Using ChatGPT in Azure OpenAI Services

Microsoft has recently launched support for the language model that powers ChatGPT as part of their OpenAI services on March 9th. Previously, I wrote a blogpost on comparison between OpenAI and Azure OpenAI here –> OpenAI vs Azure OpenAI – However, with the availability of ChatGPT in Azure, some aspects have changed. Hence, in this blog post, I will discuss the differences and capabilities between the consumer edition of ChatGPT and the version of ChatGPT that is available in Azure.

Since the feature is currently in preview it is only available in two regions East US and South-Central US. So, after I deployed a new instance of the OpenAI services in East US I got access to the new language model, dubbed gpt-35-turbo. When I go into the OpenAI Studio I have a new option called ChatGPT playground.

I can customize the initial behavior of AI through supervised training and the definition of various parameters such as temperature and top-p. “Temperature” and “top-p” are parameters used to control the creativity and randomness of the generated responses.

The “temperature” parameter determines how random and unpredictable the generated responses will be. Higher temperatures result in more surprising and varied responses, while lower temperatures result in more conservative and predictable responses.

The “top-p” parameter, also known as nucleus sampling, controls the diversity of the generated responses by selecting the most likely words from the model’s probability distribution. The parameter specifies a threshold probability below which fewer probable words are excluded from consideration. The higher the value of top-p, the more diverse and unpredictable the generated responses will be.

In summary, adjusting the temperature and top-p parameters allows us to control the balance between creativity and predictability in the generated responses of the engine.

One significant difference that Azure ChatGPT offers compared to the standard ChatGPT is the inclusion of Internet access. This means that I can request Azure OpenAI’s ChatGPT to access a particular website and generate a summary for me. Allow me to provide an example to illustrate this concept. I will request the service to establish a connection to a particular website and generate a summary of its content. Not sure if this is by design, I hope so since if I use another prompt it refuses to connect to the website.

The interesting part is that Microsoft also has integrated different security mechanisms into the service so when I tried to see if the service could access specific malicious sites they were automatically detected and blocked as seen in the screenshot below. Fantastic addition!

You as with the other Azure OpenAI API’s have access to the same underlying API’s.

In terms of the data that is available, is seems to be much of the same base as with in ChatGPT when it was launched, and only has data uptil mid of 2021 and as well, we can also see that it supports different formatting such as rendering HTML tables with content, much as the same as with ChatGPT. The biggest change (if by design) is support for outbound network connectivity to parse content there.

In an upcoming blogpost I will set up this demo that is able to use company data to parse, index and display the information back to the user

Leave a Reply

Scroll to Top