Verification with Lateral Reading
Errors and Hallucinations
Large Language Models (LLMs) are confident even when they are wrong. They have made improvements with each model, but Copilot, Claude, ChatGPT, and others frequently hallucinate. This has to do with how the tools are built. The companies that build these tools are aware of the error rate yet take no responsibility for their errors. That means you need to put in the work to verify facts and figures generated by LLMs.
Verification
Every fact and citation found through a LLM should be verified using information found through a library database or other scholarly source. This is called lateral reading, that is, using authoritative sources to verify the accuracy of information. It is also sometimes referred to as fact checking. You may feel that the work involved in fact checking the AI is too much, and in many cases we agree. Therefore, use AI with hesitation in your literature searches and other research. Use AI as an experimental aspect of your work that requires extra verification.
Many library databases are starting to include AI assistant search tools. These are prone to the same errors as the AI tools you are more familiar with. However, they are citing articles within the library collection, so verification of facts and figures is less essential.
Fact checking through lateral reading is a valuable skill to develop as it helps you bring a critical lens to everything you hear and read. So, take the opportunity to experiment with AI and lateral reading to help you develop your fact checking skills.