Skip to Main Content

DKU AI Literacy: AI Research Tools

Introduce Artificial Intelligence resources, tools, and ethics for the DKU community.

Literature Review Tools

ChatGPT

ChatGPT is an AI product established by OpenAI. It creates conversational dialogs to help draft or edit content, brainstorm ideas, program code and learn new things. The most updated version is GPT 4 (released on 03/14/2023). The latest version can offer a faster response time and internet plugins, also handle more complex tasks. Duke Kunshan has launched ChatGPT 4.0 here. Currently (05/21/2023), this DKU ChatGPT can only support text generating, not image. You can also customize some settings in it.

1. Attached Messages Count: this refers to the number of previous messages or parts of a conversation that the model retains in its context. It helps the model maintain context and coherence in the conversation. The larger the number, the more context the model has to refer back to.

2. Temperature: this parameter controls the randomness of the model’s output. It ranges from 0 to 1:
    •    Lower values (e.g., 0.2) make the output more deterministic and focused, often resulting in more repetitive and predictable responses.
    •    Higher values (e.g., 0.8) make the output more random and creative, potentially leading to more varied and less predictable responses.

3. Max Tokens: this setting determines the maximum number of tokens (words or parts of words) that the model can generate in a single response. It includes both the input and output tokens. Setting a limit helps control the length of responses and prevents overly long outputs.

4. Presence Penalty: it is a parameter that discourages the model from repeating phrases or words it has already mentioned in the current context. A higher presence penalty encourages the model to introduce new concepts and avoid repetition. It can help make the conversation more engaging and diverse.

Consensus

People can ask a scientific question or a phrase search in Consensus, the Consensus will run OpenAI’s GPT-4 language model over it. The top 10 results produce a simple one-sentence summary of the top studies to your questions. Current source materials used in Consensus come from the Semantic Scholar database. Consensus is a search engine, not a chatbot; it aims to build a search engine to find the best and most relevant research papers to ground its response.

Consensus is a search engine, not a chatbot, and the purpose is to build a search engine to find the best and most relevant research papers to ground its response.

Query, the research question is very important to get good result. The system recommended four different kinds of search queries:

  1. Single word search to learn about a subject or topic, “Cancer”;
  2. Yes/no question, “does the death penalty reduce crime?”
  3. Relationship question, “what is the impact of climate change on GDP?”
  4. Start a command, “review the literature on the relationship between socioeconomic factor and healthcare outcomes in America”

Search function: over the abstract and title of all papers, with keyword search + Vector Search

                 

Scite

Scite is research AI tool, its unique analysis method is that not only counts citations but also categorizes them as supportive, contrasting, or mentioning. This analysis provides more context about how a research paper has been cited, allowing users to see whether subsequent studies have supported or challenged its findings. 

Elicit

Elicit is a research tool designed to summarize research findings, compare studies and assist with especially systematic reviews. Users can customize the results by adding the needed information to “tables” for export and comparison. Results can be filtered by study types. 

There are three ways to use Elicit services:

  • Find papers: find literature based on a research question like “What are the benefits of taking l-theanine?”
  • Extract data from PDFs: upload a PDF and let Elicit extract and summarize the information from the file
  • List of concepts: find literature based on broad concepts like “treatments for hair loss”

 

Perplexity

Perplexity indicates its answer is backed by citations from trusted news outlets, academic papers, and established blogs to provide up-to-date information.

They provide two ways to search, Quick Search is for fast, straightforward answers by quickly summarizing information from indexed sources; Pro Search goes deeper into the specifics of your question and may ask follow-up questions to refine the search and deliver more tailored responses.