Table of Contents
Introduction – What Is GPT-4
Generative Pre-trained Transformer 4 (GPT-4) is the fourth iteration of the OpenAI‘s multimodal model that uses deep learning to conduct natural conversations with high accuracy through access to proprietary technology and research methods. Months in the making, this new iteration goes beyond text inputs to include image inputs to produce responses.
Unlike the other iterations, OpenAI has not specified how many parameters GPT-4 has built-in. It is likely higher than the 175 billion stated for GPT-3. Below is the developer’s live stream from March 14, 2023, which provides more technical information on the GPT-4. Beyond the capabilities this system has, Greg Brockman, President and Co-Founder of OpenAI spoke of its limitations we will discuss below.
Differences Between GPT-4 And The Previous Versions
Using the updated ChatGPT Plus, I asked the model to provide a summary of the different iterations of GPT starting with GPT-1 released five years ago.
ChatGPT: GPT-1 was the first version of the GPT series of language models developed by OpenAI. It was released in 2018 and had 117 million parameters. Despite its relatively small size compared to later versions, GPT-1 was already able to perform a wide range of natural language processing (NLP) tasks such as language translation, question answering, and text generation.
GPT-2, released in 2019, was a significant improvement over GPT-1 in terms of size and capabilities. It had 1.5 billion parameters, making it one of the largest language models available at the time. GPT-2 was able to generate high-quality text that was almost indistinguishable from human-written text. However, due to concerns about its potential misuse, OpenAI did not release the largest version of GPT-2 to the public.
OpenAI GPT – From connectors to artificial intelligence based communicator (Image from OpenAI)
GPT-3, released in 2020, was an even more significant improvement over GPT-2. It had up to 175 billion parameters, making it the largest language model available at the time. GPT-3 could perform a wide range of NLP tasks, including language translation, question answering, text summarization, and even computer programming. It was also able to generate high-quality text that was often difficult to distinguish from human-written text.
ChatGPT is an artificial chatbot product created from GPT-3. It was released in November 2022 and quickly became the world’s premier natural language communicator. Its popularity convinced Microsoft (via its web browser Bing) and Google (via its Office suite) to incorporate the technology. While the community praised its speed and natural communication style, many are concerned over response accuracy and data source.
Pros And Cons For The Retail Industry
GPT-4 and ChatGPT provides many pros and cons for the retail industry. Here are a list of three key ones for each:
1. ChatGPT – Users can quickly search for information about a brand (e.g., location, operating hour, refund policy). For example, we asked ChatGPT about Costco’s return policy. Here is what it came up with (as at March 15, 2023):
Retail Mashup: What is the return policy for Costco?
ChatGPT: Costco has a generous return policy for most products purchased in their stores or online. They offer a satisfaction guarantee on all their products, which means that customers can return items for a full refund at any time if they are not completely satisfied.
For electronics, appliances, and other select items, the return period is 90 days from the date of purchase. Some items, such as cigarettes and alcohol, cannot be returned.
Costco’s return policy also allows for returns without a receipt. However, in these cases, the customer will receive a refund for the lowest selling price of the item.
For online purchases, Costco offers free returns through their website. Customers can initiate a return online and print out a shipping label to send the item back to Costco.
It is important to note that some products may have specific return policies or restrictions. For example, certain products may have a limited manufacturer’s warranty that supersedes Costco’s return policy.
Overall, Costco’s return policy is known for being very customer-friendly and accommodating. If you have any questions or concerns about a return, it’s best to contact Costco’s customer service department for assistance.
The actual policy can be found here. As a comparison, ChatGPT was able to pick up on the general 90 day return policy and included stipulations and exceptions.
GPT-4 – Brands with access to GPT-4’s API can create their own version of ChatGPT that could improve the customer experience journey from engagement (e.g., searching for a product or service) to acquisition (e.g., purchasing the product or service), and onboarding (e.g., maximizing product or service capabilities) to maintenance (e.g., addressing positive or negative experiences). The natural language communication style reduces the frustration and friction customers echoed with the current chatbot technologies. In the future, tone and words used can be analyzed for a smoother interaction.
The biggest drawback to ChatGPT is accuracy. OpenAI’s President continued to exert that the GPT model system is applying publically available and third-party sourced/licensed data in a real-time learning mode. As such, it is possible that the system will provide an incorrect output (such as the example shown below on time availability). Additionally, hallucinations1 will continue to occur though at a reduced frequencies.
1 – Hallucinations are ChatGPT responses that appear correct, logical, or rational but provide users with slightly to widely inaccurate information. Since the system response comes from different data sources, the term “garage in, garage out” would apply.
OpenAI stated in their technical paper that GPT-4 included more built-in filters to restrict or reduce the number of inaccurate responses. For subjective or less known searches that would introduce bias, users are asked to verify the outputs through other channels.
ChatGPT vs GPT-4 – accuracy (Information provided by OpenAI)
We continue our Costco example by asking ChatGPT about returns on diamonds:
Retail Mashup: What is Costco’s return policy for diamonds?
ChatGPT: Costco has a generous return policy for most products, including diamonds. According to Costco’s website, their return policy for diamonds is as follows:
“Diamonds: If you are not completely satisfied with your purchase, you may return it in accordance with Costco’s return policy.”
This means that you can return a diamond purchased from Costco for a full refund at any time if you are not completely satisfied. There is no specified time limit for diamond returns, so you can return the diamond even after you have worn or used it.
However, it’s important to note that the diamond must be in its original condition with all the accompanying paperwork and certificates, including the original diamond grading report, which typically comes from an independent third-party laboratory.
Costco’s return policy also requires that the diamond must match the grading report, so any alterations or damage to the diamond may result in a refusal of the return.
In summary, Costco’s return policy for diamonds is very customer-friendly and allows for a full refund at any time if you are not satisfied with your purchase, provided that the diamond is in its original condition with all the accompanying paperwork and certificates.
This is a very detail response to the question and is mostly accurate. However, it missed out on the following from Costco’s website:
Diamonds: 1.00ct or larger: Members returning a diamond over 1.00ct must also present all original paperwork (IGI and/or GIA certificates) at which time they will receive a Jewelry Credit Memo. Within 48 hours, our Costco Graduate Gemologist will inspect for authenticity.
What Should Retailers Do Next?
There are many new opportunities and pitfalls of using this new technology for retailers. Here are some items to consider as retailers continue to assess their position with the technology:
1. Ensure their customer-facing information/policies are updated and consistent. There is an increase chance that ChatGPT’s response will be incorrect if information is disorganized and inconsisent.
2. Monitor ChatGPT responses to key questions about the brand for accuracy regularly. Any flaws or incorrect responses regarding policies, customer directives, or news reporting should be reported to OpenAI as soon as possible.
3. Review how ChatGPT API can be used to replace the chatbot technology being deployed by the brand currently. Natural language processing could enhance the interaction between the brand and its customers/employees.
4. Review the GPT-4 technical report to assess how the brand can use its capabilities for research, reporting, data analysis, and communication, etc. While ChatGPT is the main product being circulated through the GPT system, there are other potentials available (e.g., inventory management, appointment scheduling, forecasting, product reserach, and compensation management, etc.)
5. GPT-4 and ChatGPT introduce new methods to collect data and engage. As they are not a pure Informaton Technology (IT) topic, others departments (including legal, marketing, operations, and human resources, etc.) should be involved in determining the implications of this technolgy to how they operate the customer experience journey and business operations.
GPT-4 and ChatGPT are relatively new technologies with significant media coverage. While they are not perfect, retailers should not sit idle and believe the fad will blow over. This is especially if they have been using chatbots or implementing articifical intellgience/deep learning. Retail Mashup will regularly explain new use cases for retailers especially in respect to customer experience and revenue generation. Subscribe for updates.
Follow this link for more information about the Retail Mashup content platform
Follow this link for more insights
Follow this link for more podcast episodes
Follow this link to participate in regular polls