Getting to know Large Language Models
The large language models (LLM) have revolutionised artificial intelligence, by giving machines the power to understand human language and generate answers similar to the ones of a human. These models are mainly massive neuronal networks created from a huge quantity of data, such as the entire textual content available on the Internet or transcription of every conversation held by humans.
Language models have become more and more popular. They are using unsupervised learning, meaning they learn to recognize models et make predictions without actual guidance on what they need to look for and include billions of parameters.
GPT-3 (Generative Pre-trained Transformer 3), developed by OpenAI, is probably the most famous one. It is made of 175 billion parameters and has an impressive range of capabilities, such as translation, creating text summaries, or even creative writing. GPT-3 is now used to create articles, stories, or websites. Aside from this one, a lot of new large language models are being developed, such as Google’s T5 or Nvidia’s Megatron. LLM are powerful tools used for natural language processing (NLP) and their use will keep on increasing. The focus is on improving such models so they can hold more complex conversations or go as far as understanding human emotions. And thus, forever changing the way we interact with computers. And the way we work.
How are they applied to real estate?
Real estate information is still mainly paper-based, with lots of documents kept in archives. One single asset can have more than 3000 associated documents. Bills, diagnostics, rent contracts, models, etc. are mixed together, making it hard to take decisions based on factual information. These large amounts of documents are constantly evolving and must be processed, checked, and updated. Data extraction is a tedious and expensive process because of the volume of documents. To manually sort it and search is bound for human error and idle work, with dozen of excel sheets to go through.
With technology and especially the latest improvements in terms of language models such as LLM and NLP, it is now possible to easily handle large volumes of texts and documents. Technology is now able to read and understand your data. And thus, able to search and extract useful information, automatically index your documents, or analyze a large quantity of data at once. As the real estate industry has a written culture (leasing agreements, contracts, on-site inspection reports…) rather than a quantitative one, NLP is crucial.
Not only will it help you gain time, as you will no longer spend hours looking for specific information, but instead a simple query will suffice. But it will also improve the accuracy of your data, by checking the veracity of information coming from diverse sources. The extraction time will be reduced to seconds so you can use accurate insights to make informed decisions. Another perk is that it scans and reads documents, highlighting missing fields or errors, and empowering teams to work with accurate information.
Language models have become powerful tools for asset managers as they can help them to extract valuable information from large amounts of data, make sure of its accuracy, and keep it fresh and updated. All of that, in order to support asset managers make better decisions about their assets.
The future of AI applied to human language looks bright with strong impacts on jobs. But in what proportions? Time will tell.