounded impersonal and the bot did not effectively change its tone when asked to be more upbeat. It used the wrong pronouns to refer to someone, giving away that the memo was written by someone other than the sender.
5. Self-assessment reports and cover letters: The bot listed generic accomplishments, such as "I am always willing to go the extra mile," and, "I am proud of the contributions I have made." Overall, the results served as good rubrics, the Post concluded.