Utility-based billing (for example, one search costs $0.01) is a good solution for surfacing costs to customers and mapping usage (and value) to how much they pay. However subscription models (pay one fixed monthly price) are more understandable for customers. (View Highlight)
But the reality is that this is a key area of UX for us, and will be so for anyone doing products that rely on serious use of AI in this way. (View Highlight)
If the user passes some arbitrarily-defined level of scale, should the UI switch to a batch mode with different affordances? Instead of trying to respond instantly to edits, users would queue up a batch of work and then start the process, returning later to see the results. (View Highlight)
A knowledgeable human operator works together with the computer in a process of refining the search query, scanning/filtering the papers, experimenting with different column names (themselves a type of query that drives the data extraction), and verifying the AI's work. (View Highlight)
in the conventional workflow, a researcher would have to read the entire paper to figure out what data is needed and then copy-paste or paraphrase the data items into their Excel results set. (View Highlight)