AI | Artificial Intelligence | University of Wisconsin-Whitewater

Understanding how generative AI works helps our community engage more critically with its strengths and limitations in academic, creative, and professional contexts.

Training a generative AI model involves several key steps:

1. Data Collection: The model is fed vast amounts of data relevant to the task at hand. For instance, a language model might be trained on books, articles, and websites to understand human language patterns.
2. Learning Patterns: Using neural networks, the model identifies patterns and structures within the training data. This learning enables the model to generate content that is coherent and contextually relevant.
3. Fine-Tuning: After the initial training, models often undergo fine-tuning using specific datasets or techniques like reinforcement learning from human feedback (RLHF) to improve performance and align outputs with desired behaviors.
4. Evaluation and Iteration: The model's outputs are evaluated for quality, accuracy, and relevance. Based on this evaluation, the model may be further refined to enhance its performance.

Training and refining these models requires substantial computational resources and human expertise to ensure ethical, effective performance.

For a visual explanation of how Large Language Models (LLMs) work, you might find the following video helpful:

Source: 3blue1brown.com

If you're interested in exploring generative AI tools or have specific questions about their applications, feel free to reach out to CATLST.

• Protect personal information: Avoid inputting sensitive or personally identifiable information (e.g., student records, health data, unpublished work, internal documents or other university/institutional data) into AI platforms, as these tools may store and utilize your data beyond your control.
• Respect others' privacy: Do not upload or share others' work, including assignments, course content, or either published or unpublished research, with AI tools without explicit consent.
• Review privacy policies: Before using any AI tool, familiarize yourself with its data handling and privacy policies to make informed decisions about your data.

Refer to Institutional Research Assessment and Planning and UW System Administrative Policy 1031 to understand more about institutional data risk levels, specific examples of data classification (SYS 1031 Guidance: Data Classification Examples) and refer to Allowable or Prohibitive Uses uses for more details.

• Recognize AI limitations: AI models may reflect societal biases present in their training data, leading to outputs that are biased or discriminatory.
• Critically evaluate outputs: Always assess and apply human judgment to AI-generated content for potential biases, inaccuracies, or gaps in logic. Do not accept AI outputs at face value without scrutiny.
• Promote belonging: When using AI tools, strive to ensure that the content generated is inclusive and does not perpetuate stereotypes of marginalized groups.
• Promote equity in access: Be mindful that not all users have access to paid or premium AI tools. When assigning or expecting AI use, aim for free, accessible options whenever possible.

• Be clear and open in AI use: Disclose when, where, and how AI tools were used, including the nature of the task, prompts you entered, outputs generated, and how outputs influenced your work. Cite your AI use appropriately following academic or professional standards.
• Understand expectations and boundaries: AI use may vary widely across courses, departments, and instructors. Students should follow the course-specific guidelines outlined in the syllabus or assignment, including information on when AI use is permitted, restricted, or prohibited.
• Set clear expectations: Faculty/staff are encouraged to clearly communicate expectations around AI use in course syllabi, assignment instructions, research protocols, and/or professional workflows.
• You are accountable: Users are responsible for the accuracy and ethical implications of any AI-generated work they submit or share.