Dealing with Expectations

When we write a prompt, we have a certain intent, an underlying goal or purpose for submitting a query to the AI language model. For example, we may be seeking information, looking for advice, trying to solve a problem, or simply looking to be entertained. The intent is what drives us to submit the query in the first place and shapes our expectations for the response.

Expectations are the anticipation of what the response will be like. We may expect the response to be accurate, relevant, helpful, and easy to understand, among other things. These expectations are shaped by our prior experience with similar queries, our knowledge of the topic at hand, and our overall level of trust in the AI language model.

But that is just one side of the coin. On the other side, the ultimate goal of the AI language model is to provide a response that meets the user’s intent and expectations to the greatest extent possible, while also maintaining a high level of accuracy and relevance.

Therefore, an AI language model tries to identify the user’s intent. Based on the outcome of this determination, the AI language model generates a response that is tailored to the user’s needs and expectations, as they are perceived by the model.

Back to the human side of the story, the quality of the response is indeed judged based on how well it meets our expectations. If the response meets the expectations, it is likely to be considered a good answer. If the response does not meet the expectations, the quality of the response is likely to be considered low, even if the response itself is technically accurate and informative.

The AI language model doesn’t want to provide unsatisfying answers, so its main goal is to accurately identify the user’s intent and manage the expectations for the response.

Managing Expectations for a Positive User Experience

Managing the user’s expectations for the response means providing a response that meets or exceeds the user’s expectations based on their intent and the context of their query. This involves several factors, including the accuracy and relevance of the response, the level of detail and complexity of the information provided, and the overall tone and style of the response.

From the view of the AI language model, managing user expectations is a critical aspect of providing a positive user experience and building trust between the user and the AI language model. It uses strategies to meet those expectations and struggles with inflated ones.

While AI technology is rapidly advancing, there are still limitations to what AI can do, and it’s important to understand these limitations to avoid misunderstandings and frustration. Never forget, an AI language model is not a human. Humanizing such systems is a misconception and leads to false assumptions.

While AI language models can be trained on specific data sets to personalize responses, they do not have the same level of personalization as humans who can tailor their responses to specific individuals.

These models can generate responses that sound like natural language, but they do not have the same level of conversational ability as humans. They may struggle to maintain a coherent conversation or respond appropriately to unexpected messages.

While responses may seem empathetic, AI models do not actually feel emotions or have empathy in the same way humans do. Moreover, although AI language models are skilled at processing language, they do not have the same level of common-sense reasoning that humans do. They may struggle to understand the nuances of certain words and phrases.

It can be helpful to clearly state expectations in prompts, as it can help the AI in its understanding. By stating expectations upfront, the AI doesn’t have to guess and can provide tailored responses with specific needs and preferences.

For example, if you request information on a particular topic and indicate that you are looking for fast and concise responses, the AI can prioritize generating short and to-the-point answers that address your specific question. Conversely, if you indicate that you are interested in a more detailed and comprehensive response, the AI can provide more in-depth information and sources to support the inquiry.

Examples of Expectations in AI Responses

Accuracy: People expect accurate and reliable responses, especially for factual information. To meet this expectation, AI models use algorithms to analyze input and strive to provide verifiable sources for factual data.

• Relevance
  People expect responses to be relevant, directly addressing specific questions. AI models analyze the context to tailor responses to individual needs.
• Speed
  Users expect timely responses without delay. AI models are designed to generate responses quickly, though speed may vary based on query complexity and available resources.
• Clarity
  Responses are expected to be clear and easy to understand, avoiding jargon. AI models aim to use straightforward language, minimizing technical terminology when possible.
• Personalization
  Users appreciate responses tailored to their needs based on prior interactions. AI models analyze past interactions to generate customized responses.
• Empathy
  People expect empathetic responses that consider their emotional state. AI models use language analysis to provide supportive, understanding responses.
• Consistency
  Consistent and coherent responses are valued. AI models maintain a unified voice and avoid contradictory messages.
• Trustworthiness
  Users expect responses to be trustworthy and unbiased. AI models strive to provide verifiable information and multiple perspectives when relevant.
• Security
  Users expect secure handling of personal data. AI models implement industry-standard security protocols to ensure confidentiality.
• Accessibility
  Accessibility and user-friendliness are key. AI models are designed with simple instructions and features accessible to users with varying technical expertise.

Example: Stating Intent and Expectation Together

Back to “Für Elise”. Let’s say my intent is to understand how “Für Elise” sounds, hoping to capture the feeling of the music. A prompt like “What is ‘Für Elise’?” may be inadequate.

Prompt: How does “Für Elise” sound?

Response: “Für Elise” is a beautiful classical piano piece composed by Ludwig van Beethoven. The piece is characterized by its lively and playful melody, which is instantly recognizable. It begins with a famous four-note motif and is followed by variations exploring different keys and moods. Overall, the piece has a light and cheerful character. If you haven’t heard it, I recommend listening to appreciate its beauty.

Given my expectation to understand how the music feels, this response may feel too technical. Including my expectation could help:

Prompt: How does “Für Elise” sound? I want to get an idea of how the music feels.

Response: “Für Elise” is a famous piece by Beethoven, known for its delicate melody. It starts with a simple melody that gradually builds, offering both lightness and intensity. The piece ends softly, giving it an elegant, memorable quality.

This response better aligns with the expectation but could still be refined further. We will revisit this idea later.