Breaking the Barrier of Miscommunication With Prompt Engineering

Use Cases & Projects, Dataiku Product, Scaling AI Christina Hsiao

If there's one thing Generative AI has taught us, it's that people aren't always great at articulating what they want or even knowing what they want. Think back to your first interactions with ChatGPT, Stable Diffusion, Bard, or any other prompt-driven Generative AI application. Like me, you probably underestimated their capabilities, posing a relatively simple question or request with phrasing similar to a Google search. However, the response, while shockingly impressive, wasn't quite what you were after. 

It became clear that the more detail and precise instructions you provided with each subsequent query, the more satisfying the output became. Reflecting upon the final iteration, it probably would be fair to say that what you accepted in the end was actually pretty different from what you asked for the first time. This phenomenon, though frustrating, is not limited to AI — it's a universal human quirk that often prevents us from clearly articulating our desires until we see or experience something that's NOT right.

A lot of times, people don't know what they want until you show it to them." -Steve Jobs

This elusive, “moving target” experience surfaces in many everyday situations: dating, job hunting, buying a house, and the classically infuriating, “What do you want to eat for dinner?” exchanges. 

But this phenomenon is not confined to personal matters; it frequently occurs in the workplace, leading to wasted time and frustration between business and technical teams. Imagine a scenario where the business identifies a seemingly straightforward need, such as a data report, BI dashboard, or custom application. However, after months of back-and-forth iterations, it becomes evident that something was lost in translation. Does this sound familiar?

Similar to our encounters with ChatGPT, the common thread here is that people often struggle to ask the right question or define requirements accurately from the start. It's not anyone's fault — it's simply the reality that sometimes you need to see an imperfect deliverable before identifying what's truly needed. Properly articulating the desired outcome from the beginning is crucial for efficiency in any data initiative, especially when Generative AI models come into play.

Enter Prompt Engineering: The Gateway to Better Results:

Prompt engineering has emerged as a new discipline, fueled by the recent interest in Generative AI and Large Language Models (LLMs). It is the practice of developing and optimizing effective prompting techniques to interact with LLMs and other tools, ultimately achieving better results from these powerful models. 

While using LLMs for one-off inquiries is useful, the real opportunity for enterprises lies in embedding these models into existing data pipelines and projects, applying them at scale against large datasets. The key to success lies in crafting precise, thoughtfully constructed AI prompts infused with relevant business context and clear instructions.

Introducing Prompt Studios in Dataiku

At Dataiku, we've harnessed the power of prompt engineering through our innovative Prompt Studios feature (coming soon). This tool allows you to design, test, and operationalize the optimal AI prompts to achieve your business goals.

Prompt Studios in Dataiku

The Prompt Studios interface provides sections to explain tasks in plain English, add examples of inputs and expected outcomes, and even include test cases or reference existing datasets for real-world performance evaluation. With each prompt test, you can assess the quality of results both qualitatively and quantitatively, along with reviewing the estimated cost of running the query against 1,000 records. This empowers you to compare and evaluate different prompts and models effectively.

test cases

Sample results and estimated cost for a prompt asking GPT to detect topics of interest in financial news articles

Once you're satisfied with the results, the final step is deploying the prompt to the project Flow in Dataiku. This saves the AI enrichment step as a visual recipe, making it not only efficient to reuse but also visually apparent to your team that Generative AI has been applied. Others on your team can easily inspect and validate both the final prompt logic and resulting outputs. In short, Dataiku's Prompt Studios provide the ultimate tool for teams to engineer impeccable AI prompts, maximizing the value derived from Generative AI models.

image6-Jul-12-2023-04-26-23-2946-PMPrompt Recipe: Reusable, Transparent, and Responsible AI Enrichment

Bridging the Communication Gap and Boosting Efficiency

Just like the responsibility lies with us to provide clearer intent when asking our children to complete a chore, it also falls on us to refine our communication when working with Generative AI models and peers alike. As prompt engineering and conversational AI tools become increasingly commoditized, we have high hopes that as a whole, the human race will improve at communicating more effectively.

Prompt engineering is the key to bridging the gap between technical and business teams, streamlining data initiatives, and unlocking the true potential of Generative AI to drive transformative outcomes in your organization.

Are you ready to revolutionize your data initiatives and achieve unparalleled efficiency with prompt engineering? Explore the possibilities with Dataiku's prompt studios and embark on an exciting journey into the world of Generative AI.

You May Also Like

How to Address Churn With Predictive Analytics

Read More

What Is MLOps?

Read More

4 Do's and Don'ts of Hiring and Upskilling for AI Talent

Read More

Technoslavia: Navigating the Data World in the Age of Generative AI

Read More