If another search still comes up empty, record what you searched and which papers you checked. More on separating a search miss from a gap: https://t.co/H3El0umg3z
“No results” is a frustrating answer when you remember reading the finding. Try synonyms and a narrower question, then check likely papers directly. How do you find a passage a search missed?
Your advisor can now open your literature map, read every claim, and click through to the cited page. Your co-author can edit it with you, live.
New in Agent Bayes: share a project, a library, or one map, at View or Edit. Edit the same claim at once and neither save is lost.
@vsbuffalo A reader should not have to search scientific code for its assumptions. Ask the AI to place each assumption beside the calculation that depends on it.
A folder full of paper notes is not a draft plan. Group notes by the argument or section they support, with the source passage beside each claim. How do you move from reading notes to a chapter outline?
A literature review should make disagreements easier to see, not smooth them away. Put competing claims side by side with their evidence, then compare methods and samples. How do you organize conflicting findings?
Make an AI review finding useful by showing the manuscript claim beside the source passage. Authors can point to a missed passage, and reviewers can judge the evidence. What would you want to see beside a review comment?
A correct citation can still give the wrong researchers credit. Before citing a review, check whether its authors conducted the study or reported another team’s result. If you read only the review, say so. How do you handle that distinction?
A citation can be correct while the claim beside it goes too far. Before submitting, check whether the paper reports a cause or only an association. Which sentence in your draft deserves another look?
An AI review can involve more than the person uploading a draft. Before you upload, check any coauthor agreements and institutional rules. Read the account terms for retention, training and outside processors too. What is on your checklist?
@snigdhac25@ReviewAcl@naaclmeeting Readability should cover the methods, not just polished sentences. A reviewer should be able to tell what the authors did well enough to repeat it.
@geerling_lab The final “concerns addressed” is hard to assess when readers cannot see the concerns. Which criticism led the authors to change their conclusion?
Before asking AI to explain why it missed a finding, check the imported PDFs. The paper you remember from Zotero may never have been included. A better prompt cannot supply a missing paper. Ever checked the file list?
@james_y_zou “The model failed this task” becomes “LLMs cannot do this” surprisingly easily in a literature review. Keeping the finding in the past tense helps keep the claim honest.