Common Sense or Intelligence in Generative AI?

In the evolving landscape of Generative AI, particularly with Large Language Models (LLMs), the interplay between intelligence and common sense becomes crucial. Intelligence is essential for leveraging these models effectively, as they rely heavily on the quality of inputs, including instructions and context. However, common sense plays a pivotal role in shaping these inputs to guide the AI towards desired outcomes.

So, what is a prompt in GEN AI

Instruction + Context + Input Data + Output Format = PROMPT

The Role of Prompt Engineering: Prompt engineering is not merely about crafting instructions; it's about strategically designing prompts that blend common sense with intelligence to steer AI responses. This involves various techniques:

• Zero-Shot Prompting: In this approach, the AI generates a response based on its general knowledge, without specific prior training.

Example: It is like asking a kinder garden kid - what is Fourier series?Given the complexity of the mathematical topic, a child may respond with some answer, despite not being specifically trained or informed about the topic.

• Few-Shot Prompting: Various examples and demonstration is given to the model as prompt; and LLM is expected to learn and then produce output.

Example: Prompt:?"Identify the color of the following objects and match them to their names."

Example 1:?"A banana is yellow."

Example 2:?"The sky is blue."

Example 3:?"Grass is green."

Task for the AI:?"A strawberry is ___."

• Chain-of-Thought Prompting: Guides the AI through a logical reasoning process.

Example: Solving a simple addition problem. Visualise how would you teach a child on? how to solve a simple addition problem, such as adding two numbers together 2+3 . You would use a step-by-step approach to guide the child through the process.This method mirrors the Chain-of-Thought (CoT) prompting used with large language models (LLMs).

• Self-consistency Prompting: Ensures the AI's responses are consistent by evaluating multiple reasoning paths.

Example: Take an example of? teaching animal Identification to a group of school kids. The teacher wants to ensure that the children not only recognize the animals but also understand the reasoning behind each identification.

Steps: Teacher shows a picture of dog and explain it's a mammal because it has fur and gives birth to live babies. And she shows a fish and explain it's not a mammal because it lays eggs and doesn't have fur. Then, the teacher asks about a rabbit.

Context: This method ensures the AI's responses are consistent with the examples provided, enhancing accuracy.

• Tree of Thoughts Prompting: An advanced method that mimics human decision-making processes, allowing the AI to explore and select the most viable solutions from different reasoning paths.This method is believed to outperform other prompting methods according to Google DeepMind researchers.

In an era where technology often outpaces human adaptability, it's ironic that common sense can also become a rare commodity, especially in white-collar professions where individuals struggle to craft effective prompts to fully harness the capabilities of Generative AI. This paradox highlights the need for a balance between sophisticated technical knowledge and the intuitive grasp of human context that common sense provides. Both intelligence and common sense are indispensable in maximizing the potential of Generative AI.

For a deeper dive into the techniques and their applications, check out here and

here