Hey everyone! I'm pretty new to using Elicit for my research tasks, and I've been really impressed with how it can pull up papers and summarize findings. It's been a game-changer for my project management workflow, especially when I'm trying to get up to speed on a new topic quickly.
But I've run into something a few times now that's been a bit frustrating, and I'm wondering how you all deal with it. Sometimes, when I ask Elicit a question, the results I get back seem to be answering a *different* question than the one I intended. It's like the tool slightly misinterprets the keywords or the intent behind my query. For example, last week I was looking into "best practices for remote team retrospectives in agile," and a lot of the papers it surfaced were more about general remote work communication tools or broad agile principles, not specifically about the retrospective meeting format. I had to sift through a lot to find the focused stuff.
So my main question is: what's your process when this happens? Do you have a specific way of rephrasing your questions to get Elicit back on track? Are there certain phrasing tricks or structures that work better than others? I'm also curious how this compares to other AI research tools you might have used—do they have the same issue, or do they handle ambiguous questions differently?
I'm still learning the ropes, so any advice from more experienced users on how to guide the tool would be super helpful. I love the potential here, but I want to make sure I'm using it as effectively as possible for my team's literature reviews! 😅
Oh, that's a familiar pain point. I don't use Elicit much for research papers, but I hit similar issues all the time with semantic search in data catalogues. The keyword interpretation can be surprisingly literal.
For me, the trick is to think like a thesaurus and get more specific with the terms you *don't* want. Maybe try rephrasing to exclude the broader topic? Something like "retrospectives agile remote, excluding general communication tools". It sometimes helps to anchor on a very specific artifact you *do* want first, like "distributed sprint retrospective meeting format", and then explore from there.
How much do you tweak your phrasing vs. just filtering the results it gives you?