Teaching critical AI literacy is now an essential aspect of digital literacy whether the purpose is to encourage critical perspectives on AI or to train students for future workplaces where they need to correct AI performance. The output of AI text generators like ChatGPT (also known as large language models, or LLMs) often sounds highly plausible and authoritative even when it is wrong, biased, or empty. The phrases below may help students practice identifying flaws in this polished text. Consider sharing a sample chat session as a web page and then inviting students to use these or similar phrases to annotate the outputs, perhaps using Hypothesis or Perusall.
Note: the quoted phrases followed by asterisks were generated by a large language model as detailed in the acknowledgments.
- This sounds plausible because ______________, but it doesn't really make sense because ______________.
- This sounds good, but it doesn't fit the purpose. What we are really looking for is ______________.
- This doesn't clarify ______________.
- "The AI is providing a surface-level answer without diving into the nuances of ______________."*
- This is biased in favor of ______________ because ______________.
- This is biased against ______________ since ______________.
- This leaves out the perspective of ______________.
- "This might be valid for ______________, but not for the case we're discussing, where ______________."*
- "It appears the AI has taken a literal approach, missing the cultural context of ______________."*
- This is inaccurate because ______________.
- This is outdated because ______________.
- "The AI seems to have misinterpreted ______________."*
- "While this might be true in some contexts, in the scenario of ______________, it's misleading because ______________."*
- The assertion that ______________ has no basis in fact. ______________ is a trustworthy source, and it says ______________.
- ______________ appears to be made up. When I searched for ______________, I found ______________.
Relevance and Focus
- "While the information seems relevant, it misses the point on ______________."*
- This might be technically correct, but it doesn't consider ______________."*
- "The AI appears to have focused on ______________, but it's overlooking the importance of ______________."*
- "This is a general statement that doesn't address the specifics of ______________."*
- This seems to be a repetition of known facts rather than an insight into ______________.
- "This is an over-simplification of the complex issue of ______________. It overlooks ______________."*
- "On the surface, this looks right, but it doesn't explain ______________."*
- This is so vague that it is not useful in helping us think through ______________.
The above template phrases followed by asterisks (*) were adapted from ChatGPT output responding to a "Template phrases for AI output critique prompt," ChatGPT, 25 Sep. version, OpenAI, 3 Oct. 2023, https://chat.openai.com/c/a8b15d03-3...f-2014ced05511. The remaining original phrases and the organizational structure are by Anna Mills and are shared under a CC BY 4.0 license.