A new LLM-powered summarization

Key Takeaways

welcome to the Azure AI video series in this video you will learn about the new large language model powered summarization in Azure AI language service Azure AI is the backbone of copilot and in every key Microsoft product today Azure AI brings a One-Stop shop for you to experti app building in there you have pre-trained models that fulfill your needs right away customization that allows you to find tune on top of pre-trained and generative models for your specific needs as well as large language models that power you to build what was previously impossible summarization is one such capability case a fine-tuned larg language model targeted for summarization use cases that you will see in a minute summarization in Azure AI is a fine-tuned large language model that offers a fast trustworthy and cost effective way to condense content we classify summarization into two main categories based on input General documents and conversations with all apis you have access to a fine-tuned pre promp engineered AI models including GPT and other large language models this frees customers from the complexities of model building no concern about prompt tuning maintenance and model upgrades these powerful models is just a simple API callway however if you wish to retrain the model using your own data we offer a userfriendly interactive customization process that anyone can use the summarization capability is now available in containers for both connected and disconnected in addition to our Cloud offering this offers you the flexibility that will you need to apply our service for your specific needs I will now walk through a quick demo of the new summarization capabilities I will be using the rest API if you wish to code in your preferred programming language you can download the SDK following our documentation first you need to provide your API endpoint and key Additionally you need to specify the API version in general we recommend using our GA version for your production today's session we will be focusing on the new capabilities which are in preview so I specify the AI version as 2023 11 15 preview now let's dive into document summarization first let us see what we can get with the default settings I've select a recent news article from Microsoft price release about Microsoft merging Activision Blizzard this summary condenses the entire article and summarize key information out of it besides the summary itself you can receive many useful response information in Json for format which will help you streamline your automation workflow in this new release we introduce a descriptive way for you to specify your desired summary length for abstractive summarization we offer three options short medium and long understanding that customers often have varying interests and preferences with in enabled query focused summarization this feature tailor summary generation to focus on a specific topic with the same input content by specifying a query like settlement date we get a different summary than the default one the summary is now focused on settlement date in this deal instructive summarization also Al support summarizing on a specific Focus extractive summarization will preserve the original awarding and directly extract the relevant sentences now let's see the new capabilities for our conversation summarization we introduce two new aspects meaning recap and followup tasks in addition to our existing capabilities like narrative chapter title issue and resolution the recap aspect condenses a lengthy meeting or conversation into a concise one paragraph summary to offer a quick overview and the followup tasks aspects summarize action items and tasks that emerge during a meeting I will demonstrate the two new aspects in two scenarios a call center solution and a long meeting recording this is a call center case about a customer issuing a return the recap summarization gives a quick short descriptive report about this customer case then let's move on to a long meeting which lets about three minutes and spends more than 4,000 words we generate a recap to quickly grasp the meeting's core content then divided in two chapters after that we will identify the call to actions established during the meeting and provide a detailed Narrative of the entire meeting allowing you to gain a deeper Insight without the need to listen the entire recording you can also try the same functions in language Studio without a need of writing any code to get started with summarization check it out in the language Studio at aka.ms summarization D TR out all visited our documentation at aka.ms summarization D dos and and finally to learn more about Azure AI language please visit aka.ms aure Das language