Doing research with AI tools -- avoiding the trap of fabricated URLs
In this short podcast, I explore using AI tools to do research, the potential for fake URLs, and how to deal with the fabrication. I started by using Claude to summarize a podcast and provide a list of salient points, including the potential counterargument. What I didn't expect was for Claude to fabricate a list of imagined research and then summarize the fictitious research to conclude that it lended support for the counterargument. I took Claude's list of research and pasted it into ChatGPT with Bing to browse the real-time web to validate the sources. Using Claude and ChatGPT in combination worked pretty well, but overall this is a tale of caution. You have to be suspicious of research provided by AIs and know how to use each tool according to its strengths.