ChatGPT exploded onto the scene around a year ago now, by now you have surely heard of it, but if you’re a little unsure as to what it is, ChatGPT is an AI model designed to generate detailed text responses. The quality of these responses heavily depends on how well the prompt is crafted. A good prompt requires clear, specific information and an understanding of your goal.
Effective AI prompting is a vital skill for professionals. It’s about communicating your queries and requirements to generative AI models like Chat GPT, helping you achieve your objectives more efficiently. By crafting effective prompts, you can optimize your interactions with AI, enhancing the quality of outputs and understanding AI’s limitations.
When setting up your prompt for Chat GPT, think about what you’re trying to get out of it, who your audience is, and the subtleties of your language. Research suggests that choosing the right prompt can significantly boost performance in models like GPT3. This is especially true for larger models or more complex tasks. Prompt engineering – the art of refining prompts by selecting the right words, phrases, symbols, and formats – is key to getting the best possible result from AI models.
Clear and Detailed Instructions are the Key
The key to getting the most out of Chat GPT is specificity in your prompts. The more detailed and precise you are, the more tailored the AI’s responses become. Being precise cuts down on the AI’s need to guess, leading to more relevant and accurate answers.
Make sure to clearly lay out the steps you want the AI to take. This not only ensures clarity but also helps in breaking down complex tasks into manageable actions.
Providing examples can greatly aid the AI in understanding the context and the level of detail you’re expecting in the response.
Utilising Reference Materials to Enhance Reliability and Reduce Fabrications
If you’ve been on social media lately, you’ve probably noticed two sides to the AI debate: those for and against its use. Critics often argue that AI merely repurposes existing content, lacking originality. However, for those of us using AI to streamline our daily tasks, this capability is exactly what we need.
When working with AI, guide it to use specific reference materials, like links to PDFs or websites. This approach helps anchor its responses in factual content. Instruct the AI to refer to your previous work if you’ve tackled similar tasks before. Enhancing the credibility of AI-generated content can be as simple as asking the AI to include citations from these reference texts, particularly for specialized subjects. This not only adds a layer of trust but also ensures the information is grounded in established knowledge.
Simplify Complex Tasks for Better Accuracy, Patience Leads to Precision
Like any large complex task, break them into simpler, modular components. This reduces error rates and makes the AI’s responses more manageable and coherent. For example, if you need to summarise lengthy documents, complete the summary piece by piece to stay within input limits, this will allow you to construct a comprehensive summary iteratively with more accurate outputs.
Encourage the AI to take its time in formulating responses. This can significantly improve the accuracy and depth of the answers. Asking the AI to review its previous responses can help refine and enhance the information quality.
Request a “chain of thought” approach for complex queries and use follow-up questions to refine and expand on initial responses.
Continuous Refinement of Prompts
The arrangement and design of prompts can profoundly affect a model’s output. Specific patterns and sequences can enable models like GPT3 to achieve state-of-the-art performance, highlighting the importance of thoughtful prompt construction. Regular refinement of chat GPT prompts will ensure their effectiveness. This includes revisiting and rephrasing prompts based on performance data and audience feedback.
A well-designed prompt can also help reduce biases and enhance the factuality of model responses. It’s possible to guide models towards more neutral and accurate outputs by incorporating certain instructions and selecting prompts carefully.
Contextual details act as catalysts in enhancing the performance of AI chat models, especially with ChatGPT. They assist the model in discerning the precise scenario, improving the accuracy of its responses. These details range from background information to user preferences and specific facts, tuning the responses towards more relevance and precision.
Context sets a clear path for the AI, providing a greater understanding of the scenario aids in producing appropriate responses. The more accurate the details are, the better the AI will be at generating relevant responses. In essence, the inclusion of these details is an integral part of an effective chat prompt.
But Why Is Any of This Relevant to Me?
With ChatGPT’s emergence, AI is now accessible to everyone, marking the start of what I believe is a massive leap forward across all industries and markets. The conversation around AI’s impact on employment is heating up. Many fear AI might replace human jobs, and some see people who utilise AI as “less capable” than their non-AI counterparts.
However, I see it a little differently. In my view, those who embrace AI to enhance their day-to-day tasks will be the real winners. The threat to jobs doesn’t come from AI itself, but from a reluctance to adapt, evolve, and make the most of these innovative tools at our disposal.
Examples (click to expand)
Here are two examples of me asking ChatGPT to do the same thing. In the first example, I ask
Write a section for a BIM Execution Plan that describes the proposed level of accuracy and tolerance of the model.
and the results are as follows:
This section outlines the intended level of accuracy and tolerance specifications for the Building Information Modeling (BIM) project. Our aim is to ensure that the model provides precise and reliable data, aligning with the project’s requirements and industry standards.
Accuracy Standards
- Dimensional Accuracy: All model elements will be represented within a tolerance of ±10mm, ensuring the integrity of spatial coordination and alignment with the actual construction.
- Geometric Accuracy: The BIM model will maintain geometric accuracy by closely representing the shape, size, and orientation of each component as per the design specifications.
- Level of Detail (LOD): The model will adhere to LOD 300 for architectural and structural elements, and LOD 350 for MEP (Mechanical, Electrical, and Plumbing) components, as defined by the BIM Forum’s LOD Specification.
Tolerance Specifications
- Structural Components: A tolerance level of ±5mm will be maintained for structural elements such as beams, columns, and trusses.
- Architectural Elements: Non-structural elements, including walls, doors, and windows, will adhere to a tolerance of ±10mm.
- MEP Components: MEP components will follow a stricter tolerance of ±5mm due to their critical nature in building services.
Quality Control Measures
- Model Auditing: Regular auditing will be conducted to ensure adherence to the prescribed accuracy levels. This will include spot checks and comprehensive model reviews at key project milestones.
- Collaborative Reviews: Regular coordination meetings will be held with all stakeholders to review model accuracy and address any discrepancies immediately.
- Software and Tools: Advanced BIM software, equipped with precision tools and error-checking algorithms, will be utilised to maintain high levels of model accuracy.
The proposed levels of accuracy and tolerance are designed to ensure that the BIM model accurately represents the intended design, facilitating effective decision-making and coordination throughout the project lifecycle. Adherence to these standards is crucial for the successful delivery of the project.
Okay, so not bad, right? Anyone can write a BEP now! Well, not quite. As someone who has been using ChatGPT since it was released, it is easy to spot both text written by ChatGPT and even easier to spot when people without an understanding of the topic at hand are relying on GPT to produce their work.
The real issue arises when the author doesn’t grasp the content ChatGPT produces. If they don’t understand the ‘why’ behind what’s been written, or if there are any errors in the text produced, enforcing or implementing it becomes a major challenge.
Now, let’s look at a refined version of the prompt, incorporating documents I’ve personally written as reference material:
Using the attached examples as source information, write a section for a BIM Execution Plan that describes the proposed level of accuracy of the models. Provide any tolerances for specific disciplines in a table format. Do not include any information about LOD. Write the section using plain english. The section must be succinct. Do not use unnecessary adjectives.
this time around, the response is much more appropriate for what we’re after:
Section: Model Accuracy and Tolerances
The accuracy of BIM models is critical to the successful delivery of our project. This section outlines the expected precision levels for different disciplines and their respective tolerances.
Discipline Tolerance Architectural ±5 mm Structural ±5 mm MEP (Mechanical, Electrical, Plumbing) ±10 mm Civil ±15 mm Landscape ±20 mm Interior Design ±10 mm Note: These tolerances are set to ensure that the models provide a reliable representation of the physical components. They are applicable across all stages of the project, from initial design to final construction.
I’ve always maintained that BIM and design automation aren’t here to replace us, but to enhance what we can do. Artificial Intelligence is no different. You need clear, specific communication and a strategic approach to use AI effectively. Stick to these principles, and your interactions with AI will be both efficient and fruitful, yielding more precise and dependable outcomes. Our methods of engaging with these advanced systems must evolve as technology progresses. This adaptation is key to unlocking more creative and efficient uses of AI in our everyday work.
Excellent article on ChatGPT prompt writing. This echoes on my understanding that clear communication and strategic approaches are vital for effective AI use.