In today's rapidly advancing digital landscape, artificial intelligence (AI) is more than just a buzzword -- it's a transformative force reshaping industries, enhancing our daily lives, and pushing the boundaries of what technology can achieve. At the heart of many AI-driven innovations are concepts like machine learning, Large Language Models (LLMs), and specific applications such as ChatGPT. But how do these elements connect, what roles do they play in financial services, and how are we leveraging them at Moody’s in particular?

In this post, we’ll try to break down the fundamental concepts of AI so you can better understand the skills you need to utilize AI-driven solutions like Moody’s Research Assistant to their maximum potential.

The foundation: machine learning

Machine learning is the backbone of modern AI. It refers to the process by which computers learn from data, identifying patterns and making decisions without being explicitly programmed for specific tasks.

Imagine teaching a child to recognize animals in pictures: instead of telling them the exact characteristics of each animal, you show them dozens of examples until they naturally start to differentiate a cat from a dog. In a similar way, machine learning algorithms analyze vast datasets, learning from examples to make predictions, classify information, or recognize patterns.

This ability to "learn" from data allows machine learning models to improve over time, adapting to new information and refining their outputs. Whether it's powering recommendation systems on streaming platforms or enabling self-driving cars to navigate streets, machine learning is the critical process that enables AI to become smarter and more effective.

Building on machine learning: Large Language Models

One of the most impressive applications of machine learning is in the development of LLMs. These are specialized AI models trained on massive amounts of text data, allowing them to understand and generate human-like language. Designed to predict the next word in a sequence based on the context provided by the preceding words, they are at the heart of many generative AI (GenAI) systems, enabling them to craft new content that can mimic human creativity, making them central to the advancements in AI-driven text generation.

LLMs like GPT (Generative Pre-trained Transformer) are trained on diverse datasets, which include everything from books and articles to websites and social media. This extensive training enables them to grasp the intricacies of language, such as grammar, tone, and even subtle nuances of meaning. The result is an AI that can parse instructions, engage in sophisticated conversations, translate languages, generate creative content, and much more.

Bringing it all together: GenAI

Generative AI is a prime example of what LLMs can achieve, and one of the first and most successful examples of GenAI is ChatGPT. Built on the GPT-4 architecture, ChatGPT leverages the power of a Large Language Model to interact with users in natural language. Whether you're asking questions, seeking advice, or getting help to draft an email, GenAI platforms use their deep understanding of language to generate responses that are not only accurate but also conversationally engaging.

The connection between GenAI  and machine learning is foundational: without the underlying machine learning algorithms, there would be no way for GenAI to learn from the vast datasets it was trained on. Likewise, the effectiveness of the GenAI model as a conversational agent is a direct result of the advanced capabilities of the LLM on which it is based. This synergy between machine learning, LLMs, and specific applications like ChatGPT, Replika, or Moody’s Research Assistant illustrates how different aspects of AI work together to create powerful tools that can be used in a variety of contexts.

Introducing Moody's Research Assistant

The origins of Moody’s Research Assistant, a tool developed in collaboration with Microsoft, can be traced back to the period shortly after the release of ChatGPT, as we came to the realization that AI models, particularly those based on LLMs, were becoming increasingly capable of understanding and generating human-like text. As the AI landscape began to shift, the need for a tool that could leverage these capabilities in a more structured and useful way became clear.

The idea for Moody’s Research Assistant was born from a simple question: can we make it easier for our customers to find the insights they need using GenAI? Moody’s Research Assistant was envisioned as a solution that could combine the advanced text-processing abilities of AI with practical applications, making it easier for users to access and utilize the vast amounts of information generated by these models within the financial landscape.

The development process: challenges and solutions

Developing Moody’s Research Assistant was not without its challenges. One of the main hurdles was ensuring that the tool was both accurate and easy to use.

Let’s recall that an LLM is built to generate text based on patterns it has learned from vast amounts of data. The problem here is that this means they may also have a tendency to produce content even when it’s not entirely accurate or relevant, simply because they are designed to "talk". This is what we call a “hallucination”. Ensuring Moody’s Research Assistant produced reliable results, without hallucinating, required careful and continuous iteration.

This challenge was addressed through the use of a technique called Retrieval Augmented Generation (RAG). RAG is an AI pattern that combines an LLM's ability to understand natural language with access to information stored in large and reliable knowledge sources. It works by first retrieving relevant information from a database and then using a generative model, like GPT-4, to produce coherent and contextually accurate responses based on that information. This approach enhances the quality and relevance of the generated text, making RAG particularly useful for tasks that require both accurate information retrieval and natural language generation, such as question answering and content creation.

The beauty of RAG is that it can be designed to retrieve data from either a closed or open system, depending on how it's implemented. In a closed system, RAG retrieves information from a predefined database that may be updated in real time, but only after information is validated, and where access is closely monitored and controlled. This could include a company's internal documents, specific datasets, or any other controlled repository of information, which is exactly what we did for Moody’s Research Assistant.

Another factor in overcoming these challenges was the collaboration with Microsoft, which brought together the strengths of both organizations: Moody's expertise in data and analytics, and Microsoft’s cutting-edge AI technology. By working closely with Microsoft, the team behind Moody's Research Assistant was able to integrate advanced AI models into the tool, ensuring that it was not only powerful but also reliable and user-friendly.

This partnership was a significant contributor in transforming the initial concept of Moody's Research Assistant into a fully functional tool that could meet the needs of diverse users. It highlights the importance of collaboration in the tech industry, where combining different areas of expertise can lead to innovative solutions that might not be possible in isolation.

The result: a tool for the future

Moody’s Research Assistant is the result of a vision to make AI more practical and accessible. Through thoughtful development, collaboration, and a focus on user experience, Moody's Research Assistant has become a tool that not only showcases the potential of AI but also provides tangible benefits to users. Whether it’s helping with complex queries, providing insights from large datasets, or simply making information more accessible, Moody's Research Assistant represents a significant step forward in the application of AI technology, particularly within finance.

As AI continues to evolve, tools like Moody’s Research Assistant will play an increasingly important role in bridging the gap between advanced technology and everyday use in the workplace – one day soon, AI will likely be as ubiquitous as email or social media, and we will be wondering how we ever got by without it.

Now that we have it, how do we use it?

AI is a powerful tool, but it is ultimately shaped by the people who develop, refine, and use it. By demystifying AI, breaking down its complexities and acknowledging its imperfections, we hope that you can start incorporating powerful new solutions like Moody’s Research Assistant into your workflows – saving you time while giving a more holistic view of risk and trends through the use of a wider spectrum of data and insights that Moody's Research Assistant can call within seconds.

So how can you best use Moody's Research Assistant?  There are now a lot of GenAI chat platforms, each espousing a different flavor of conversation and filling a different need, but for our specific purposes, it might help to anthropomorphize a little: it’s not Google or a search engine; rather it’s a smart new hire with a photographic memory, but who doesn’t know how you like things done. So let it know you want a credit memo. Let it know how many years back you want it to go.  Give it a concrete deliverable and specific parameters.

Ultimately, it's best to dive in with an open mind, but if you need more guidance, it may be beneficial to read our Prompting 101 series.

It’s worth remembering that as AI continues to evolve, it is you, the human element – with your creativity, intuition, and willingness to embrace nuance – that will guide its development in the months and years to come