3 Key Differences in ChatGPT and Azure OpenAI

In this vLog I discuss some misconceptions around ChatGPT and Azure OpenAI, to include:

  • Who owns ChatGPT, OpenAI, and how Microsoft got involved
  • Security and privacy concerns about Azure OpenAI and ChatGPT
  • How each of the services is consumed and billed

Take a look to find out more!

 ChatGPTAzure OpenAI
OwnershipOwned by OpenAI LP a for profit arm of OpenAI the non-profit who’s mission is to development of societyPart of Azure AI offerings as APIs, and investor in OpenAI for exclusive rights to technology generated
SecurityInsecure and open to the public. LLM is trained on dataset created in GPT 3 and currently only references data from 2021 and earlier. Questions and interactions are captured and can be used for further trainingSecure to an Azure tenant using GPT-4, GPT 3.5, Codex and Dall-e Requires access for tenant to reduce the chances of the AI to be used for malicious purposes Data is not stored by the prompt, and model is not trained on data added
CostsFree during preview stages, or paid for better hardware availabilityBased on a consumption model like other Azure Cognitive Services. Biggest expense is re-training the model you’ve deployed for your data

Here comes your CoPilot

The new age of large language models (LLMs), and the ability to accelerate various forms of novel thought is being cast upon us at a rapid pace. Just like an airplane copilot, we are seeing an explosion of tools in various areas to help us do our everyday jobs, making us more productive and freeing up additional time to enhance our creativity… Or play candy crush.

You have already seen a plethora of announcements from Microsoft about various copilot tools that are being added to their Office productivity suite to assist the common office worker, GitHub CoPilot for assisting the software developer with writing, analyzing, and documenting code, the data analyst using Power BI and Microsoft Fabric for simplifying the analysis and report building process that could be tedious, and this is just the beginning from their standpoint. They’ve also announced the AI CoPilot software development kit that allows developers to add a CoPilot to any number of applications that are used throughout the business and consumer worlds for assisting people with their everyday tasks and simplifying the process by which they were able to develop and create new pieces.

The real question that comes to mind, however, is “who gets credit for the work that gets created?”. When we see situations such as in the entertainment industry, where movie scripts are being created by these tools, thousands of songs have been recently removed from Spotify due to the fact that they were generated with AI, images and videos are being developed and manipulated with AI, this question comes to mind frequently. And this is just the beginning of what I anticipate will be a massive explosion of questions around who really should get the credit for what’s being created. If AI is helping individuals complete their work at a faster pace, and the broader community is benefiting as a result, does it really matter? If I read 10 articles on copilot, and I am able to retain a significant portion of what I read, then turn around and form an opinion, and write my own article, such as I’m doing here about how I see things happening, is that still my work? Is it my work even though I am writing it based on a whole bunch of other material that others produced, and I’m summarizing in a slightly different way? This is the process by which the majority of research has been based for centuries now, as well as many fiction and non-fiction works have been created. Is that really different when we look at the technology that underlies LLMs?

In the world of data science, we can see tremendous opportunity to take advantage of already created machine learning models based on the algorithms that were used to be able to then replicate the findings for any number of various data sets and opportunities to create new algorithms and predictive models. Is this somehow “cheating” suddenly? Are these data scientists working, hopefully, towards the greater good, and having the novel inspiration for what they want to build, but are using these tools to help them produce it more quickly, cheaters? I think these are the questions we need to be asking ourselves rather than pointing fingers at who the people are using the tools to produce the work that they are producing.

The other major concern coming from all of this are the privacy and security implications of training these LLMs with information that ultimately should not be shared. Microsoft is providing excellent options with regard to these concerns by allowing customers to create their own instance of the various tools provided, such as ChatGPT or Dall-E APIs, that allow their customers to isolate the models that are trained specifically for them in their own Azure Subscription and are not using those individual models or data that is collected to train any of the other models. Using tools such as Google’s, Bard or OpenAI’s ChatGPT interface, you do not have the same luxury, and those models are being re-trained by all the data that is fed into them. This very example was made loudly public recently when some engineers from Samsung fed some of their data into ChatGPT, exposing corporate secrets unknowingly. This is also causing rash decisions by CxOs to widely ban the tools that are helping their employees be more productive. Clearly more education is needed for helping with these decisions and scenarios at all levels of corporations, education, and individuals across the board.

As a writer myself, recently just finishing publishing my first book, the days of writer’s block and challenges with getting started on various topics in chapters, still haunt me. I see these tools as an opportunity to help us get past that and produce work that much more quickly. Truthfully, the answer here is subjective to each person’s opinion, similarly to how a person feels about a painting, song, or written piece. We already base so many of our works of art, whether in the form of an application, methodology, algorithm, and more traditional artistic stylings on the knowledge we’ve acquired through experience, that the notion of “original thought” is now so uncommon, and even when introduced, rejected by the greater society, how is this any different? I say we take advantage of the tools we are given and make the copilot as pervasive as possible to help gain efficiencies in every aspect of the modern world!