New Elicit Beta for Systematic Review and Meta-Analysis || Ai Literature Review || Hindi || 2023

Elicit’s newly released beta is positioned as a faster, more systematic way to run literature reviews and meta-analysis workflows—especially when the goal is to extract structured data from many PDFs and then turn those extracts into usable summaries. The core pitch is straightforward: upload up to 100 papers (PDFs), let the system generate structured outputs, and then download results in formats that fit downstream analysis—reducing the manual grind of reading, extracting, and reformatting findings across studies.

In the walkthrough, the beta is accessed through a dedicated website (noted as “elicit…dot…in”), with login handled via Google sign-in or an email-based account. Once signed in, the interface presents three main actions, with the most emphasized being PDF-driven extraction. The demo shows uploading a small batch first (five papers), then scaling the same workflow up to larger sets. After upload, the system produces a structured “final answer” view tied to the uploaded documents. From there, users can select what to include—such as intervention and outcome measures—and remove fields that don’t apply to their review. The user can also adjust the level of detail in the generated summaries (including options described as “more” or “full” output), and verify that the extracted fields align with what’s actually in the papers.

A key practical feature is the ability to download the extraction results. The walkthrough references downloading outputs in spreadsheet-friendly formats (including CSV and Excel-style options), and then “organizing” the data into tables for analysis. The emphasis is on keeping the workflow consistent: instead of reading papers one by one and manually copying results into a spreadsheet, the platform aims to produce a structured dataset directly from the PDFs, ready for synthesis.

Beyond extraction, the beta adds concept-level support through a “Discover” function that generates research concepts across papers. The demo uses a health-related example around sleep disorders and cardiovascular conditions, where a user supplies a broad concept prompt and the system generates related sub-concepts (e.g., links involving inflammation, immune function, and other indirect pathways). The output includes relevance signals and hyperlinks back to source material, with a noted behavior of removing duplicates so the evidence list stays cleaner. The concept view is framed as a way to check whether an idea has scientific grounding and to surface candidate papers for deeper exploration.

The workflow also includes query-based research assistance: users can run targeted searches (e.g., focusing on a specific signal or classification), then review “summary of top” answers and add them into the extraction workflow. The demo concludes with a comparative note—encouraging users to test the beta against other tools they already use, since different systems have different strengths—while stressing that the beta’s structured extraction and concept generation can help improve the speed and quality of literature review work.