HP Bets On Startup To Help Developers Create Trustworthy AI Models

Thanks to a technology integration with a startup called Galileo, the PC and print giant says its upcoming AI Studio software will give developers the ability to ‘detect and correct hallucinations, drift and bias in their data’ as they develop custom models with proprietary information.

HP Inc. said it is partnering with a startup to enable the development of trustworthy AI models through its forthcoming Z by HP AI Studio software.

Unveiled Monday, the Palo Alto, Calif.-based company said it will use technology from San Francisco-based Galileo to integrate a trust layer into AI Studio, which it called the “world’s most comprehensive workstation solution for AI development.”

HP plans to release AI Studio in North America and the U.K. early next month and follow that up with Galileo integration in October.

“We’re enabling data scientists to work more effectively, faster and more collaboratively than anywhere out there. That’s why we’re so excited today to add to this tool: the ability to create a trust framework,” said Jim Nottingham (pictured), senior vice president and division president of advanced compute solutions at HP, at an event last Thursday.

The trust layer enabled by Galileo is designed to tackle a major challenge that continues to blemish the quality of large language models (LLMs) and their appeal: so-called hallucinations, which is a term used to describe when a model provides incorrect or misleading information.

Galileo is also taking on two related issues in the world of LLMs: drift, which happens when the accuracy of a model degrades over time, and bias, a well-documented problem that shows up in different forms for a variety of reasons.

Using Galileo’s technology, Nottingham said, AI Studio will give developers the ability to “detect and correct hallucinations, drift and bias in their data.”

He added that the software will let developers “set and fully configure guardrails, which is ultimately going to let them increase the quality of these models.”

“It will come with tools that enable rapid experimentation, real-time monitoring and proactive detection against inaccurate or biased outputs. Ultimately, it’s going to enable these companies to accelerate the deployment of their AI applications,” Nottingham said.

Founded in 2021 by former Google and Apple software engineers, Galileo provides an “end-to-end platform for generative AI evaluation, experimentation, observability and protection,” according to the startup’s website.

The startup has raised $23.1 million in investor funding to date, and it said its customers include banks, telecom firms and retail companies in the top 50 of the Fortune 500 list.

“Allowing businesses to leverage their proprietary data without compromising security is critical for AI development, and Galileo is proud to join forces with HP to deliver solutions that add visibility, control and trust to enterprise generative AI projects,” said Yash Sheth, Galileo co-founder and COO, in a statement.

Nottingham said HP is introducing the feature after “consistently hearing feedback that they have significant concerns about the accuracy of results” provided by AI models.

Feedback also informed the creation of AI Studio in the first place, according to Nottingham, who said the software platform is meant to address customer challenges related to enabling developers to collaborate on model development from different locations using various types of compute infrastructure, whether that’s a workstation PC or the cloud.

So far, early customers have provided positive feedback, saying that it “enables them to centralize their workflow,” the executive said.

“It enables them to streamline data management, and it significantly improves collaboration,” Nottingham added. “What we’re hearing from the customers also is that it provides significant value when it comes to managing and sharing your data, your test steps and your results across all these complex workloads.”

This, in turn, is saving customers a “tremendous amount of money,” he said.

“They’re also especially finding value in the ability to select the compute and get the right compute to the data, to share and manage the data,” said Nottingham.

Revealed by HP last fall, AI Studio “enables users to create shared projects, connect to the data needed and invite collaborators in just two clicks,” the company has previously said.

Users can connect to the platform locally with a workstation PC—made by HP or others—or they can access it via the cloud. It also offers them the ability to deploy models “wherever and whenever they choose,” whether that’s in a PC or in a cloud instance.

Another key feature is the platform’s integration with Nvidia’s NGC cloud platform, which provides AI developers with access to fully managed services such as the GPU designer’s NeMo LLM customization service.