Few-shot Prompting: What it is and why it matters for small businesses

Giving small businesses a leg up

Few-shot promting, by Midjourney
Few-shot promting, by Midjourney

For decades, research in AI has been making great strides. Various techniques, such as machine learning (ML) and Natural Language Processing (NLP) have been increasingly put into production over the past few years. However, actual, practical use has only been within reach of larger organizations.

For smaller businesses, the two main impediments to adopting AI have been (1) the huge quantities of data required to build models that are capable of providing useful results, and (2) the cost of hiring data scientists and implementing processes to corral all that data. As AI usage becomes mainstream, these costs have been slowly reducing. Nonetheless, the hurdle has remained quite high.

That is, until the introduction of ChatGPT.

ChatGPT has stirred the pot. It is not such a giant technological leap in itself, but rather a flashy announcement proclaiming: "AI is here!" People have taken notice in a big way.

The reason that ChatGPT is such an exciting prospect for small businesses is that it makes AI immediately available for daily business use. This possibility is in large part due to its ability to provide quality results with few-shot prompting.

In this article, I will describe what few-shot prompting is, and why it matters so much to small businesses that desperately want to gain an edge.

Working with ChatGPT

ChatGPT does not "think". It is not human, and it is not an independent being. It has to be told what to do.

ChatGPT is a large language model trained by OpenAI to generate human-like text based on some information provided to it. The model will "react" to the input by generating text that it calculates as being the most relevant.

These inputs are called "prompts", and the output generated by the model is called a "completion". So the way that you use the technology is by prompting it to generate a completion. The prompt provides the model with context, and the completion is the model's response based on that context.

Training

Working with ChatGPT by prompting it for a completion is possible because the model has already been trained on a large corpus of text data, which allows it to have a general understanding of language and the ability to generate text that is grammatically correct and semantically meaningful.

The model learns patterns and relationships in the training data, which allows it to understand the meaning of words, phrases, and sentences. This understanding allows the model to generate text that is similar to the input it has seen.

So, because the model has already been pre-trained on such a huge quantity of data, all you need to do is prompt it. The training data provides the knowledge base it will call upon when you trigger it with your prompt.

Prompting

Training the model is the hard part. The heavy lifting was already done by OpenAI, so ChatGPT is ready for use by you. To make practical use of it: (1) you need to integrate it somehow into your existing workflows, and (2) you need to prompt it to get the right completion. This article focuses on the second point: prompting.

To elicit a response from ChatGPT, you need to tell it what to do. This is called "prompting". Prompting provides the context that ChatGPT will use to do its thing. There are many ways of prompting, but for our purposes, let's just say that prompting is like providing an instruction. For example: "write an email reply informing that I will be away for the weekend, but that I will gladly reply on Monday."

Even with just this instruction, ChatGPT will write a plausible email reply. If you also provide the original email, it will use information from that email, such as the sender's name, to craft a customized reply. Magic!

However, without more context, ChatGPT can tend to go a little wild. You are running a business after all, so the text needs to be professional and somewhat consistent. The goal is to automate work so that you can serve your customers better, not automate rude or incorrect responses that will make your customers seek out your competitors.

0-shot, 1-shot, and few-shot Prompting

Without providing further, specialized training, additional context can be provided in two ways: (1) by including more details in the prompt, or (2) by providing some examples along with the prompt. The latter is called "few-shot prompting". Few-shot prompting includes the special cases of 0-shot and 1-shot prompting.

A 0-shot prompt is used to prompt the AI to generate text without any additional training data or examples. An example of a 0-shot prompt is: "Write a product description for a new coffee maker called 'SmartBrew', it has a programmable timer and a built-in grinder." If you were to prompt ChatGPT like this, it would happily chirp out an answer. You provided enough context for it to generate text. The text would appear to be eerily human-like. It would have no grammatical errors and sound logical enough. However, most likely it would not express concepts or ideas consistent with your business.

A 1-shot prompt uses a single example as input to generate text. The model is given one example of text that is similar to the output it should generate, and it learns from that example to generate text that is similar to the input. If a small business has already spent significant time designing a product description for an existing product, they could use that as input for their second product. The model will likely pick up on their intended style and produce output that feels similar to the original. An example of such prompt could be: "Generate a product description for our new product 'GenuisBrew', similar to the description of our 'SmartBrew' coffee maker." [sic, because it's funnier in this case to leave in my spelling mistake]

A few-shot prompt uses a small number of examples as input. These examples are similar to the output it should generate, allowing the model to learn what you expect it to do. OpenAI recommends that you "show and tell" when prompting ChatGPT, meaning that you will get better results if you show the model what you want, rather than just telling it what you want. The example below of a few-shot prompt is taken directly from the OpenAI website:

Marv is a chatbot that reluctantly answers questions with sarcastic responses:

  You:  How many pounds are in a kilogram?
  Marv: This again? There are 2.2 pounds in a kilogram.
        Please make a note of this.

  You:  What does HTML stand for?
  Marv: Was Google too busy? Hypertext Markup Language.
        The T is for try to ask better questions in the future.

  You:  When did the first airplane fly?
  Marv: On December 17, 1903, Wilbur and Orville Wright made the
        first flights. I wish they’d come and take me away.

  You:  What is the meaning of life?
  Marv: I’m not sure. I’ll ask my friend Google.

  You:  Why is the sky blue?

Why does this matter for small Businesses?

Let me be clear: if you are a small business owner, ChatGPT matters.

Yes, there is a lot of hype and noise. No, it is not a perfect technology and can produce spectacularly bizarre junk. Sure, some people are up in arms about how it could allow students to cheat, or criminals in Nigeria to produce grammatically correct phishing emails. These are all real concerns.

How the technology will be used and offered commercially is still up in the air. The computational resources required are quite expensive compared to "traditional" computing methods. How ChatGPT can be integrated into a business setting safely and non-intrusively is something we are only beginning to figure out.

However, the fact of the matter is that AI is now mainstream. It is now within reach of ordinary people and small businesses alike. Gone are the days when millions of dollars were necessary to initiate an AI project. OpenAI has built this sophisticated language model so that you don't have to. You no longer need petabytes of high-quality data or teams of data scientists to spend months on end developing a highly specialized recommendation engine for you. You can now start using AI directly and immediately in your day-to-day work.

Thanks for few-shot prompting, AI is here, right now, and is ready for you. Are you ready for it?

AI4B2B is a cutting-edge technology company that specializes in applying AI to help businesses work together more efficiently.

If you're looking to improve your business processes and take advantage of the latest AI technology, look no further than AI4B2B. Let us help you unlock the full potential of AI for your business today.

Quick index:

  • Deep Learning
  • Few-shot Prompting
  • Language Model
  • Machine Learning (ML)
  • Natural Language Processing (NLP)