Since OpenAI, Google DeepMind and Mistral developed enterprise-grade generative AI models, organizations have been able to tap into the rich capabilities that these foundational models offer. Over time, businesses found that these models were limited because they were trained using vast amounts of public data. Customization is the practice of adapting large-scale language models (LLMs), to better suit a specific business’s needs. This can be done by incorporating a company’s data and expertise, teaching new skills or tasks to a model, or optimizing prompts or data retrieval.
Customization has been around for a while, but early tools were rudimentary and the technology and development teams often didn’t know how to do it. The customization methods and tools are changing and giving businesses more opportunities to create unique value with their AI models.
To learn how these leaders are leveraging these opportunities, we surveyed 300 technology executives in large organizations across different industries. We also spoke with a few of these leaders in depth. They all customize generative AI models, and shared with us the reasons for doing so, as well as the tools and methods they use, the challenges they face, and the steps they are taking to overcome them.
According to our analysis, companies are moving forward with customization in a very ambitious manner. They are aware of the risks, especially those related to data security, and are using advanced methods and tools such as retrieval augmented generation (RAG) to achieve their desired customization gains. Download the full report.This content was created by Insights – the custom content arm at MIT Technology Review. This content was not written by MIT Technology Review editorial staff.
