LLMs tools

🦆 Warm-up activity with duck.ai

Open duck.ai and try the same question with two different AI models (you can switch models in the duck.ai interface):

What are three good strategies for keeping on top of the literature in a fast-moving research field?

This exercise shows that different models can give noticeably different outputs even to identical prompts. The model you choose matters—for both the quality and the style of the response. There is no universally “correct” answer; critical evaluation always remains your responsibility.

Large Language Models (LLMs) have been rapidly introduced to workplace and research environments since ChatGPT’s breakthrough launch in late 2022. The landscape is quickly evolving, with major tech companies and AI startups releasing increasingly sophisticated models tailored for different use needs. When selecting an LLM consider factors such as data privacy policies, integration capabilities, cost structure, and specific feature sets that align with the University’s workflows and the sensitivity of your research data. The following list covers some of the most prominent options available today.

duck.ai duck.ai (used for activities in this course)

  • Provider: DuckDuckGo
  • Strengths: Privacy-first; does not save conversations or use them for model training; free; provides access to multiple underlying models (including GPT-4o mini, Claude, Meta Llama, Mistral)
  • Best for: Learning, experimentation, and general tasks where you want privacy by default
  • Considerations: Designed for privacy rather than deep integration with productivity tools; may have usage limits. As with any external tool, do not enter confidential or personal research data

Microsoft Copilot copilot.microsoft.com

  • First Release: March 2023
  • Strengths: Deep integration with Microsoft Office suite
  • Best for: Document creation, email assistance, Office workflows, enterprise productivity
  • Considerations: Enterprise versions offer better data protection; requires Microsoft 365 subscription
University of Bristol AI guidance

The only recommended tool for the time being is Microsoft Copilot if you are looking to use generative AI to minimise data and privacy concerns. duck.ai is used in this course for hands-on practice because it is privacy-respecting and requires no login, but it is not the right tool for any research task involving unpublished or sensitive data.

The goal isn’t to avoid AI tools entirely, but to use them responsibly and effectively.

Find more →

ChatGPT (OpenAI) chat.openai.com

  • First Release: November 2022
  • Strengths: Conversational, good for general tasks, coding assistance
  • Best for: Writing assistance, brainstorming, general Q&A, code generation
  • Considerations: Data usage policies vary by version; enterprise versions available

Claude (Anthropic) claude.ai

  • First Release: March 2023
  • Strengths: Helpful, harmless, honest approach; excellent for analysis
  • Best for: Analysis, research assistance, ethical reasoning, document processing
  • Considerations: Strong focus on safety and accuracy; large context window

Google Gemini gemini.google.com

  • First Release: December 2023
  • Strengths: Multimodal capabilities, integration with Google Workspace
  • Best for: Research, document analysis, creative tasks, Google ecosystem integration
  • Considerations: Various models available (Nano, Pro, Ultra) with different capabilities

Google NotebookLM notebooklm.google.com

  • First Release: July 2023
  • Strengths: Document-grounded AI, source-based research, podcast-style audio summaries
  • Best for: Research analysis, document synthesis, creating audio overviews, academic work
  • Considerations: Works with your uploaded sources; Plus version offers 5x higher usage limits

Meta Llama llama.meta.com

  • First Release: February 2023
  • Strengths: Open-source, customizable, strong performance
  • Best for: Custom applications, on-premise deployment, research
  • Considerations: Requires technical expertise; various sizes available (7B to 405B parameters)


Proper Attribution

When artificial intelligence tools significantly contribute to your work, proper attribution is essential for maintaining transparency and ethical standards. Failing to acknowledge AI assistance where it played a substantial role can mislead audiences about the true nature of your work’s creation. In research, this connects directly to academic integrity and to journal and funder policies on AI use.

Example Attribution: “Initial drafts of this literature review were developed with assistance from Claude (Sonnet 4). All sources were independently verified, analysis was conducted by the author, and conclusions represent the author’s professional judgment.”

Do you require more thorough record-keeping? Use a documentation form instead

LLM_Usage_Documentation_Form.docx