In the realm of data and artificial intelligence (AI), the buzz often surrounds the “wow” factor, but for many business managers, the focus is on achieving practical, real-world results. Amid the inundation of news about cutting-edge AI like ChatGPT and Large Language Models (LLMs), managers in non-tech or traditional industries may find themselves wondering if these advanced technologies are relevant to their daily operations.
Large Language Models, or LLMs, offer more than just text generation capabilities. These models, like GPT-4, are incredibly versatile and can be harnessed to perform a wide array of tasks. The key to unlocking this potential lies in crafting precise instructions, often referred to as “prompts,” which guide LLMs to execute specific tasks. This evolving skill of crafting effective prompts is commonly known as prompt engineering.
Consider scenarios where an LLM can identify languages in customer reviews, translate corporate documents, or assess the sentiment of customer complaints. In the pre-LLM era, achieving these tasks would have necessitated the development or procurement of separate systems for each function, entailing additional expenses for data collection and hardware. However, with LLMs’ cost-effective pay-per-use pricing, these use cases become more economical than developing numerous specialized models. Consequently, LLMs are evolving into a versatile technology for various applications, essentially becoming a dominant design.
Seamless integration into workflows
While many have explored ChatGPT through its web interface, LLMs and their underlying technology are available via software APIs, making integration into applications and business workflows remarkably affordable and straightforward. Examples like Duolingo utilizing GPT-4 for improved user feedback and Morgan Stanley leveraging LLMs for financial recommendations demonstrate the diverse applications of LLMs. Open-source projects, such as langchain, enable the creation of programs by composing chains of LLM-based instructions, paving the way for even more innovative solutions in the future.
The disruptive aspect of LLMs is the way they enable programming with AI. Developers can specify tasks using simple language instructions and combine results like Lego bricks, offering an alternative to the traditional “buy-vs-make” dilemma, where you can “make vs. buy or adapt from an LLM.”
Interrogating company data with LLMs
Another promising application of LLMs is querying company data. LLMs can provide answers to data-related questions when the data is included as part of the prompt. This approach was tested by Boston Dynamics, which used ChatGPT to inquire about a robot’s battery level or internal state. This opens up opportunities for corporate knowledge bases. It’s now possible to input a bundle of business documents or data records into a prompt and ask questions about them, potentially replacing traditional SQL queries in the long term.
Efforts are underway to connect document databases with LLMs more efficiently, such as searching for the most relevant documents and only incorporating those into the context. Despite challenges, LLMs are emerging as a new interface for interacting with data.
Interoperability and action-taking with LLMs
LLMs are also breaking ground in their ability to “connect” with other systems and take actions when necessary. The ReAct prompting strategy, introduced by Google and Princeton researchers, enables LLMs to generate answers that include calls to services. For instance, an LLM could connect to a calculator service to provide an answer to a calculation request. Premium versions of ChatGPT offer plugins like Expedia and Kayak, providing access to booking information and handling complex mathematical queries.
Overall, this ability for LLMs to interact with other systems has the potential to transform them into platforms or marketplaces for various services, akin to how phones serve more functions than just making calls.
LLMs: Affordable, adaptable, and accessible
In closing, business managers should not dismiss LLMs due to perceived complexity. LLMs are readily available as flexible and cost-effective tools on a pay-per-word basis, often offering rates cheaper than alternatives. They are rapidly evolving into all-purpose, generic technologies, replacing numerous smaller models and offering a new type of programming through word-based instructions. LLMs efficiently handle data queries and have the potential to serve as platforms for various services.
