Over the past several months, large language models (LLMs) have established themselves as the premier methodology for generative tasks in natural language processing, most notably through the news around ChatGPT. Most likely your board, financial analysts, employees, and customers are asking what you are doing with generative AI.

Generative AI is a branch of artificial intelligence that uses algorithms to create or generate new and original content, such as images, text, source code, audio, or video. This technology utilizes machine learning models trained on large datasets to create new content similar in style, format, and content to the input data.

While researchers predict that it could take up to five years to start seeing real business impact from this technology, listed below are the areas of investment that can set your organization up for financial and competitive success in the mid to long term.

The initial development of large language models requires massive amounts of data, and generative AI outputs are built on advanced data inputs, including acquisition, processing, and pipelines.

Data engineering and mastering is not as glamorous as artificial intelligence model building and has historically been underinvested in — many have learned the “garbage in, garbage out” rule the hard way. Your data scientists might be distracted with doing this cleanup work when they should be building models. In sum, data quality drives model performance, fairness, and scalability; poor data quality decreases effectiveness, adoption, and value realization.

Your CIO, CHRO, CFO, chief legal, and business line leaders should contribute to your organization’s rules of the road for generative AI because each will ultimately benefit from it if managed properly. As a first step in that direction, update your employee handbooks, communications, and policies to ensure you’re covering topic areas such as:

• Intellectual property loss
• Security vulnerabilities in AI-generated code
• Confidential information and data leakage
• Regulatory compliance
• Guidelines for responsible use
• Webinars, FAQs, and forums

If you’re using a generative AI chat interface that is not your company’s, then you’re most likely helping to develop someone else’s IP. This arena is evolving rapidly, and currently there are both vendor products and open-source options. Open-source options such as the Pile and GPT-NeoX serve as a starter kit for you to customize for your own strategic purposes while retaining your IP and avoiding operating expense outlays to train the models.

News about bias in artificial intelligence is common and top of mind for many leading AI organizations and governments. There is still more to learn about how bias develops through a model training process as well as how bias amplifies beyond the training dataset. Long story short, your organization should develop an operating model and knowledge base about bias in data, model development, and monitoring.

Seeing the impact of AI in your income statement will take time, but by following the steps above you will simultaneously reduce the chances of a setback and set yourself up for driving revenue growth, efficiency, and new customer experiences.