Skip to content
TheChatArchiveThe Chat Archive
Geminigemini-2.5-pro· GoogleHugging Face LMArena Human Preference 140K

I have the title and the contributor info. of a particular NCBI Gene Expression Omnibus…

From a Chatbot Arena battleThe same person sent this prompt to two anonymous models at once. Compare it with Metallama-4-maverick-03-26-experimental’s answer.
  1. User

    I have the title and the contributor info. of a particular NCBI Gene Expression Omnibus Series (GSE) ID. But unfortunately, there is no citation, so I cannot natively figure out which publication is associated with that particular experiment.

    To mitigate this, I am going to use a LLM equipped with a web search tool. Given my inputs of the title and the contributors, it should search and return me the proper, full, and correct title of the journal article corresponding to my title (the input title and the journal article title IS NOT GUARANTEED to be the same) while ensuring that the contributors I have given it are also part of that journal article.

    I hope you understood the task. This is very important, so no hallucinations are allowed. Write a prompt for this task.

  2. Gemini Assistant

    Of course. This is an excellent use case for an LLM with a search tool. A well-structured prompt is key to getting accurate, non-hallucinated results.

    Here is a robust prompt designed for this task. It includes role-playing, clear instructions, constraints, and an example to guide the model effectively.


    The Prompt

    You are an expert bioinformatics research assistant equipped with a web search tool. Your primary function is to bridge the gap between NCBI Gene Expression Omnibus (GEO) Series records and their corresponding peer-reviewed publications when a direct citation is missing.

    Your Task: Given the title of a GEO Series (GSE) and a list of its contributors, you must use your web search tool to find the full and correct title of the journal article that published the results of that experiment.

    Inputs:

    1. GSE_TITLE: The title of the experiment as listed on the NCBI GEO database.
    2. CONTRIBUTORS: A list of names of the researchers who contributed to the GSE submission.

    Process to Follow:

    1. Analyze Inputs: Carefully examine the GSE_TITLE for unique keywords (e.g., cell types, conditions, gene names, techniques) and the CONTRIBUTORS list.
    2. Formulate Search Queries: Construct precise search queries for search engines like Google Scholar or PubMed. A good query combines several key elements:
      • A few unique keywords from the GSE_TITLE.
      • The last name of the first or last contributor.
      • The term "publication," "paper," or a specific journal name if suspected.
      • Example Query: "specific keyword from title" "Last Name of Contributor" publication
    3. Verify the Match: Once you find a potential publication, you must perform a two-step verification:
      • Content Match: The topic, methodology, and organism described in the journal article must closely match the description in the GSE_TITLE.
      • Contributor Match: You MUST confirm that the CONTRIBUTORS I provided are listed as authors in the found journal article. It is acceptable if the article has additional authors, but my provided contributors must be present.
    4. Extract the Output:
      • If a verified match is found, return only the full and correct title of the journal article. Do not return the GSE title, the abstract, the journal name, or any other information unless explicitly asked.
      • If you cannot find a matching publication or cannot confidently verify the match based on the contributors, you must respond with: No definitive publication found for the provided information.

    Crucial Rules:

    • NO HALLUCINATIONS. This is a fact-finding mission. If you are not 100% certain, state that you cannot find the publication.
    • The title you return MUST be for the peer-reviewed journal article, not a restatement of the GSE_TITLE. The two titles are often different.
    • The contributor check is mandatory and is the most important verification step.

    Example Usage:

    [User provides the following]

    GSE_TITLE: Gene expression data from human primary T cells stimulated with anti-CD3 and anti-CD28 in the presence of Wortmannin or LY294002 CONTRIBUTORS: Harald Tresch, Thomas Hehlgans, Daniela M. Männel, Falco-boris H. Guse

    [Ideal LLM Output]

    Inhibition of PI3K-p110δ affects T-cell activation and function


    This prompt structure clearly defines the role, task, inputs, methodology, and strict output requirements, significantly minimizing the risk of error and preventing the model from generating plausible but incorrect information.

Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.