Over the last three months I’ve had the good fortune to speak at two data conferences, moderate a panel in Chicago and attend a Financial Services Data conference in Boston. While most data conferences tend to focus on the traditional intersection of People, Processes and Technology, what was top-of-mind at all of these events was (you guessed it) Chat GPT! Virtually every conversation was sprinkled with questions and theories regarding the rapid emergence of Generative AI, the implications of Large Language Models (LLMs) and how organizations can adopt them.
Although the building blocks for LLMs date back at least seven years and are based on early neural network innovations, OpenAI’s novel integration of a powerful LLM with a simple chat interface has created a tidal wave of media coverage, experimentation and hype.
At their core all LLMs function in a similar fashion – by ingesting large amounts of text and applying probabilities to the words (or tokens) they can predict the next word in a sequence and therefore provide seemingly intelligent responses to inquiries. What makes LLMs impressive is their ability to generate human-like text in almost any language, something that hasn’t been possible until now.
While there are a number of LLMs now available (and my colleague Gery has written an excellent article about them), Chat GPT is currently the champion. This is likely due to OpenAI’s first-mover advantage and the enormous volume of text (175 billion) that has been used to train it.
Seemingly everyone is experimenting with these models – college students, entrepreneurs, journalists, lawyers, and even your neighbor seems to be doing something with them! If you work for a large organization, chances are you’ve either tried Chat GPT or your team is trying to figure out how to use it.
TOP FIVE THINGS TO CONSIDER
The key to embracing this new technology is balancing the justifiable excitement with pragmatism: we are VERY early in the adoption of Artificial Intelligence and LLMs are a very recent evolution. Below are some considerations if you jump in.
Data Privacy and Security
This is probably the most vital challenge for most organizations. Companies need to ensure that the data collected and processed by LLMs (and Chat GPT in particular) is secure and protected from unauthorized access. Additionally, companies must adhere to existing data privacy regulations such as GDPR (Europe) and CCPA (California) to ensure the necessary security, data residency, accuracy and purpose limitations are followed. If you are in a heavily regulated industry, there are also model transparency and governance considerations to navigate.
Integration with Existing Systems
Before implementing an LLM, companies need to evaluate how this type of AI model will integrate with their existing systems. This includes evaluating the compatibility of the technology with the company's current infrastructure, processes, and workflows. Some LLMs (such as Chat GPT and Google Bard) are cloud-native, creating additional integration challenges for organizations. Other LLMs (such as Alpaca) can be installed within enterprise data centers but typically require specialized and expensive hardware to operate efficiently and at the scale necessary.
Quality of Data Accuracy
LLMs vary dramatically in the quality of their output – this is partially due to the quality of data used to train them as well as the volume of both training data and queries (which continually optimize the model itself). Before enabling this technology for ANY customer-facing function, companies need to ensure that the data used to train the model is accurate, unbiased, and representative of the type of queries and requests it will receive. This improves the probability that the responses generated by an LLM are relevant and helpful to customers.
Given the rapid evolution of LLMs, there are a lack of industry frameworks to address the ethical considerations of these models. As a result, companies need to consider issues such as bias and discrimination, which are already emerging with various use cases. In addition, organizations should keep an eye on industry and government efforts to develop ethical frameworks to guide usage.
As we see with the explosive adoption of Chat GPT, the user experience is a critical factor to consider when adopting an LLM. The integration of this technology should be intuitive, user-friendly and a value-add to ensure a positive experience for both internal and external customers. This requires thinking through how this type of technology fits into a customer’s experience, what it can enable, and what is the most elegant way to introduce it. In addition, there are performance and scaling considerations given the heavy computational nature of LLMs.
In conclusion, adopting generative AI technologies such as Chat GPT can provide significant benefits for organizations interested in enhancing their internal functions, improving their customer communications, and empowering their customer service. However, the companies most successful in harnessing this power will be those that also skillfully navigate the considerations above.