OpenAI / ChatGPT in business use cases with Azure

We know that its the talk of the town, so we also planned to explore using OpenAPI in real life business scenarios.

So requested the early access on Azure and we received the same in couple of weeks time, so we are sharing the experiencing the same for your learning.

At the moment the services are not in general availability, its still under limited access, so you need to request to get the access in azure, so that you can create OpenAI resources in Azure on your subscription

What’s OpenAI

Open AI is a company who is currently creating AI/ML Models in multiple areas by providing a generous mount of data available in internet and train them.

What all are the available models

OpenAI creates multiple models in multiple areas as we can see below,

Instruct Models

language Models deals with natural language processing, and the old one among the same is ADA, then BAGGAGE, CURIE and DAVINCI in sequence. DAVINCI is the latest and capable one at the moment,

Image Models

DALL-E is the model used for the same and that will generate image from text prompts

Audio Model

Whisper is the model used for the same and that will translate audio to English

Now the instruct models are available in azure using OpenAI resource in azure, we can use the model and deploy in our azure environment for further exploration. Currently DAVINCI in azure not supporting custom fine tuning, rest all allows the same to fine tune. But still still capable and more efficient than others

As we don't have other models available in azure at the moment, we will be discussing only about the instruct models for our use cases. Now lets see the specialties of the model compared to others

Specialties

1.Advanced Natural Language Processing (NLP) Technology and Accuracy

ChatGPT uses advanced NLP techniques to understand and respond to customer inquiries, leading to higher accuracy and relevance compared to other AI chatbots

2.Personalization and Contextualization

ChatGPT can analyze and remember customer data and preferences, leading to more personalized and contextualized responses.

3.Training and Data Requirements

ChatGPT requires large amounts of high-quality training data to improve its performance, leading to potentially higher costs and data requirements compared to other AI chatbots.

4.Integration and Customization

ChatGPT can be integrated with existing systems and processes, and customized to specific business needs and use cases.

5.Cost and ROI

ChatGPT may have higher initial costs and training requirements, but can lead to higher ROI and cost savings in the long term due to its accuracy and efficiency.

Primary Use cases

There are N number of ways we can use the same, but as its name suggests it more related to chat.

• Customer Service and Support
• Sales and Marketing
• HR and Recruitment
• Operations and Management

OpenAI in action

Now lets see the same in action, for that lets see a conventional bot application flow, Bot will get triggered by user and then it will provide options based on the same and it may collect some information from its services and share the data in specific format, but its not capable enough to answer a deviated question than its been configured

Now lets see the same in the world of OpenAI, here the response from either API or Cognitive Search or Any similar source data will be given to OpenAI model for summarization, post the same, the session will be able to answer any kind of question around the same.

Here the advantage is that the subsequent questions around the data will answered with the intelligence using the model.

Unlike the conventional bots available in the market, this needs no specific questions to be framed to get the answers,

Now lets see the same in code, here we are giving the Chat GPT playground instead of code, we can use the Rest API endpoints along with he key to accomplish the same

Now if am providing a json to the service and provide “Tl;dr” at the end of the prompt, the same data will be summarized by the model. Along with the same, the same will be considered for subsequent questions to personalize the response on the same session

Now I am giving a dummy sales json and trying to summarize,

After the same if am asking some questions around the same will be answered accordingly.

This will provide endless options.

Hope this will help you to get some understanding around the same.

along with positives, this will also have negatives.

Things to note

Language and Cultural Limitations

• ChatGPT may not be able to fully understand and respond to queries in languages or dialects it is not trained on, or with cultural nuances it is not familiar with.
• It may also have difficulty with slang, sarcasm, or humor, leading to misinterpretation of customer inquiries.

Lack of Emotional Intelligence

• ChatGPT may have difficulty understanding and responding to complex emotional states or providing empathy, leading to poor customer experience.
• It may also have difficulty detecting and responding to customer frustration, anger, or dissatisfaction.

Limited Creativity and Innovation

• ChatGPT may not be able to generate novel or creative ideas, leading to limitations in certain use cases such as marketing or product development.
• It may also have difficulty adapting to new or unexpected scenarios, leading to inaccurate or inadequate responses.

Data and Privacy Concerns

• ChatGPT relies on large amounts of data to train and improve its performance, leading to potential privacy concerns for customers and businesses.
• It may also be vulnerable to data breaches or cyber attacks, leading to potential data loss or misuse.

Misconceptions about Capabilities

• ChatGPT may be perceived as a fully autonomous and intelligent entity, leading to misconceptions about its capabilities and limitations.
• It may also be perceived as a replacement for human customer support or management, leading to potential resistance or misunderstanding from employees and customers.

Its too early to take a decision, as its still mature and need to explore more.

Now you can also do your exercise and provide feedback around the same.

Happy coding.. :)