Of Bricks and Bytes: How data strategy is changing real estate | PROPMODO BOOKMARKS→

ChatGPT’s Awe-Inspiring Real Estate Uses Have Plenty of Limitations, Too

Unless you’ve been living off the grid for the past few months, you know the enormous hype about ChatGPT. The chatbot launched by San Francisco-based OpenAI in November garnered more than one million users in less than a week and has people excited and even a bit scared. It has been deemed the best AI chatbot ever released to the general public that, according to one writer, is part of the generative AI eruption that “may change our minds about how we work, how we think, and what human creativity really is.” 

Google may be the company most concerned about ChatGPT. The chatbot is seen as a real threat to its search engine business, prompting Google to issue a “Code Red” within the company. OpenAI, the artificial intelligence startup with strong backing from Microsoft, is reportedly valued at $29 billion after the chatbot’s release. ChatGPT has plenty of limitations, which we’ll touch on later, but its ability to provide detailed responses in real time across several knowledge domains is impressive. The chatbot is already making waves in the real estate industry, according to many of the professionals I spoke with.

ChatGPT (Generative Pre-trained Transformer) is built on top of OpenAI’s GPT-3 family of large language models. It has been fine-tuned with supervised and reinforcement learning techniques. Large language models are a type of machine learning trained to comprehend the meaning of words from enormous amounts of text data pulled from the internet. Language models are nothing new and something we interact with all the time. They’re used in search engines, to automate translations, and in the auto-complete features on smartphones.

So, ChatGPT isn’t exactly a new technology. Many companies, including in real estate, have used GPT-3 or prior versions of it for several years. The reason why the chatbot has generated such buzz is because of the ease of access. GPT-3 was released in 2020, but it was more complex to use for the layman. ChatGPT has a more user-friendly interface, the big reason so many are experimenting with it outside of strict professional uses. It’s free (for now), so why not? Controversies have erupted over students using it to cheat, but it also has plenty of legitimate business uses already, from writing social media copy to generating business strategy ideas to writing basic HTML code.

A thought-provoking tech tool

Structurely is a company that has been using a version of GPT-2 for about six years. The company uses conversational artificial intelligence in real estate, among other industries, to convert sales leads into qualified appointments. Structurely’s best users are salespeople who use the product to sort hot leads from cold ones. The company employs AI trainers that fine-tune their conversational model. Despite the fear that AI will replace workers, Structurely’s Co-Founder, Nate Joens, said the effect will be more like a rearrangement of jobs and industries. AI training could become a more prevalent industry, and so could something called prompt engineering. “Sometimes, it takes weird, smart prompt engineering to get the most out of AI like ChatGPT,” Joens told me.

ChatGPT is also surprisingly good at programming and writing code. Christian Lessard, Co-Founder of Steelbasis, said he’s used it to take complicated Excel models and create quick Python versions of them. “I think ChatGPT will help a whole generation of commercial real estate people learn to think like a programmer,” said Lessard, whose PropTech company provides vendor management software. He also noted that people are already creating companies with GPT models, so there will be a bit of an arms race in that regard. But there’s a significant barrier to entry to create things using GPT. It’s costly to scale the large language model, and it needs to be reinforced locally with domain real estate knowledge.

Perhaps the best use of ChatGPT in real estate has been generating copy for marketing pitches, emails, and other content generation. 30 Lines is a multifamily communications strategy firm that has used it for this purpose. They took one of their lengthy software support docs, fed the content into ChatGPT, and had it rewritten into a friendly email template for their support team. The process took less than 15 minutes. Previously, 30 Lines used GPT-3 to help with blog posts, video scripts, and advertising headlines. It can also help write real estate listings, though in one humorous example, it completely fabricated details of the ceiling heights and hardwood floors.

Those who I talked to told me ChatGPT works best as a tool for thought, helping generate, clarify, and edit ideas. The AI gives teams “super-powers” to work faster and test ideas quickly, but it still needs the personal touch of a real estate professional. DoorLoop, which provides property management software, has dabbled with the chatbot but hasn’t used it for content generation (yet). The firm isn’t sold on the extensive use of chatbots. DoorLoop’s Co-Founder, David Bitton, told me communication and customer service are areas where he thinks AI can’t fully replace human interaction. “Chatbots are great for streamlining conversations and helping us pinpoint the help we need, but conflict resolution and meaningful customer support are best controlled by humans,” Bitton said.

Beyond the hype

ChatGPT may be “scary good” to some, but there are plenty of limitations. The chatbot is a beta version now, and while it’s trained on a vast data set, it doesn’t yet have access to the broader internet. The data it was trained on only goes up to 2021, so don’t expect legitimate answers about information after that. Many people have also noted the answers are not always correct but come back in a confident tone. This has been called the “confident bullshitter” aspect of ChatGPT. This has led many to compare ChatGPT to a team of high-powered interns: it gets things done quickly and efficiently, and it’s pretty damn good, but you have to verify it most of the time.

The flaws have been well-documented, so the best way to use ChatGPT at this point is to experiment with it. Outside of real estate, many are using it for frivolous ways, like playing with a new toy. Sam Altman, OpenAI’s CEO, agrees with this and tweeted recently that “it’s a mistake to rely on it for anything important right now.” It’s also important to remember we’re still in the early stages. Some real estate professionals have made bold statements about ChatGPT, saying things like, “If you’re a company not leveraging GPT, you’re going to be doomed.” This may be true. It’s smart to learn how to use it now, but it’s too early to tell if it will transform the real estate industry the way some people are expecting it to.

Artificial general intelligence like ChatGPT is great at casual conversations, but it’s not an expert in niche areas. Rumors about a new version from OpenAI (GPT-4) have already spread, which could be built on an enormous data model, generating more accurate responses with more nuance. Unlike this version we’re playing with now, GPT-4 could perhaps take our awe (and fears) even further. This spiffy new version could be unveiled as early as this year.

When we look beyond the hype, ChatGPT is simply another tool for real estate teams to utilize, whether for writing marketing copy or helping with programming. There are many things that living, breathing, human real estate professionals can do that GPT can’t, and this probably won’t change, no matter how good GPT-4 is and whatever versions come after that. ChatGPT has offered us all a glimpse at the stunning progress of artificial intelligence, but the dawning of the “machine age” in real estate is still far off if it ever comes to pass.

Image - Design