How Customization of Models is Bringing General AI to Business
View all sessions on demand from Smart Security Summit This.
There’s been a lot of hype and activity around generalized AI models, especially in the short time since ChatGPT debut for the first time.
ChatGPT — and GPT-3 large language model (LLM) it’s based on — trained in public data, serves as a great foundation for consumer apps, but lacks the customization, privacy, and security a business loves bridge. That’s where Rodrigo Liang, co-founder and CEO of Samba Nova Systemis looking to make a difference with the launch of his company’s SambaNova Software Suite today, aimed at helping businesses build and deploy custom artificial intelligence model.
>> VentureBeat’s constant tracking artificial intelligence insurance <
SambaNova started operations in 2017, mainly focusing on hardware for AI, amazing increase $676 million in April 2021 to support its efforts. In recent years, the company has expanded beyond its initial focus on hardware to build support for both machine learning and inference training on various models, with data flow as a service offer for sale. The new SambaNova Suite expands on the data flow service with a set of capabilities that allow organizations to customize both proprietary and open source AI models to meet their specific requirements.
“The focus of SambaNova is on how to bring more general AI capabilities to the enterprise,” Liang told VentureBeat. “There are some things you need for businesses to succeed and we are doing it today for them.”
Nvidia isn’t the only AI hardware vendor for synthetic AI
More and more vendors are building on generic AI models to help enable enterprise use cases.
Content creation is a particularly vibrant area of creative AI customization for business. Jasper artificial intelligencefor example, recently announced Jasper for business offers designed to help customize AI for a particular business. font emerged from the shadows on Monday with an enterprise content creation platform for creative AI, along with a $65 million endowment. quantitative last week announced its foray into generalized AI, helping organizations execute their business strategy.
SambaNova’s key differentiator from others in the general AI space is that it has its own hardware to help optimize enterprise use cases. Instead of relying solely on Nvidia GPUs like many in the industry, SambaNova has developed its own custom silicon optimized for both training and machine learning inference.
“What we did was… use an AI software stack that people really wanted to use PyTorch, TensorFlow and complex models like GPT, and we take them all the way down to silicon,” says Liang. “We have really custom-built silicon to run these large models for the enterprise, compared to the other way around.”
The team behind SambaNova, including Liang, has experience building microprocessors for Sun Microsystems and Oracle. Liang says that SambaNova has created an extremely efficient and energy-efficient AI processor to run these innovative enterprise AI applications.
Liang emphasized that custom silicon also allows for continuous training and inference capabilities, so that the data that feeds general AI models can be updated.
“In business, you need real-time information, and so you don’t want your models to really fall behind,” he said.
Privacy, Responsible AI, and Enterprise
With the new SambaNova Suite, Liang says his company is looking to address the key challenges businesses face with generalized AI. Among those challenges are the customization of company-specific data, as well as the ability to limit bias and deliver accountable and explainable AI.
With its platform, SambaNova allows an organization to conduct customized training in a private environment on any data the organization has, including unstructured data that can be found in discussion channels. The company’s Slack discussion. Taking it a step further, Liang said the platform provides transparency to organizations about how a given AI model actually works.
“SambaNova has been built on its ability to give you precise information about how the model came to a certain conclusion,” says Liang. “We store all the processes around how we train and refine the model, so that when an auditor comes in or someone wants to check for bias or why something is happening in a certain way, you can really work through the process and verify that your results were done properly.”
VentureBeat’s Mission is to become a digital city square for technical decision-makers to gain knowledge of transformative and transactional enterprise technology. Explore our Briefings.