Solving Data Wrangling Problems: The Opportunities and Threats of Decmocratizing AI and LLM’s

A lot has been written about artificial intelligence (AI) and machine learning (ML) over the past few years. But for all of the promises and predictions about what’s coming next, so far, AI has remained in the early-adopter stage.

The technology remains something only a select few people know, understand and are able to use and benefit from.

But that may be about to change as AI becomes both more mainstream and more commercialized, allowing the democratization of the technology to extend to everyone. In fact, according to a recent Gartner report, manual data tasks could be reduced by as much as 45% by 2022 thanks to machine learning.

The reason for the rapid adoption is simple. Your company’s data has gotten too large, complex, diversified and too decentralized for mere humans to manage alone. Not only do they need to embrace scalable cloud solutions to manage it all, but soon they’ll also need to embrace AI and ML to automate processing and analysis.

As datasets grow in size, the speed of analysis only grows in importance, meaning AI will soon be a vital component to getting the job done in a timely manner.

Another factor in the democratization of AI and ML will be the transition from open source to commercial applications. Up until now, open source solutions have dominated that AI and ML landscape, but it’s expected that 75% of new end-user solutions are likely to be built with commercial solutions rather than open source alternatives by 2022.

Arman Eshraghi, CEO and Founder of Qrvey, hosts a podcast, “SaaS Scaled.” Our latest episode featured Michael Katz, CEO and Co-founder of mParticle an AI-powered customer data platform. You can watch or listen to the podcast with Michael here, as well as our episode featuring Devvret Rishi, CEO & Co-Founder at Predibase, the low-code AI platform for developers here. We’ve covered some talking points of their discussion below on data wrangling and democrazting access to AI and LLM’s.

Have you seen movement towards democratization of LLM and AI?

“Yes, the first thing I want to say is I think this movement is critically important. I think technologies become interesting when they are put in the hands of the average software developer. One of the most fundamental shifts that happened in the last two decades was the shift to the cloud. And when that actually became useful was when your average software engineer could spin up something in AWS. We lived in a whole new world, and it led to this proliferation of SaaS applications overall. “So, the high-level context I want to say for your question is I think it’s a very important trend. But to answer directly, ‘Are we seeing it in AI?’ I would say it’s a two-part answer. 

1. We see a lot of interest.
2. I don’t think we’ve yet seen that interest materialized into successful production applications.

“According to a survey conducted by Predibase, 85% of organizations are either actively using LLMs or have immediate plans to do so, but only 13% of those organizations have an LLM application running in production.  “Where we are as an industry today is a lot of people have seen that promise and that desire for a more democratized interface to machine learning, and they’re ready to invest the dollars in it. But the tooling in the platforms or whatever else it is that they need to fill that last mile have not caught up yet, because the vast majority of them have not put in an application to production just yet.”

Predictions for the future 

“I think that the future is not going to look like a single large model, like ChatGPT, dominating the SaaS application landscape. I think it’s going to look like many, many task-specific, fine-tuned models that are each good at doing their own individual thing. One of my favorite customer quotes is, ‘Generalized intelligence is great, but I don’t need my point of sale system to recite French poetry.’ “And so, I don’t think we’re going to have these individual single models that dominate as much as we will have individual SaaS applications building and fine-tuning small LLMs that are task-specific.”

Will LLM apps replace or enhance SaaS apps?

“I think that this is true every time there is a large secular shift in technology, there’s going to be three types of organizations. 

1. Incumbents that integrate that new technology and build a better workflow
2. New players that basically replace an existing incumbent because their technology is now their competitive advantage, and they’re able to operationalize it faster or they’re AI-native.
3. A new class of products and technologies that are just unlike anything we’ve seen before because their underlying capabilities were not possible

“I think the future is going to break down into these three herds. And it’s important to note that folks that do not integrate these technologies over a 5-year time period will probably no longer be competitive.”

Democratizing Access to Data Tools & AI

Michael continued, “The tricky thing about AI is that it benefits the entrenched. The larger enterprises with existing data infrastructure and data engineers have an inherent advantage. But much of what we’re focused on today is democratizing access to the same sets of tools and resources used by the world’s largest companies, like Facebook, Google, and Netflix.”

Michael also noted that even when you think you’ve solved challenges with data, that state doesn’t persist. There are always new challenges and changes can lead to a cascade of consequences.

“Teams are drowning in data. What they’re not drowning in is talent.

“We see a huge opportunity in helping bridge that divide between the haves and have-nots. Through automation, we can up-level existing resources for companies and help make them more strategic so they’re not constantly in the muck.”

Opportunities and Threats of AI

Michael continued, “We should be able to automate things and make them much more objective. Imagine if you could tap into everything that you experienced in your life that’s stored in your subconscious, having that accessible in real time in an objective capacity.

“It should provide an opportunity for smart, creative, thoughtful people to get out of the low value-add activity and be more strategic. While it’s going to benefit a lot of people, I also think it’s a threat to some people if their work could be automated and they’re not committed to continuing to grow and learn.”

What Generative AI Can Never Do

“Sometimes human intuition does win, especially when we’re talking creative output. Generative AI, like ChatGPT, could never have come up with Nike’s ‘Just Do It’ campaign. You still need human intuition to create statements that stand out from the average of everything that gets dumped into the data blender.”

Michael and Arman concluded by discussing companies that will break out. Winners will build a strong foundation of first-party data and use it to get close and stay close to their customers and provide them with better experiences. Specifically, they will:

• Create feedback loops
• Improve their processes
• Systematize their business with high quality information
• Automate what gets repeated

They will win among the winners.

As one proof point, McDonald’s is at an all-time high in terms of enterprise value. Much of their expansion is driven by automation and intelligence, of which mParticle has played a part.

You can listen to more SaaS Scaled podcasts here.

Qrvey’s Take on AI Solutions

Open source solutions often provide a lot of initial innovation, but they fall short in many key areas. That’s why there are many players already hard at work building connectors into existing open source AI systems in order to provide the features, security and scalability that enterprises need in order to deploy AI systems more broadly throughout their organization.

Once these key features have been addressed, enterprises will finally be able to fully realize the benefits of AI and ML systems and deploy them reliably at scale.

As with most new technologies, commercialization also brings with it new tools and user interfaces to make it more approachable and usable for everyday users.

This means choosing and deploying AI and ML models will no longer be just for data scientists and IT professionals, instead, everyone in your company will have the self-service tools they want to get the insights they need from that growing mountain of data your company generates every day.

Qrvey is a modern embedded analytics platform that provides self-service integrated with AL/ML capabilities to your entire organization.