Sponsored content
ChatGPT and similar tools based on large language models (LLM) are awesome. But they are not all-purpose tools.
It's like choosing other tools to build and create. You must choose the right one for the job. You wouldn't try to tighten a bolt with a hammer or flip a hamburger with a whisk. The process would be uncomfortable and would result in dismal failure.
Language models like LLMs make up just one part of the broader machine learning toolset, which encompasses both generative ai and predictive ai. Selecting the right type of machine learning model is crucial to align with your task requirements.
Let's dive into why LLMs are better suited to helping you write copy or generate gift ideas than tackling your business's most critical predictive modeling tasks. There is still a vital role for “traditional” machine learning models that preceded LLMs and have repeatedly proven their value in businesses. We'll also explore a pioneering approach to using these tools together, an exciting development we at Pecan call ai/predictive-modeling-lp/?utm_medium=partner_affiliate&utm_source=kdnuggets&utm_motion=selfserve&utm_campaign=kdnuggets_jan2024&utm_content=llm+alone” rel=”noopener” target=”_blank”>Predictive GenIA.
LLMs are designed for words, not numbers
In machine learning, different mathematical methods are used to analyze what is known as “training data,” an initial data set that represents the problem that an analyst or data scientist hopes to solve.
The importance of training data cannot be underestimated. It contains the patterns and relationships that a machine learning model will “learn” to predict outcomes when it is then given new, unseen data.
So what specifically is an LLM? Large language models, or LLMs, fall under the scope of machine learning. They originate from deep learning and their structure is developed specifically for natural language processing.
You could say that they are built on a foundation of words. Your goal is simply to predict which word will be next in a sequence of words. For example, the iPhone autocorrect feature in iOS 17 now uses an LLM to better predict which word you're likely to try to type next.
Now, imagine you are a machine learning model. (Bear with us, we know that's a stretch.) You have been trained to predict words. He has read and studied millions of words from a wide range of sources on all types of topics. Your mentors (aka developers) have helped you learn the best ways to predict words and create new text that fits a user's request.
But here's a twist. A user now provides you with a massive spreadsheet of transaction and customer data, with millions of rows of numbers, and asks you to predict numbers related to this existing data.
How do you think your predictions would turn out? First, you'll probably be upset that this task doesn't match what you worked so hard to learn. (Fortunately, as far as we know, LLMs don't have feelings yet.) More importantly, you are asked to perform a task that does not match what you have learned to do. And you probably won't do so well.
The gap between training and task helps explain why LLMs are not suitable for predictive tasks involving numerical and tabular data, the primary data format most companies collect. Instead, a machine learning model designed and tuned specifically to handle this type of data is more effective. He has literally been trained for this.
The efficiency and optimization challenges of LLMs
In addition to being better suited to numerical data, traditional machine learning methods are much more efficient and easier to optimize for better performance than LLM.
Let's go back to your experience posing as an LLM. Reading all those words and studying their style and sequence sounds like a lot of work, right? It would take a lot of effort to internalize all that information.
Similarly, the complex training of LLMs can result in models with billions of parameters. That complexity allows these models to understand and respond to the complicated nuances of human language. However, intense training comes with intense computational demands when LLMs generate responses. “Traditional” numerically oriented machine learning algorithms, such as decision trees or neural networks, will likely require far fewer computing resources. And this isn't a case of “bigger is better.” Even if LLMs could handle numerical data, this difference would mean that traditional machine learning methods would still be faster, more efficient, more environmentally sustainable, and more cost-effective.
Also, have you ever asked ChatGPT how it knew to give a particular answer? Your answer will probably be a bit vague:
I generate responses based on a combination of authoritative data, data created by human trainers, and publicly available data. My training also involved large-scale data sets obtained from a variety of sources, including books, websites and other texts, to develop a broad understanding of human language. The training process involves running calculations on thousands of GPUs over weeks or months, but the exact details and timescales are proprietary to OpenAI.
How much of the “knowledge” reflected in that answer came from human trainers versus public data versus books? Even ChatGPT itself isn't sure: “The relative proportions of these sources are unknown, and I don't have detailed visibility into which specific documents were part of my training set.”
It's a little disconcerting that ChatGPT provides such confident answers to your questions but can't trace your answers back to specific sources. The limited interpretability and explainability of LLMs also pose challenges in optimizing them for particular business needs. It may be difficult to understand the basis for their information or predictions. To complicate matters further, certain companies face regulatory demands that mean they must be able to explain the factors that influence a model's predictions. Overall, these challenges show that traditional machine learning models (generally more interpretable and explainable) are probably better suited for enterprise use cases.
The right place for LLMs in companies' predictive toolkit
So should we leave LLMs to their word-related tasks and forget about them for predictive use cases? Now it might seem like they can't help predict customer churn or customer lifetime value after all.
Here's the thing: While saying “traditional machine learning models” makes those techniques sound widely understood and easy to use, we know from our experience at Pecan that companies still struggle to adopt even these more familiar forms of ai.
<img decoding="async" src="https://technicalterrence.com/wp-content/uploads/2024/01/1706638030_738_Why-LLMs-Used-Alone-Can39t-Address-Your-Business39s-Predictive-Needs.png" alt="42% of companies in North America have not started using ai at all or are just beginning to investigate their options.” width=”100%”/>
Recent research from Workday reveals that 42% of companies in North America have not started using ai or are only in the early stages of exploring their options. And it's been more than a decade since machine learning tools became more accessible to businesses. They have had time and there are several tools available.
For some reason, successful ai implementations have been surprisingly rare despite the huge hype around data science and ai, and their recognized potential for significant business impact. There is a lack of any important mechanism to help close the gap between the promises made by ai and the ability to implement it productively.
And that is precisely where we believe LLMs can now play a vital liaison role. LLMs can help business users bridge the gap between identifying a business problem to solve and developing a predictive model.
With LLMs now on the scene, business and data teams that do not have the ability to manually code machine learning models can now better translate their needs into models. They can “use their words,” as parents like to say, to begin the modeling process.
Merge LLM with machine learning techniques designed to excel in business data
That capability has now come to Pecan's Predictive GenAI, which is merging the strengths of LLM with our already highly refined and automated machine learning platform. Our LLM-based predictive chat collects insights from a business user to guide the definition and development of a predictive question – the specific problem the user wants to solve with a model.
Then, using GenAI, our platform generates a Predictive Notebook to further facilitate the next step towards modeling. Again leveraging the capabilities of LLM, the notebook contains preloaded SQL queries to select the training data for the predictive model. Pecan's automated data preparation, feature engineering, modeling, and deployment capabilities can accomplish the rest of the process in record time, faster than any other predictive modeling solution.
In short, Pecan's Predictive GenAI uses the unparalleled language skills of LLMs to make our best-in-class predictive modeling platform much more accessible and friendly to business users. We're excited to see how this approach will help many more companies succeed with ai.
So while the L.L.M. only While they are not adequate to handle all of your predictive needs, they can play a powerful role in advancing your ai projects. By interpreting your use case and giving you an edge with automatically generated SQL code, Pecan's Predictive GenAI is leading the way in bringing these technologies together. Can ai/predictive-modeling-lp/?utm_medium=partner_affiliate&utm_source=kdnuggets&utm_motion=selfserve&utm_campaign=kdnuggets_jan2024&utm_content=llm+alone” rel=”noopener” target=”_blank”>check it out now with a free trial.