Generative AI is becoming increasingly prevalent in academic research, offering tools that enhance both the productivity and creativity of researchers. For master's and PhD students, as well as staff researchers, leveraging advanced AI models and research tools can significantly improve various aspects of the research process, including literature review, data analysis, and academic writing. This article introduces advanced AI models, such as OpenAI's GPT series, Anthropic's Claude models, and Google's Gemini models, which are renowned for their capacity to assist in the exploration of research ideas, provide analytical support, and improve the clarity and coherence of writing. Additionally, it explores AI research tools like Elicit, COnsensus, and SciSoace, which facilitate tasks such as literature discovery, summarisation of academic papers, and the organisation of research findings. By understanding and effectively utilising these technologies, researchers can greatly enhance their research productivity, deepen their insights, and elevate the quality of their academic work 

 

 

 

 

 

For research purposes, GPT-4o represents the most advanced model currently available from OpenAI, offering advanced capabilities that significantly enhance the research process. It is effective in conducting literature reviews by summarising articles, performing advanced data analysis, and assisting in the organisation of complex information. Researchers can upload full PDF files of papers into GPT-4o and prompt the model to explain the content, enabling them to quickly grasp key insights. A distinctive feature of GPT-4o is its ability to manage relatively large text and visual inputs simultaneously, providing rich context and precise responses (OpenAI, 2024). Its extended context window allows for the integration and referencing of large volumes of information across different modalities—such as text, images, and soon audio—all within a single session, which is particularly useful for synthesising diverse sources of information. Additionally, GPT-4o’s vision capabilities allow researchers to analyse visual elements within their research materials, such as figures, charts, and screenshots, thereby facilitating a deeper understanding of complex data. By assisting in structuring literature reviews and providing comprehensive analyses, GPT-4o serves as a valuable tool for researchers, offering a strong foundation for the further exploration and development of their research ideas and projects.

 

 

 

Figure 1

 

 

 

Claude 3.5 Sonnet (as seen in Figure 1) is the latest model released by the lab Anthropic, designed to enhance performance in reasoning, coding, and safety. It surpasses previous models and competitors in various benchmarks. Claude 3.5 Sonnet is distinguished by its advanced reasoning capabilities, strong coding proficiency, and improved safety features. A unique feature known as “Artifacts” allows for more efficient data handling. For instance, when handling data, Claude opens a dedicated section on the right half of the window screen, where users can interact with the data more directly. This section enables real-time tracking of modifications to Claude’s data output in response to user prompts, allowing for dynamic interaction with the data. This capability facilitates immediate observation of changes, thereby enhancing the efficiency and accuracy of the data analysis process within a single interface (Anthropic, 2024). This feature significantly improves the model’s usability in research by providing seamless editing and data visualisation within the same platform. While Claude 3.5 Sonnet is comparable to GPT-4o in terms of capabilities, it has certain limitations regarding the formats, number, and size of documents it can handle compared to GPT-4o. For example, Claude 3.5 Sonnet may struggle with processing very large PDF files or handling multiple files simultaneously, while GPT-4o is more adept at managing these tasks. Additionally, the rate limit for Claude 3.5 Sonnet is significantly higher, which can be a constraint for users who require continuous or high-volume processing. Despite these limitations, the “Artifacts” feature remains a notable advantage due to user-friendliness and convenience.

 

 

 

Google’s Gemini 1.5 Pro, the latest AI model from Google, features a groundbreaking 1 million token context window. This advancement vastly surpasses previous limitations, allowing researchers to upload unprecedented amounts of data within a single prompt. Such capacity enables more complex, multi-modal analyses that were previously unfeasible with other models. Notably, Gemini 1.5 Pro maintains high levels of performance even as its context window increases. In the Needle In A Haystack (NIAH) evaluation,—a test designed to assess a model’s ability to locate specific information within large volumes of text— where a small piece of text containing a particular fact or statement is purposely placed within a lengthy text block, Gemini 1.5 Pro accurately identified the embedded text 99% of the time, even when the text was hidden within blocks containing up to 1 million tokens (Pichai & Hassabis, 2024). While Gemini 1.5 Pro excels in its context window capacity, it is somewhat less proficient than GPT-4o and Claude 3.5 Sonnet in areas such as reasoning, coding, and attentiveness to instructions. Despite this, the model’s expansive context window and its ability to upload videos provide a distinct advantage. This feature makes Gemini 1.5 Pro particularly useful for researchers who need to work with long texts, extensive datasets, and multimedia content, offering a unique advantage, especially for researchers who require the integration of multimedia data into their analyses. For instance, researchers engaged in projects involving both textual and video data can more effectively analyse these different data types within a single framework.

 

While advanced AI models such as GPT-4o, Claude 3.5 Sonnet, and Gemini 1.5 Pro significantly enhance the research process, their full potential is unlocked when used in conjunction with specialised AI research tools. These tools are designed to address specific needs in the research workflow, such as literature discovery, data extraction, and the organisation of research findings. By integrating these AI research tools into their practices, researchers can make more informed decisions, and extract deeper insights from their data. The following section introduces three powerful AI research tools—Elicit, Consensus, and SciSpace—each of which offers unique features tailored to different stages of the research process.

 

 

 

 

 

Elicit AI has emerged as a highly powerful research assistant, offering a comprehensive set of features designed to enhance the research process. The platform is centred around notebooks and provides three primary ways to interact with data: “Find Papers,” “Extract Paper from PDFs,” and “List of Concepts” (Elicit Support, 2024), as shown in Figure 2.

 

Figure 2

 

 

 

1. Find Papers

 

This feature functions as an advanced search engine for research questions. Users can input specific research queries in the interface, making it ideal for investigating any emerging query and new topics. Elicit generates a notebook and searches for relevant research papers addressing the query. Users are provided with a summary of the top papers, offering a snapshot of relevant information along with references. Clicking on these references directs users to the corresponding papers for further detailed review. Elicit also displays columns listing the papers and abstract summaries, accompanied by sorting and filtering tools to refine the search to the most specific and useful information.

 

2. Extract Paper from PDFs

 

This feature is exceptionally powerful for individuals who already have PDFs of research papers and wish to interact more deeply with that data. Users can upload papers to Elicit’s library and select the ones they want to analyse further. This tool enables users to conduct in detailed examinations of the data by posing specific questions about the papers and delving into their content more comprehensively.

 

3. List of Concepts

 

This feature is invaluable for researchers entering a new field or exploring new branches within their current research. It provides a broad overview of the research field, ensuring comprehensive coverage of all relevant concepts. Traditionally, gaining a thorough understanding of a research topic required extensive hours of reading. However, with Elicit, all necessary information is systematically presented in a straightforward table format, making it easier to grasp the scope of the research area.

 

 

 

Figure 3

 

 

 

Consensus, as illustrated in Figure 3, is an excellent tool for researchers with specific questions in mind. By entering a research question, Consensus provides an initial summary that synthesises findings from relevant papers (Consensus, 2024). This summary serves as a concise takeaway message from the analysed literature. Below this summary, the Copilot feature, powered by ChatGPT, is available to offer additional insights. The true strength of Consensus emerges as the Copilot generates insights on how each individual paper addresses the posed question, providing a detailed and useful perspective.

 

Consensus provides results in the form of reference cards, each containing pertinent information about the study type, citation count, and its influence, facilitating the evaluation of its relevance and utility. These cards function as concise snapshots of the research, helping researchers quickly assess the relevance and utility of each paper for their specific inquiries. The Consensus metre is particularly effective for addressing yes-or-no questions, offering a clear overview of the prevailing consensus in the research field. AI filtering based on study types, such as meta-analyses, systematic reviews, randomised controlled trials (RCTs), non-RCT trials, and observational studies, further refines the Consensus metre results, providing more precise insights.

 

Consensus also integrates seamlessly with reference management tools like Zotero, facilitating easy referencing in Word or other word-processing software. The results are powered by SciScore, which not only provides the H-Index or Impact Factor of a journal but also assesses the journal’s rigour. This ensures that the papers are sourced from reputable sources.

 

 

 

Figure 4

 

 

 

SciSpace, as illustrated in Figure 4, combines elements of both Elicit and Consensus, creating a versatile platform for research inquiries. Upon entering a question, SciSpace generates a summary of the top five relevant papers, followed by a table with detailed information about the articles referenced in the summary, as well as additional relevant papers. This structured presentation allows for efficient review and analysis.

 

Users can save interesting papers in “My Library” for easy access and reference. Additionally, PDF files can be uploaded into the library, with options to select specific types of data for extraction. The Copilot feature in SciSpace, powered by OpenAI’s ChatGPT technology, further enhances its functionality by acting as an AI-powered research assistant. This feature allows for inquiries about the uploaded PDFs, offering detailed explanations, summaries, and insights based on the content. The Copilot feature is available in both a standard version and a high-quality paid version, providing flexibility to accommodate various user needs.

 

To utilise the Copilot feature, sentences within a paper can be highlighted, and then Copilot can be accessed in the sidebar to generate responses to questions about the highlighted sections. This interactive approach enables researchers to explore and understand the research material more deeply.

 

Overall, each of these AI research models and tools presents distinct features that cater to different aspects of the research process. OpenAI’s GPT-4o excels in handling complex and diverse research materials, providing comprehensive summaries, advanced data analysis, and the ability to integrate multiple modalities such as text and images. Anthropic’s Claude 3.5 Sonnet stands out for its reasoning and coding proficiency, particularly enhanced by its unique “Artifacts” feature, which facilitates real-time data manipulation and visualisation. Google’s Gemini 1.5 Pro, though slightly less advanced in reasoning and coding, offers a substantial 1 million token context window and multi-modal analysis, making it exceptionally useful for researchers working with large-scale datasets.

 

In terms of research tools, Elicit allows users to efficiently locate, extract, and conceptualise research material with ease. Consensus helps synthesise research findings into concise summaries and provide detailed insights through its Copilot feature, making it particularly useful for researchers seeking prompt, informed responses to specific questions. SciSpace, on the other hand, integrates the strengths of both Elicit and Consensus by offering a platform that supports both summarisation and detailed interaction with research papers. While Elicit is ideal for in-depth exploration of new research fields, Consensus provides clarity and precision in understanding the consensus within a research area, and SciSpace offers a balanced approach that enables both high-level overview and detailed analysis. Collectively, these tools empower researchers to navigate and engage with academic literature more effectively, each contributing unique strengths to the research toolkit. Given the dynamic nature of these tools and their continually evolving functionalities, researchers are encouraged to conduct independent evaluations to determine which tools best align with their specific research needs.

 

 

 

 

 

Anthropic. (2024, June 21). Introducing Claude 3.5 Sonnet. Retrieved August 5, 2024, from https://www.anthropic.com/index/claude-3-5-sonnet

 

Anthropic. (2024). Claude AI platform. Anthropic. https://claude.ai/

 

Consensus. (2024). How it works & Consensus FAQ’s. Retrieved August 5, 2024, from https://consensus.app/home/blog/welcome-to-consensus/

 

Consensus. (2024). Consensus AI platform. Consensus. https://consensus.app/

 

Elicit Support. (2024). Building your first notebook. Retrieved August 5, 2024, from https://support.elicit.com/en/categories/123777-getting-started

 

Ought. (2024). Elicit AI platform. Elicit. https://elicit.com/

 

OpenAI. (2024, May 13). Hello GPT-4o. Retrieved August 5, 2024, from https://openai.com/index/hello-gpt-4o/

 

Pichai, S., & Hassabis, D. (2024, February). Our next-generation model: Gemini 1.5. Google. Retrieved August 5, 2024, from https://blog.google/technology/ai/google-gemini-next-generation-model-february-2024/#performance

 

SciSpace. (2024). SciSpace platform. SciSpace. https://typeset.io/