LLMs for the Rest of Us
Large language models (LLMs) have exhibited one of the most rapid trajectories of any newly introduced technology. The initial release of GPT-3 by OpenAI to developers and researchers in May 2020 was followed in November 2022 by the release of ChatGPT to the general public. In the first week of its release, ChatGPT acquired 1 million users. Meta soon followed with its open source Llama in February 2023, and Anthropic debuted Claude in July 2023. It was a technology whose time had come.
LLMs are the product of statistical analyses of large amounts of data from public sources such as the internet and document repositories. By converting words and phrases to numerical values, referred to as embeddings, the LLM can perform calculations that allow it to generate natural language responses to inquiries or commands. LLMs are built on neural networks, which identify patterns and relationships among data elements that allow them to predict sequences of words that emulate human speech. Parameters are settings in the LLM that control the output, such as the length of the response. LLMs contain billions or even trillions of parameters.
Together, these elements have created complex applications that are qualitatively different from any in the past, allowing unprecedented interaction with vast amounts of data in a natural language mode. Even in the early stages of use, their value has been broadly recognized. According to Grand View Research, the global market for LLMs was $5.6 billion in 2024 and is expected to grow to $35.4 billion by 2030, reflecting a growth of 23.7% per year. Chatbots and virtual assistants accounted for 27% of total revenue.
Although the early adopters of LLMs were large corporations in healthcare, finance, and manufacturing, individuals and small and midsize businesses (SMBs) were soon able to access them because of the availability of free or low-cost options and ease of use. The interface allows users to simply type in questions or instructions. Some small organizations that experimented with LLMs internally found that they had marketable products.
Evolution Through Experimentation
When a small pizza shop in San Francisco wanted to make a fully functional website for its customers a decade ago, it was held back by the lack of technical expertise. Fortunately, expertise was available from two patrons of the establishment who jumped in to help revamp the website and provide core functionality. They quickly realized that many similar small businesses lacked the resources and knowledge to enable their websites to perform certain tasks. They founded POWR (powr.io) to fill this gap. The two had marketing and technical expertise and eventually produced 60 widgets and other products that now reach 5 million end users globally.
As AI emerged, POWR began integrating it into its products, including a process for generating images for its slider carousel and a library of over 10 million forms from which Form Builder by POWR can select elements to construct a customized form for users. POWR also applied its expertise in-house. Mick Essex, who joined the company in 2022 as head of growth marketing, analyzed his own time, looking for patterns and redundancy, and found that he was spending 10-plus hours a week editing content submitted by its stable of writers and guest authors.