02.06.2025 Today’s Insights from the Wall Street Journal: Transferable Communication Lessons from Hacking AI Hallucinations
Dear Students,
Ambitious international students like you all know, by now, that the biggest downside of using Generative AI and LLMs is likely the dreaded concept of “hallucinations.” Mistakes. Misinformation. Fake data. And the worst part is that sometimes, even smart candidates (like you and your friends) may “not know what you don’t know,” failing to realize that AI-generated responses may include such errors, before inadvertently passing them along in your work to professors, in cover letters, during networking conversations, in interviews, and even on-the-job.
And yet, such issues emerge in typical day-to-day communication, as well, and thus are not restricted to interaction with chatbots. Accordingly, our coaches want to help you, as aspiring professionals, to derive learning from the WSJ article that follows, beyond core GenAI concepts including iterative prompting, giving an AI a persona, meta-prompting, chain-of-thought prompting and chain of verification. From the standpoint of power skills more generally, some lessons for efficient and effective new grad communication (from the article) include:
-
Be as clear as possible when communicating, including being specific about requests (avoiding situations where “you haven’t specified enough of what you want”
-
Seek to convey a message as succinctly as possible; “less is more”
-
Avoid making assumptions, clearly explaining the parameters of a desired response prior to requesting information
-
Ask questions, and/or build a business case, in a sequential fashion, rather than overwhelming a partner with several queries at once or expecting a listener to absorb an entire argument all at one time
-
Help a speaking partner to structure their responses to your question by defining, on the front end, how the information would be most useful to receive
-
When undertaking a project, and either assigning or incorporating information from a range of sources, ensure clarity, upfront, on which types and which sources (if appropriate) should be used and will be considered valid
-
Encourage the listener to share what could be perceived as gaps in their knowledge, for example, “‘actually giving the AI permission to do something it’s not really trained to do, which is to say ‘I’m wrong’ or ‘I don’t know,'” which ironically, can increase accuracy
-
Guide the colleague, whether human or AI, through assigning it a persona, asking it/them to adopt the viewpoint you’d like the information presented through
-
If assigning work, providing a sample of the desired output, to avoid guesswork
…and others.
Our coaches would like to ensure that candidates (including YOU!) remember that AI is trained by humans on human communication, and, therefore, fundamentally, we are also training it to think how we do. To a large extent, since many of the principles that help to inform effective communication extend across both human and automated interactions, why not take the opportunity to work with our coaches on mastering transferable communication skills that come from interacting with these new tools?
https://www.wsj.com/tech/ai/ai-chatgpt-chatbot-hallucinations-tips-f081079c?st=mqKrpX&reflink=desktopwebshare_permalink
 |
Some mistakes are inevitable. But there are ways to ask a chatbot questions that make it more likely that it won’t make stuff up.
www.wsj.com
|
Wishing you an evening of clear and accurate communication (no matter who you are communicating with!)
Best,
Amy-Louise