It may not be long before your company has its own ChatGPT-like artificial intelligence (AI). An April 2023 survey conducted by AI software provider expert.ai found that 37.1 percent of enterprises are planning to build customized large language models (LLMs).
The benefits of customized LLMs
LLMs like OpenAI’s GPT-4 (the AI engine behind the ChatGPT platform) and others like it have recently demonstrated just how far generative AI has come and the incredible feats it is now capable of. Their use cases range from the whimsical to the job-threatening—and naturally, enterprises of all shapes and sizes are trying to figure out the best way to take advantage of them.
Customization may be the answer, as the expert.ai survey reveals.
“Enterprise specific language models with a human-centered approach are part of the future,” said Marco Varone, founder and CTO of expert.ai, in a press release announcing the survey. Varone says that business use cases of natural language processing—the type of AI that includes LLMs—require domain-specific training beyond that of broad LLMs like GPT-4.
There are many advantages of LLM customization. Not only does it tailor a language model for an enterprise's own data and jargon, but a domain-specific LLM can be smaller, more efficient and faster than general language models without sacrificing performance, according to Varone. Knowledgeable enterprise employees can make it even better, he says, by monitoring and refining training data to ensure accuracy, transparency and accountability.
How to create your own ChatGPT
The expert.ai results are provided in a downloadable report called Large Language Models: Opportunity, Risk and Paths Forward. For those wondering how to create a custom enterprise LLM, the report assures it’s not as hard as you might think.
“Teams often believe the myth that adopting LLMs in the enterprise requires large repositories of proprietary data for training. It’s just not true. In fact, there are many ways to train language models to deal with very specific topics that may only have a few training resources available,” claims the expert.ai report.
It gets even easier. Chipmaker NVIDIA is lowering the barrier to custom enterprise LLMs with its recent release of NVIDIA AI Foundations, a set of cloud-based services for building domain-specific generative AI models. NVIDIA CEO Jensen Huang says that AI Foundations will serve as a “foundry” for custom large language models. Like Varone, Huang believes that generative AI will be a big part of the future—he predicts it will “reinvent nearly every industry.”