Generative AI is transforming the traditional computing stack and will ultimately form the core of a new era of software development, according to Nvidia CEO Jensen Huang.
Speaking with BlackRock CEO Larry Fink at the World Economic Forum (WEF) in Davos, Huang compared the impact of AI on modern computing to the emergence of the web, the personal computing boom, and cloud computing.
“It’s helpful to reason back to first principles about what’s fundamentally going on with the computing stack. This is a platform change,” he said.
“New applications are developed to run on new types of computers. [With the] Platforms migrated to the Internet, and new types of computing platforms hosted all kinds of new applications. ”
With each of these transformations, Huang said, “the computing stack was reinvented and new applications were created.”
Generative AI essentially represents the latest step in this evolutionary path, and Huang pointed out that ChatGPT and Anthropic’s Claude chatbot will be the tip of the spear that will deliver new applications and tools in the future.
“New applications will be built on top of ChatGPT,” he added. “For example, new applications will be built on top of Anthropic Claude, so this is a platform shift in that sense.”
AI transforms software
Later in his conversation with Fink, Huang said that looking at AI through this prism is “really easy to understand.”
As companies flock to technology to transform processes and streamline operations, AI is helping redefine not only what modern software is made of, but also how it is built.
Analysis by Bain & Company shows that AI is already transforming areas such as software-as-a-service, and the arrival of agent AI will further drive efficiency gains and speed of development.
The key difference in this case is that previous software was “effectively pre-recorded,” whereas AI tools rely on prompts, which significantly changes the way users interact. They are more intuitive and proactive and can identify context with minimal information.
From a user efficiency perspective, this is a notable improvement in a world where unwieldy software often relied on specific structured data to function.
“Software in the past was effectively pre-recorded, with humans inputting and writing the algorithms and recipes for the computer to run,” he explained. “I was able to process structured information, which meant I had to enter my name, address, account number, age, and where I lived.”
“You create these structured tables and the software pulls information from them,” Huang added. “We call this SQL query. SQL is the single most important database engine known in the world. In the past, almost everything was run on SQL.”
AI is a much more convenient mode of human-computer interaction, Huang said, because AI can understand unstructured information and can digest text, images, or audio.
“For the first time, computers are available that are not pre-recorded but processed in real time,” he commented. “Describe it however you like, and it will perform the task for you as long as it understands your intentions.”
“5 layer cake”
Huang pointed out that the foundation of AI is important. This is a delicate balancing act where multiple “layers” must work together for the technology to work efficiently.
This aspect adds a degree of complexity that was typically waived in previous technological changes. Simply put, this is not a simple click-and-go technology that operates independently. The infrastructure stack that supports AI is essential.
“When you think about AI, you think about AI models, but it’s really important to understand industrially. AI is essentially a five-layer cake, and at the bottom is energy,” he said.
“AI processes in real time and generates intelligence in real time, so it requires energy to do that. Energy is the first layer. The second layer is the layer I live in. It’s chips, chips, computing infrastructure.”
“The next layer on top of that is cloud infrastructure, cloud services. The next layer on top of that is AI models.”
According to Huang, there is a common misconception that the core of technology occurs in the model domain. However, while this is important, it is at the application layer. On top of that Where real value and economic growth is created.
However, Huang acknowledged there are challenges here. Improvements in model performance are outpacing application development, meaning organizations are struggling to extract value from the technology.
“The reason last year was such a great year for AI is because models were much more advanced than the layers above them, which ultimately are the layers we all need to succeed at.”
“This application layer could be in financial services,” Huang added. “It could be in the medical field, it could be in manufacturing. This is where the economic benefits will ultimately come later.”
Follow us on social media
Please be sure to follow Google News ITPro Keep an eye on all the latest news, analysis and reviews.
Also, Follow ITPro on LinkedIn, ×, facebookand blue sky.
