“What seems consistent is that, for now, the greatest value comes not from surrendering control entirely to AI or clinging to entirely human workflows, but from finding the right points of collaboration for each specific task—a skill we’re all still learning.”
– Ethan Mollick
As AI tools rapidly evolve, specific advice on which tool to use or the “most effective” ways to interact with these tools is ever changing. Ethan Mollick’s Quick Guide to using AI Right Now is updated every few months. New features such as varied linguistic styles or tones, and the frequency of hallucinations (i.e., instances where the tool presents incorrect or fabricated details) change with each new model. Moreover, different flavors of models appear: from quick chats, to advanced reasoning models that enhance problem-solving capabilities, and and deep research that includes live website links to sources and information they suggest. Regardless of how these systems evolve – it is still true that how you interact with the system matters! In other words, the prompts you use can drastically increase the relevance and accuracy of the output.
Talking with AI – Strategies for Crafting Better Prompts
Before using AI for your work and learning at Tufts, be sure to review the Tufts Generative AI Guidelines
You will interact with many common generative AI systems through a “chat” box across the Internet. A prompt is the input you provide to an AI model to generate a response. It can be a question, command, or statement that guides the AI’s output.
Different prompts can lead to drastically different output and impacts the relevance and accuracy of the generated response (see How to Get the Best Results from ChatGPT). The practice of designing and refining prompts to achieve more accurate, relevant, and useful AI-generated responses (Prompt Engineering) helps users maximize the capabilities of AI while reducing errors and biases. Prompts can be delivered over multiple points in a conversation and can be quite long, see for example a prompt to help a teacher develop project ideas, a prompt to give a student feedback about their work, or this prompt to prime AI to help you use LaTeX. As models evolve, specific formulas of prompt engineering become less critical, but there are some fundamental qualities of AI Fluency (See for example this free Course from Anthropic, The AI Fluency Framework.
However, there are also some quick strategies for improving your prompts, including advice to:
- Be Specific – The more detailed and precise your prompt, the better the AI response.
- Provide Context – Offering background information helps AI understand the question better.
- Use Step-by-Step Instructions – Breaking down complex tasks improves output accuracy.
- Experiment and Iterate – Keep tweaking your prompts based on responses.
- Set Format Expectations – Specify if you need a list, summary, structured output, etc.
Other effective prompting techniques rely on interactively using the AI itself to help you improve it’s output. Other common techniques to increase the relevance and accuracy of the responses involve including additional context and asking an AI system to outline its steps in finding an answer. For example consider the following prompts:
| Before proceeding with this task, please ask me any questions you need to provide the most helpful response possible. Consider aspects like additional context I might be able to provide, specific requirements that could help shape relevant output, my response format preferences, and any constraints that would be helpful to be aware of. Once we’ve clarified the details of this task, please first outline the steps you will take to develop a solution and wait for any feedback I might provide. Once we’ve agreed on your process, proceed step-by-step, pausing after each step to check in with me. |
Prompts are the main way you interact with an AI system, yet as you do so, it’s critical to consider the continuum of ways AI can be used as well as the larger ethical questions of where, when and how it’s appropriate to use AI.
Using AI Generated Content
Keep in mind when you are considering using AI in your work, you are always ultimately responsible for the output.
As Ethan Mollick writes in Co-Intelligence: Living and Working With AI, “be the human in the loop”; anything co-created with a generative AI should be checked, edited and reworked as needed before being shared.
As we introduced in the first section, understanding how it works is essential for making informed decisions about when and how it can be useful. AI models predict and generate outputs based on patterns in the data they were trained on, not through independent reasoning or human-like cognition. While it might be expressed in a confident or authoritative voice, even asking for search citations or confidence in an answer doesn’t lead to reliable responses. It is important to understand and explore the limitations of the AI tool you are using and to always critically analyze the content you generate.
Quick things to keep in mind when reviewing AI generated content:
- Cross check content with credible sources. AI may generate incorrect information (hallucinations).
- Read carefully to check for inconsistencies and misleading information. AI responses
- Understand and explore the limitations of the AI tool you are using.
- Look for biased content: AI reflects biases present in its training data, leading to skewed outputs both in the text that is generated and what is not generated.
- Inconsistencies in responses – AI-generated answers may vary even with similar prompts.
Ultimately, while generative AI tools can be impressive creators, they still aren’t perfect—and likely never will be. There may always be hallucinations that an expert can spot, and getting the tool to independently produce exactly what you are looking for can be time-consuming and frustrating. Therefore, the most effective use of generative AI is to treat it as an assistant that can help you achieve your goals, rather than relying on it as an independent creator.
Overall, it can require effort to deliberately stop and evaluate seemingly polished language in a generated response. When we are too quick to accept the suggestions generated by AI tools it can bias us away from other solutions we might come up on your own, and we might be tempted to adopt language/expressions that differ from your. Be careful not to lose your own voice and style, do not think that AI generated text is superior to your own voice and ideas.
In the next section we will dive deeper into how to make the decisions of where/when/how to use AI.