What the Metal CEO Wishes Every Company Knew About Data and Generative AI Models

Throughout history, the world’s greatest philosophers have argued whether nature tends toward disorder.

It certainly abhors a vacuum.

And while the answer to existence’s degree of entropy may still be up in the air, at least one thing is certain: Internal company data, for its part, is frequently messy and often disordered.

This loose bit of IT housekeeping might have been no more than a nuisance or a minor frustration even just a year ago, but now, with the rapid commercialization of artificial intelligence (AI) across today’s operating landscape, it’s becoming more important than ever for enterprises to get their legacy data right to remain competitive.

“If you think about a typical organization, one that’s been in business for 10-plus years, and consider all the data that they have amassed, particularly unstructured data — things like Word docs, slide decks, knowledge-based articles, customer feedback, emails and more — until recently, this data was generally produced and then shelved,” Taylor Lowe, CEO and Co-founder of large language model (LLM) developer platform Metal, told PYMNTS.

He explained that all this information, thanks to AI, can be incredibly useful once unlocked. It just needs to be structured to remove unnecessary complexity.

That’s why data readiness is not only crucial, but increasingly represents foundational table stakes for firms looking to win in today’s AI-driven business environment, he said.

Read also: Growing Enterprise AI Adoption Shows Integration Friction Is No Fiction

Taking a Turnkey Approach to Training AI

“As exciting as AI technology is, it’s still new for most, and expertise is hard to come by,” Lowe said.

At a high level, there are three primary pieces to building AI: data pre-processing, AI training and AI inference, which is when the model goes live.

Data pre-processing comes first and requires organizing and structuring the previously loosely gathered information meant to be used for training the AI, generally by labeling and cleaning the data. It’s a laborious process that is full of friction and frequently outsourced by even the largest leading tech companies.

Lowe explained that his company’s mission is to help firms of all sizes looking to build an AI application win at the outset by removing these frictions and transforming their own, frequently disorganized company data into something that their internal engineers, who may be new to the AI stack, can work with seamlessly and easily.

“The data is transformed, made searchable, and can be plugged into things like chatbots or similar types of AI applications for semantic search and retrieval,” Lowe said. “Often, the information isn’t necessarily new. It’s just much, much easier to access and retrieve, and people can just move much faster as a result.”

This dramatically improves how information is stored and even thought of within a company, effectively giving new life to years of data, he added.

See also: Why B2B Tech Will Drive the Next Innovation Cycle

Unlocking and Repurposing Data

In this period of transformation in the AI field, the conversation revolves around the immense possibilities and the responsibility that comes with it.

“Engineers need to be involved no matter what,” Lowe said. “And that’s because they’re going to be able to tweak the right knobs internally, making sure the application is performing and behaving as expected.”

As more companies embrace AI technology across a variety of use cases, the need for data privacy, employee education, and refining the AI models speeding up processes becomes paramount.

“A lot of people are using OpenAI’s [application programming interfaces (APIs)] to test and sort of play with this technology,” Lowe said. “This won’t last forever in our view, and enterprises are right to push for data privacy, up to and including running the whole stack on their own network. After all, you need guardrails if you are going to ask an organization to entrust its most valuable asset, its own data, to these types of applications.”

He added that enterprise organizations “have the most to gain” from AI in the near term, particularly given the volume of data that they have and continue to produce, and the ability of AI to impact and streamline their current workflows.

But in order to take advantage of AI’s potential to transform and reshape historic processes, the right controls around data privacy need to be put in place, as does employee education around some of AI’s own inherent pitfalls, such as certain models’ tendency to produce hallucinations.

Still, Lowe emphasized that AI technology is already starting to reshape the landscape of enterprise programs — and there is no stopping now.

“This technology simply didn’t exist before, and companies should really understand the power that they have right now because they can help shape this technology and unlock the most promising use cases,” he said.