National Housing Market Outlook

Unlocking the Potential of Generative AI and Large Language Models for Your Business

Wilson Becker

April 19, 2024

We are deviating from our usual insight on the housing market to demystify Large Language Models (LLMs) and Generative AI (Gen AI).

Hopefully, this will help you understand the core concepts behind LLMs, their strengths and limitations, and how you can leverage them to drive revenue growth and cost reduction in your business.

Since ChatGPT went viral in November 2022, an entire industry has sprung up around Gen AI. This technology is nothing new. The discipline is older than most of its practitioners. The real breakthrough is LLMs. 

Spoiler alert—this technology won’t replace humans anytime soon. Still, there are ways that executives can use LLMs to speed up their staff’s text generation and processing tasks. 

Introduction to Generative AI

Gen AI is a subset of deep learning. If you’ve ever heard the term “neural network,” you already know a little about deep learning. Neural networks function similarly to the human brain. The network accepts images or sounds as input and transforms them into language it can understand and respond to meaningfully.

Deep learning processes information in the depths of the network. Its response improves over time through a process called model training.

Gen AI models are one kind of neural network. AI Developers designed them to create new content, such as text, images, or music. In contrast, developers designed discriminative models to classify and categorize input. Say, for example, that you want to build an AI to create new marketing content. You’d use a generative model. If you wanted to build an AI to determine whether your latest marketing content will sell the product, you’d use a discriminative model.

What is a Large Language Model?

LLMs are a type of generative model, not a discriminative model, meaning they are good at creating new content but cannot sort inputs into categories. We call them large language models because they train on vast amounts of text—nearly every book you could imagine, every Wikipedia article, and every blog post on the internet. During the training process, you feed the model all the text data you can find, and then through some clever calculus, the model “learns” how to generate new text.

What we found out with ChatGPT is that, for the most part, when you increase the size of the training data, you increase the quality of the text that comes out the other end. A common refrain among scientists in the deep learning community is that LLMs approximate (or are only as good as) the quality of the training data the user feeds them.

The actual language model itself only consists of two things:

  • A huge file containing the model weights (a bunch of numbers that represent what the network learned during model training)

  • Several hundred lines of code to make the whole thing run

Very large language models, like OpenAI’s GPT-4, were trained on so much text that you can think of them as constituting their own world model. ChatGPT has nearly all the high-quality English-language content worldwide, compressed and served to users in a free chat interface.

What are LLMs good at?

LLMs are good at generating new text. LLMs are inefficient compared to older, less experimental language model architectures (read: expensive to use). But, with the right prompting strategy, it’s possible to elicit high-quality output on a handful of tasks.

Here are some things that LLMs like ChatGPT can do reliably:

  • Text completion (e.g., Finish this sentence: The cat in the ___)

  • General question answering

  • Summarizing large bodies of text

  • Translating text from one language to another

  • Content generation (e.g., articles, stories, promotional materials)

  • Sentiment analysis and opinion mining (e.g., How many of our followers have bad things to say about us on social media?). Note that this is a classification problem, so if you use LLMs for this, you’re overpaying.

What are LLMs bad at?

If you’re a leader in your organization, it’s essential to understand that LLMs are not a silver bullet for every bottleneck and business problem.

There are several things that LLMs are bad at:  

  • They struggle with tasks that require understanding the passage of time or the sequence of events.

  • LLMs cannot effectively plan or reason about long-term consequences.

  • Since they lack genuine understanding of the words they are generating, LLMs may produce internally inconsistent, incoherent, contradictory, or factually incorrect text. This problem is known as hallucination.

  • LLMs are incapable of mathematical reasoning, logical reasoning, and computation. Counting, for example, is a big problem.

  • LLMs frequently over-summarize. If you’re looking for an in-depth analysis of a complex topic, you’ll likely need a team to put in serious effort.

The LLMs on the market today use clever tools and tricks to ensure you never find out about these limitations. Still, it is important to note that these models fundamentally lack basic general intelligence. According to Yann LeCunn, Chief Scientist at Meta, “a house cat has way more common sense and understanding of the world than any LLM.”

That said, generative deep learning has a very active and motivated research community, and much time and money is being spent on solving these problems as quickly as possible.

How can LLMs grow revenue and reduce costs?

The most reliable opportunity for leaders to capitalize on the current Gen AI upswing is to find areas in their businesses that rely on text generation and text processing. It is important to understand that your business’s Gen AI strategy should not involve reducing staff. The technology is not ready—and might never be able—to replace human-generated content. Instead, your focus should be on giving your staff AI-based tools to accelerate output and increase the quality of work.

Here are some ideas:

  • Text generation: If part of your business involves creating product descriptions, writing news articles, or brainstorming topics, giving those teams access to low-cost tools like ChatGPT, Claude, or Gemini might help speed up the writing process.  

  • Text processing: Does your data team perform article summarization, sentiment analysis, or document classification? If so, transformer models like BERT can help improve the velocity and accuracy of these tasks.

Leaders across industries find that commercially available AI solutions struggle to produce business insights at the level of their marketing claims. The technology is very new, after all, and it’s not like the FTC regulates those marketing claims. In some cases, there are opportunities for small teams to build homegrown AI solutions in short order.

I’ll continue to provide more tips on how business leaders can improve their company’s competitiveness on both the buy and the build side of the Gen AI conversation.

Stay connected with me on LinkedIn.

Building Market Intelligence™

Every week, we deliver analysis to over 40,000 subscribers with our Building Market Intelligence™ newsletter. Subscribe to our weekly BMI newsletters to stay current on pressing topics in the housing industry.

About The Author

Wilson Becker
Vice President of Engineering
Wilson manages Data, Software, and AI at JBREC. His team experiments with new ways of capturing and analyzing industry data using Machine Learning and AI.

Contact Us

If you have any questions about our services or if you would like to speak to one of our experts about we can help your business, please contact Client Relations at clientservices@jbrec.com.

Products and Services Mentioned

green check icon

US Demographics Insights and Strategies

Access current analysis and specific recommendations to navigate US demographic shifts and capture long-term opportunities.
green check icon

Research Memberships

Our research services enable our clients to gauge housing market conditions and better align their business and strategic investments in the housing industry. We provide a thoughtful and unique holistic approach of both quantitative and qualitative analysis to help clients make informed housing investment decisions.
green check icon

Real Estate Consulting

Our experienced team of consultants helps clients make sound housing investment decisions. We thrive on their success and work with many clients over multiple years and numerous projects.

Latest Insights

Charting a 22-year roller coaster of investor activity
From Boomers to Zoomers: Housing Priorities by Generation
Dear Consultant: How do we address the shortage of skilled trades workers?