Comparison of 6 Paper Search GPTs + Native Browsing (Bing)

Choosing the optimal tool in the field of paper search significantly impacts the efficiency of research and learning. As AI technology, including ChatGPT, evolves, one of the main challenges users face is the issue of inaccurate information (hallucination). One solution to overcome this is to reference information from the internet. Here, we explore the optimal approach to paper searching by comparing the search results of 6 paper search GPTs and the default browsing function (Bing).

First, the conclusion:

It has become clear that the best option for paper searching is GPT services that have built-in literature databases. These services excel in detailed search functions and the ability to delve deep into information. In particular, I think ScholarAI has the best balance and TXYZ is excellent at selecting literature (especially in slightly niche fields). Also, these services support searching in Japanese, which is evaluated as highly convenient. On the other hand, searching using the default browser Bing is excellent for information coverage, but has limitations in obtaining detailed information.

Background of the evaluation:

• Pubmed API limitations: In custom GPTs, even using the Pubmed API, only up to 3 papers can be obtained, which limits information gathering.

• Bing browsing function: The search results obtained through Bing are 8, presented in SEO order, but detailed information cannot be obtained through clicking links. However, it is possible for the user to click the links directly to obtain information.

• GPT services with built-in literature databases: Consensus, SciSpace, ScholarAI, ScholarGPT, etc., have literature databases within their own services and can provide deeper information. However, there are limitations to the features available for free.

Evaluation criteria:

Each service was evaluated based on whether it appropriately returned the following items after inputting the following prompt: Search research papers for ”search keywords” as many as possible. Return following:

• Journal name

• Author name

• Publication year

• Citation count

• Summary

Results:

Summary:

For paper searching in research and learning, GPT services that have their own literature databases have advantages in information acquisition and deep-dive questioning. In particular, ScholarAI is a well-balanced option and TXYZ is excellent at selecting literature. On the other hand, native browsing functions like Bing are excellent for information coverage but are not suitable for obtaining detailed information. Ultimately, it is important to choose the best tool from these based on the user's needs and preferences. Also, these tools have a limit on the number of search results displayed, are unsuitable for comprehensively searching for relevant papers, and I recommend using them when you want to quickly know something you are curious about, or as a starting point for your search.

Links to each GPT are below