Last year at around this time I gazed into my crystal ball of data (no, it’s not powered by an LLM) and made five predictions about data trends that would capture headlines in 2023.

They were: 

  • Data as a product 
  • Cloud First
  • Expansion of Roles
  • Privacy and AI Trust
  • Metaverse and Web3

For the most part, I think these trends have defined the previous year. Generative AI is a great example of data as a product. Cloud first is increasingly becoming the way organizations operate, especially if they want to connect their data to LLMs and do the fine-tuning of their models themselves. Roles have expanded but also been redefined, again thanks to the prevalence of AI assistants. Privacy and AI trust are everywhere these days, from the US White House Executive Order on safe, secure, and trustworthy AI to California’s report on Gen AI risks and potential use cases. And even though the metaverse and Web3 slipped onto the back burner during 2023, a lot of work continues here, especially on the AI side. So watch this space on that final trend.

Data Trends for 2024

As I did last year, I’ll offer a trend I see as important and then point out its possible implications for data leaders.

The Trend: Hardware and Data Hunger
One of the things we know about LLMs in particular – not to mention large machine learning models – is that boosts in performance come from adding the number of parameters we train in a model rather than tweaking that model’s algorithm. This requires more innovative hardware, both chips and servers. I would expect to see more chip players based on the stratospheric rise of NVIDIA as well as a push for more data (and more obscure data sets that we don’t have access to) to grow and flourish in 2024.  

Data Leader Strategy
Starting with architecture is key, since compute demand for AI is outstripping the ability of some organizations to keep up. Work with your IT leaders to determine whether your stack has the right chips, software, servers, data management software, and other infrastructure. Ensure that your cloud providers do, too. In terms of data, look at the data you have as a company that can define competitive advantage when it is used to fine-tune a base model. Proprietary data in LLMs will define the next level of value in AI in the coming years.

The Trend: The LLM Operating System
One of the reasons for the enormous push for new data and hardware is a platform shift to a new operating system defined by LLMs such as Bard, Claude or GPT. We’ve seen this many times in the past, most recently in the development of the mobile operating system, which generated thousands of apps. Already we’re seeing announcements of app-like models to support niche applications, the prime example being OpenAI’s ChatGPT Store, now slated for release in early 2024. Expect the internet to begin to take on these new model-based characteristics as well. 

Data Leader Strategy
Data leaders should ask how their strategy must evolve as we shift into a world of LLM operating systems. How might they use LLMs to create bespoke applications based on users’ needs? How should they think about optimizing privacy, memory, and performance? And what does a data environment centered on LLMs mean for your organization’s business model? 

The Trend: Thinking Models, Fast and Slow
Just as Nobel Laureate Daniel Kahneman’s best-selling book laid out two ways we think – fast, instinctive, and emotional as well as slower, more deliberative, and more logical – I think we’ll begin to see LLMs characterized by deeper reasoning and slower thinking. Any user of GenAI already has seen the “fast” models, which generate content quickly, automatically, and even unconsciously (hallucinations, unexpected new ideas) but aren’t very good on rational, logical, slower decision-making. I anticipate that in 2024, we’ll start to see the birth of more slower and reasonably thinking models. This trend started to come into focus with a recent paper on training language models with pause tokens, which prompt them to pause and check through some of their answers before presenting them to the user. 

Data Leader Strategy
Since these slower-thinking models do raise the possibility of breaking encryption methods, data leaders should tread carefully with them until more research and development has been done to safeguard them. These models may not see the public eye during 2024 for just this reason. At the same time, slower AI models may provide a vital bridge for us to understand what is happening in highly complex data areas – quantum computing, for example.

The Trend: Tool Consolidation
Anyone in today’s data space sees how many tools are required – tools for data governance, visualization, cataloging, ETL, and so on. The data trend in 2024 will see the energy continue to shift back toward consolidated master tools, not unlike what Microsoft has been demonstrating with their Microsoft Fabric analytics platform. I think we’ll continue to see companies release all-in-one solutions. Many will be used in cloud deployments to consolidate multiple functions, and will offer relief to budget-strapped IT and data departments.

Data Leader Strategy
Enterprise architects and chief data officers will need to grapple with these questions in earnest and look at their current infrastructure. Do they need to have ten different tools that do individual specialized things or do they migrate to one or two tools that span many different areas? Where does best of breed still top end to end? 

The Trend: Workforce Upheaval
Anyone who has used an AI assistant – what Microsoft calls a Copilot – has witnessed firsthand what a difference they can make. In 2024 we’ll see more and more tools that will be LLM infused or offer some type of AI automation. The anxiety is that this role change is happening at a rate faster than we as humans are able to adapt. People are rightly wondering, What does this mean for my job? How do I adapt? Another legitimate worry is that AI will increase the existing digital divide around the world while at the same time calling the necessity of a four-year college degree into question. Will you even need a four-year degree if you have base skills plus an AI assistant? Although many companies still resist this shift, the evidence from using AI tools to complement skills is that B-level players can more easily upskill to A-level performance.  

Data Leader Strategy
The good news from preliminary research is that individuals who are given access to LLMs that they can use alongside their jobs are shown to execute more tasks and perform better at them. Data leaders should seize the opportunity to redefine roles in a way that allays employee anxiety, even as the possibility for social unrest and even upheaval driven by AI continues to rise.  

A call for technology optimism
I continue to be a technology (and AI) optimist, which is why I hope more companies create safe spaces for LLM use and training in 2024, even if it’s not yet part of everyday business. Remember, 40 years ago computers entered our world as workplace “co-pilots.” They haven’t taken over, just made things run faster and more efficiently. That’s why all data leaders should always be looking to the jobs, the goals, the education, the training, and the upskilling that we need to manage what happens in our workplace. Allow your team and colleagues to follow their curiosity about what data can do, and regardless of the data trends we see in 2024 it’s an easy prediction to say you’ll have a brighter future.