LLM (Large Language Model)
LLM is a type of AI neural network trained on vast amounts of text that can generate, summarise, translate, and analyse language at scale. The technology underpins widely used tools such as ChatGPT and Claude, and increasingly powers features inside platforms like HubSpot that property operators already use every day.
What is a large language model and how does it work?
A large language model is a neural network trained on a vast amount of text for natural language processing tasks, particularly language generation. The architecture most LLMs use, called the transformer, was introduced by Google researchers in 2017 in a paper published at NeurIPS. It allows the model to process relationships between words simultaneously rather than reading text one word at a time. During training, the model learns statistical patterns across billions of examples; after training, it can complete, summarise, translate, or answer queries by predicting the most plausible next token. Well-known products built on this technology include OpenAI's GPT series; Anthropic's Claude is another widely used example.
What can an LLM actually do in a property operations context?
For PBSA and BTR operators, LLMs surface most visibly through tools already in use. HubSpot Breeze, the AI layer built into HubSpot's CRM, uses language model capabilities to summarise email threads, draft prospect outreach, generate leasing communications, and surface deal insights. Beyond the CRM, LLMs can parse lease documents, extract key dates and obligations, and turn raw reporting data into plain-English summaries. The practical effect is faster first-draft work and fewer manual steps for repetitive text-heavy tasks across lettings, finance, and compliance functions.
What are the limitations operators need to understand?
LLMs predict plausible text; they do not verify facts against a ground-truth database. This means they can produce confident-sounding output that is factually wrong, a failure mode known as hallucination. Research on hallucination detection and mitigation confirms that this cannot be fully eliminated: it is a fundamental characteristic of generative models that operate by predicting plausible sequences rather than accessing ground truth. In property operations, where figures in lease agreements, rent schedules, and compliance notices must be exact, any LLM-generated content should be reviewed before it is sent or acted on. The practical response is to design workflows with human review at critical points and to use LLMs for drafting and summarising rather than for authoritative data retrieval.
How do LLMs connect to other systems via APIs and MCP?
On their own, LLMs only know what was in their training data. They extend their usefulness by connecting to live systems through APIs (application programming interfaces) or, more recently, through the Model Context Protocol (MCP), an open standard introduced by Anthropic in November 2024. MCP gives an LLM a standardised way to call external tools and data sources, so instead of custom integration code for every combination of model and system, one protocol handles the connections. This is the technical layer that allows HubSpot's Breeze agents to pull live CRM data, or that allows a custom automation to query your property management system and generate a summary report without a bespoke build for each use case.
Key takeaways
- An LLM is a neural network trained on text that generates, summarises, translates, and analyses language; the transformer architecture, introduced in 2017, is the standard foundation.
- For PBSA and BTR operators, LLMs are already present in HubSpot Breeze: email summarisation, outreach drafts, deal insights, and leasing communications are live features.
- LLMs predict plausible text, not verified fact. Hallucinations cannot be fully eliminated, so human review at critical points is a design requirement, not an optional extra.
- LLMs connect to live systems through APIs or MCP, an open standard from Anthropic (November 2024); without integration, the model only knows its training data.
- Clean, correctly structured data is what makes LLM features reliable in practice: the AI layer is only as good as the data it runs on.
How Cloudfox Helps With LLM
Cloudfox implements and configures HubSpot for PBSA and BTR operators, which means clients get the LLM-powered capabilities in HubSpot Breeze as part of a properly structured CRM rather than bolted on afterwards. That includes AI-assisted email drafting, automated deal summaries, and prospect research, all working against clean, correctly mapped data rather than a messy portal. On the finance side, Cloudfox configures the accounting stack (Xero, ApprovalMax, Syft) that feeds the reporting layer, so when operators want to use AI tools to summarise financial data or flag anomalies, the underlying numbers are reliable. Getting the data foundations right is what makes LLM features useful rather than risky. Find out more at /what-we-do.
Frequently Asked Questions About LLM
Is an LLM the same as ChatGPT?
No. An LLM is the underlying model technology; ChatGPT is a product built on top of OpenAI's GPT models. Similarly, HubSpot Breeze and Anthropic's Claude are products built on large language models. The term LLM refers to the category of AI, not any single tool.
Can an LLM read my property management system or accounting software?
Not directly, unless an integration is configured. LLMs connect to external systems through APIs or a newer open standard called MCP (Model Context Protocol), introduced by Anthropic in November 2024. Without that integration, the model only knows what was in its training data. Properly implemented, an LLM can query live data from a PMS, CRM, or accounting platform and generate summaries or reports on demand.
How accurate are LLM outputs for tasks like lease review or financial reporting?
LLMs are useful for first-draft work and summarisation, but they predict plausible text rather than retrieving verified facts. Hallucinations, where the model produces confident but wrong content, cannot be fully eliminated. For anything where accuracy is contractually or financially critical, treat LLM output as a draft that requires human review before use.
What is the difference between an LLM and an AI agent?
An LLM generates text in response to a prompt. An AI agent is a system that uses an LLM as its reasoning core but also takes actions, executes multi-step workflows, and interacts with other software without constant human prompting. HubSpot's Breeze Agents are examples of agentic AI built on top of language model capabilities.
Do I need to understand LLMs to use HubSpot Breeze or similar tools?
No. Operators use the features, not the underlying model. Understanding what an LLM is matters mainly so you can set realistic expectations: it will draft communications well, but it will not reliably retrieve specific figures unless it is connected to a live data source, and it should not replace human judgement on legally or financially significant decisions.