LitPivot：在文献图景中动态情境化，辅助发展研究想法

Why pivoting is hard

Research ideas evolve through engaging with the literature. In our analysis of meeting notes from four completed computer science research projects, we identified 21 literature-initiated pivots across all four projects. These fell into two types: differentiating an idea from existing work to establish an idea’s novelty, and aligning an idea with existing work to build on it. This project itself went through two pivots before we settled on its contributions.

These changes can be difficult to navigate. In a formative study, five researchers developed proposals using a text editor and standard search tools. They struggled to develop their ideas while critically evaluating them against the literature. When a paper challenged an idea’s novelty, they wanted concrete suggestions for how to revise it, supported by prior work. These findings shaped LitPivot’s assessments, which pair literature-supported critiques with specific suggestions for revising the selected part of an idea.

How LitPivot works

In LitPivot, a researcher starts by writing a short proposal in an editor and enabling what we call facet highlighting , which segments an articulated idea into different facets (i.e., a problem, contribution, and evaluation). To check a proposed facet, the researcher clicks that segment in the draft. An adjacent panel groups related papers based on the selected facet (e.g., grouping papers by the evaluation methods they use when the evaluation facet is selected) and highlights the most relevant groups. The researcher selects papers, requests an assessment, and reviews a suggested revision alongside excerpts supporting the critique. The literature is then reranked based on relevance to the revised facet.

In particular, LitPivot separates a research idea into its problem , proposed contribution , and evaluation plan . Each facet has its own set and relevance ranking of related literature. LitPivot groups retrieved papers around the facet being edited, letting the researcher choose which ones the system should use to assess the draft.

Alongside scientific-writing environments such as Prism and agentic assistants such as Claude Code, LitPivot explores a new workflow for developing an early research idea: choosing literature to assess it against, using that evidence to revise it, and reconsidering which papers matter as the idea changes. Its paper groups and linked critiques make those steps explicit for each part of the proposal, while researchers control both the draft and the evidence used to assess it.

For a problem statement, LitPivot looks for evidence that the problem exists and reasons it matters. To assess a proposed contribution, it compares what the selected papers have already addressed with the limitations and future work they describe, looking for remaining gaps. For an evaluation plan, it checks whether the proposed tests address the research question and how similar work has been evaluated. The system then suggests a rewrite with supporting quotes attached.

Changing one facet of an idea can leave another out of date. LitPivot flags passages that may need revision, such as an evaluation plan that no longer adequately evaluates a revised contribution. Researchers can accept, edit, or reject suggestions, and request new papers or refresh the literature ranking as their proposal develops.

Paraphrased example from one study participant.

Our findings

In our lab study, 17 researchers each developed one assigned idea with LitPivot and another with a baseline tool—a text editor and an AI tool for asking questions about selected papers. Both conditions used the same collection of papers. Two domain experts, unaware of which system had been used, rated the resulting proposals on a single criterion: how well they used relevant literature to support and justify the idea. Proposals developed with LitPivot received higher ratings.

Participants in the same study also reported a stronger understanding of the surrounding literature and greater self-reported improvements in their ideas’ novelty and usefulness with LitPivot than with the baseline. Domain experts also reviewed the participants’ ideas and rated the ideas revised using LitPivot to be significantly better grounded in relevant literature.

One participant in our lab study started with a proposal to generate consistent, factual synthetic medical notes. LitPivot identified consistency across multiple hospital visits as a potential gap in the selected literature. The participant reviewed abstracts and removed papers about single-visit notes. After another assessment, they narrowed the proposal to detecting contradictions across notes from multiple visits. The example shows how revising an idea and choosing the literature used to assess it can inform each other.

In a follow-up study, five researchers used LitPivot on their own ongoing projects during 30-minute sessions. Participants used LitPivot across multiple project phases, not just at the beginning. Two used it to vet new ideas against the literature, while three used it to refine the framing of more mature ideas. One researcher highlighted how seeing parallel concepts from different domains helped them reframe their contribution.

These studies suggest that LitPivot can support both early exploration and the refinement of an existing research direction.

Supporting scientific discovery

At Ai2, we’re developing AI to help researchers formulate testable ideas and examine the evidence behind them. Asta supports that effort through theory generation from published findings (Theorizer) , synthesis, and hypothesis generation and testing on data (AutoDiscovery) , while HARPA generates promising hypotheses informed by emerging research trends. Scideator lets researchers recombine elements of prior work into new ideas and assess their novelty, and IdeaSynth supports iterative refinement with feedback grounded in the literature.

Our recent study, Old Ideas, Novel Problems , examines the limits of automated novelty evaluation, showing how small prompt changes can alter an AI judge’s verdict on the same ideas. LitPivot extends this line of work by making the choice of relevant literature part of the revision process. Researchers can change which papers they use to assess a proposal as their idea develops, then use critiques and supporting excerpts to decide how to revise it. The system connects those decisions across the proposal’s problem, contribution, and evaluation plan.

We hope insights from LitPivot help scientists build on existing knowledge and turn early ideas into new contributions to their fields.

More from this author

OlmoEarth v1.3: State-of-the-art embeddings for the whole planet

FairStress: A deeper evaluation of fairness in LLM decision-making