Generative AI has achieved impressive results, from passing bar exams to writing songs, but some companies struggle with how to apply it to their own operations.
They know it can be a transformative technology. But they don’t know where to start using generative AI and large language models (LLMs) to achieve their business goals, said a panel of data industry insiders recently held in Las Vegas hosted by data cloud vendor Snowflake. He said it at the Snowflake Summit, a user conference.
However, according to Ali Dalloul, vice president of Microsoft’s Azure AI platform, deciding where to start isn’t complicated.
Organizations simply need to look for the biggest inefficiencies and apply generative AI to make those business processes more efficient.
When we think about what we can do with AI, we have to start with the company’s mission, core business, and areas of greatest inefficiency.
Ali DallolVice President of Azure AI Platform at Microsoft
“It’s an enabler technology,” said Dallol. “When you think about what AI can do, you have to start with your company’s mission, core business, and areas of greatest inefficiency. I can’t sleep in the middle of the night, and that’s when I start working. ”
Business Benefits of Generative AI
Improving efficiency is one of the big promises of generative AI.
Organizations are driven by data. It informs decisions that lead to action. However, many organizations leave much of their data untouched, either because they have more data than they can manage, or because the process is too manual and time consuming, and the data is not available in real time. You are not using your data properly.
As a result, decisions are often made based only on partial data, or on data that may be out of date, leading to results that do not maximize the potential of the organization.
Generative AI has the potential to eradicate many of the problems organizations face in trying to extract the most value from their data.
Automate many of the data management processes that previously required manual monitoring and manipulation. It also virtually eliminates the need for users to know code, making data exploration and analysis possible for any business and his users, not just data professionals trained in coding languages such as SQL and Python. increase.
Generative AI should therefore be applied to inefficiencies, says Dallour.
Generative AI can be used to improve processes and decisions in customer service, marketing, fraud detection, content creation, and anywhere in your business where data is used to inform processes and decisions.
”[Generative AI] “Models are very good at summarizing, content creation, understanding and generating code, text and images. When you add these and complement them with other AI services…the companies are doing amazing things.” Mr Dallol said.
Jonathan Cohen, Nvidia’s VP of Applied Research, likewise says that efficiency gains from automation are a given as companies start applying generative AI to their businesses.
He pointed out that organizations hold large amounts of data related to their operations. Considering the capabilities of generative AI, you can make all your data meaningful.
Data management and analytics vendors do not yet offer a wide range of tools to enable organizations to build generative AI models and applications. Many features have been introduced, most of which are in preview.
But just as your organization’s data engineers and data scientists can build their own data management and analysis capabilities, they can use public tools to train their own generative AI models or develop their own language models from scratch. or build your own generative AI-powered applications. In the meantime, we wait for features from the vendor.
“Companies collect data. That’s their job,” Cohen said. “The real use case is to take that data and turn it into an automated system that can make decisions or recommend actions based on the patterns in which the business has historically worked.”
In particular, he continued, unstructured data can be used in conjunction with generative AI to improve efficiency and drive automated processes more than ever before.
Unstructured data is data such as text, audio, and video that does not have predefined numbers or tables, so it cannot be categorized and organized as easily as structured data that is predefined with numbers and tables. .
To get a complete view using unstructured data with structured data, you need to give structure to the unstructured data. It used to mean that data engineers had to assign numbers to unstructured data. Most of the data is unstructured and required a lot of manual work.
However, generative AI can be programmed to automatically give structure to unstructured data by assigning numbers to it, suddenly leaving organizations with significantly more data that can be used to train models and inform decision making. Offers.
“Today, we can take all the unstructured data, customize our models, build AI that can access all the information, formulate answers and responses, and look for patterns,” Cohen says. says. “It’s accessible today and very achievable.”
Left to right: A panel comprising Andrew Ng of Landing AI, Ali Dalloul of Microsoft, Jonathan Cohen of Nvidia, and moderator Christian Kleinerman of Snowflake discusses generative AI at the Snowflake Summit in Las Vegas.
Generative AI Governance
Targeting inefficiencies is an obvious place for companies to start applying generative AI to their business, but panelists noted that it needs to be done carefully.
Just as organizations carefully manage which employees can access what data, access controls must be applied to generative AI models and applications to ensure data and AI safety.
In fact, the relationship between data and AI is so intertwined that there is no AI governance without data governance, said Christian Kleinerman, senior vice president of products at Snowflake.
“There is no AI strategy without a data strategy, and data strategy includes governance,” he said. “It has to be adhered to.”
Therefore, as companies begin to apply generative AI to improve their business, it is imperative that they do so with AI governance measures in place that enable users to derive value from AI while protecting the organization. says Dallol.
“You’re working with your client’s data, but you’re working with your own data. That content shouldn’t be exposed to anyone. [is not supposed] “To access that information, you have to partition the data and make sure that role-based access control is properly set up and implemented within the data,” he said. [generative AI] Architecture. “
And that’s true whether organizations customize public LLMs like ChatGPT or Google Bard to their own needs, or develop their own language models, he continued.
“You need a deep understanding of roles, services, use cases, and how to manage data,” said Dallol. “That’s the baseline.”
Eric Avidon is a Senior News Writer for TechTarget Editorial and a journalist with over 25 years of experience. He is responsible for analytics and data management.
