Azure OpenAI Services in teaching and education

[Guest ajitjaokar_oxf]

Azure OpenAI Services in teaching and education

 

 

By Ajit Jaokar

 

Course Director: Artificial Intelligence - Cloud and Edge Implementations

 

University of Oxford

 

With contributions from

Ayse Mutlu Lead AI Tutor

Artificial Intelligence - Cloud and Edge Implementations University of Oxford

 

 

Introduction

 

 

With the advent of Large language models(LLMs) like GPT-3, we see a transformation in education. In this article, I present my views on the future of education in light of these developments. The views presented here are based on my teaching - but are a personal perspective.

 

 

 

First, GPT-3/chatGPT is a rapidly evolving space. For example, in my course at the University of Oxford (University of Oxford: Developing artificial intelligence), we first started working with OpenAI due to our collaboration with a liberal arts college in the USA. We helped them design a system for scriptwriters who collaborated with OpenAI GPT-3 to create characters in their script.

 

 

 

In this sense, we have worked with OpenAI longer than most people. However, after the release of chatGPT in late 2022, the rate of change has been phenomenal. Today, we see ChatGPT Whisper, Visual-GPT multimodal AI, and even talk of GPT-4.

 

 

 

Because of the rapid rate of change, my knowledge in this domain is limited. But, having said that, we have some clear ideas about where Azure OpenAI Services can apply to education.

 

 

 

Today, there is a lot of excitement and speculation about GPT-3, and it is natural to ask how intelligent GPT-3 is and whether it approaches human-level intelligence. But in many ways, that's the wrong question to ask. Instead, exploring the idea of how we can build ChatGPT-like functionality using our own data is more interesting.

 

 

 

When framed this way, we focus on the pragmatic and ignore the esoteric.

 

 

 

Also, in this blog post, we discuss Azure OpenAI Services - i.e. the integration of the Azure cloud platform with Open AI for applying large language models and generative AI for enterprise use cases. Azure OpenA Services is distinct from ChatGPT and GPT-3.

 

 

 

This distinction is the foundation of my perspective below, i.e. my responses relate only to Azure OpenAI Services.

 

 

 

Chatting to your own data

 

 

One of the criticisms of LLMS is that LLMs are merely text generators. Because they are trained on large amounts of data, they string words together based on the statistical probability of words following each other in sentence construction. More data is not the solution to this problem. LLMs need the concept of ground truth. One of the advantages of Azure OpenAI Services is that it partly overcomes this problem by taking a B2B perspective, i.e. chatting with your own data.

 

 

 

The Azure OpenAI Services allows you access to ChatGPT and lets you develop your enterprise apps using large pre-trained AI models. In addition, Azure OpenAI provides critical functionality like responsible AI, security, and REST API deployment. You can also filter and moderate the content of your users' requests and responses to ensure that coding and language AI models are used responsibly for their intended purpose.

 

 

 

The implementation of “Chatting with your own data" also involves other elements such as prompt engineering; citations and supporting content to support results; emerging interaction patterns(ex: breaking down a query and referring to external sources as needed); Semantic ranking, Summarization of responses, etc.

 

 

 

 

 

 

 

 

Image source: Microsoft

 

 

 

A co-pilot first workflow

 

 

A broader question is: How would we rethink AI workflow for LLMs like GPT-3?

 

 

 

GPT-3 democratizes AI - especially it brings in people who are not traditional AI specialists into AI. So we could rethink the workflow of AI. We call this as a co-pilot first approach.

 

 

 

Consider you are an AI product manager or AI project manager, i.e. a non-developer. We start with how to create your ideal assistant/co-pilot. And we can further break it down into:

 

• How will you select the use cases?
  
• How will you evaluate the use cases?
  
• To what extent can you use the LLM tools to generate code, visuals, and language models using prompt engineering and fine-tuning
  

 

In other words - can we start with the idea of a co-pilot/assistant collaborating with you for each task?

 

 

 

We are exploring these ideas in our teaching.

 

 

Using Azure OpenAI Services for Education

 

 

Based on this background and keeping the distinction between Azure OpenAI, GPT-3, and ChatGPT in mind, Azure OpenAI Services can create conversations with our own data providing personalized learning opportunities. Even more so, it can help create conversations for specific verticals. For example, we are considering this strategy for chatbots in Agriculture. In general. Personalization via Azure OpenAI Services-based conversations could make learning more inclusive for struggling students or students with special needs.

 

 

 

Today, learning has become virtual and hybrid. In this sense, a conversation agent fits in naturally. Tools like visual-GPT could also add multimodality in the future.

 

 

 

But deeper changes are needed. For example, I have been a fan of the 'reverse bloom/inverse bloom/flipped bloom' taxonomy. The idea is simple: i.e., flip the well-known Bloom's taxonomy and put creativity at the center of the learning process. Doing so changes the dynamic from “what you know” to “what you can apply”.

 

 

 

Not many people object to putting creativity at the center of learning.

 

 

 

 

 

Image source: plpnetwork

 

 

 

Evaluating and scaling such creativity is an entirely different matter altogether. There is also the question of creating work assisted by AI/ GPT-3, which concerns educators from a plagiarism/originality standpoint.

 

 

 

In the current system of evaluation, educators evaluate and assign a defined numeric score to compare candidates for subsequent career development. As the evaluation process is digitized, this assessment method may also change. Here are some thoughts on how we can use AI in the evaluation process

 

 

 

1. Prompts become the submission: we start with the GPT-3 response, and each student then independently evolves this baseline through a set of prompts.
   
2. A reflective process/diary created with the help of GPT-3 maintained by the student throughout the course which acts as a submission
   
3. Multiple perspectives/ modalities combining language, images, etc based on GPT-3.
   

 

 

 

We thus change the evaluation dynamic from “demonstrating what you know” to “demonstrating what you can apply”.

 

 

 

Nevertheless, this is a nascent area, and much work needs to be done. Here are some areas we need more work on. Some of these areas I am working on in my course:

 

 

 

1. A better understanding of the ground truth through linking Causal machine learning with LLMs coupled with critical thinking
   
2. An evolution of prompt engineering strategies
   
3. More details on how the co-pilot first workflow would work in the industry
   
4. Toolkits to implement these ideas ex on the lines of the inclusive design toolkit
   
5. Responsible AI toolkits in the context of GPT-3
   
6. Exploration of multimodality on the lines of visual GPT
   
7. Code development tools - especially for AI, i.e., the evolution of GitHub co-pilot, AI-builder, and Power platform
   
8. Testing strategies for domain experts.
   

 

 

 

Empowering your students for a co-pilot first world

 

 

 

Image source: Reddit

 

 

 

Much of this discussion may be even more accelerated if the industry adopts the 'co-pilot first approach.' If so, educators must follow this trend to keep up with the new job roles. We are already seeing this in the legal profession for training legal interns.

 

 

 

This will need a complete rethinking of many of the current ideas on education and the adoption of some new ideas that I proposed in this article.

 

 

 

In this case, the conversation changes from: 'chatGPT is used for exam cheating or not' to: How can I empower my students to take up jobs of the future if the co-pilot first mode of work becomes a default?'

 

 

 

Conclusion

 

 

 

 

 

 

 

I saw this image on Linkedin about teachers protesting the introduction of calculators in 1978. Of course, calculators are here in any case despite protests. That made me think: How will the world look like if we take a progressive view of change to create a more inclusive education system?

 

 

 

As a person on the high-functioning autism spectrum, creating an inclusive education system is close to my heart.

 

 

 

More enticingly: how will industry and jobs change in the co-pilot first world? What will that mean to pedagogy and evaluation?

 

 

We are exploring some of these ideas in our courses at the University of Oxford: Developing artificial intelligence.

 

Continue reading...