Posting in his SubStack Adjacent Possible, Steven Johnson discusses how “language models are opening new avenues for inquiry in historical research and writing“. He suggests they can act as collaborative tools, rather than replacements for the writer’s engagement with primary sources.
Johnson argues that NotebookLM is designed to facilitate rather than replace the reading of original sources. I t does so by making the entire source readable within the app, and by provinding inline citations linked directly to the original material.
He identifies some interesting use cases.
The AI can be a tool for collaborative brainstorming by allowing users to explore different hypotheses and see patterns within personally curated sources.
NotebookLM can be used for targeted information retrieval.
- It can help “fill in blank spots” or remind users of forgotten details from their readings.
- The tool is valuable for fact-checking against uploaded source material.
- For specific information, like in a car manual, it can provide direct answers to questions through a conversational Q&A format.
It can enhance serendipitous discovery by suggesting surprising, less obvious connections amongst the sources.
It can create mind maps from the sources, in effect indexing them on the fly.
Finally he speculates on a future where e-books could come with a NotebookLM-like interface. This would bundle together the main work with all the original sources used by the author, enabling “timelines, mind maps, and explanations of key themes, anything you can think to ask”.