The terminology we use to describe technology shapes how we think about it, understand it, and ultimately interact with it. While “Large Language Model” or “LLM” has become ubiquitous in media coverage, technical discussions, and everyday conversation, this widespread adoption of an increasingly inaccurate term poses real challenges. When we continue to use language that undersells or misrepresents the capabilities of these AI systems, we risk limiting our understanding of their potential applications, their limitations, and the ways they might impact society. This isn’t merely a matter of semantic preference – precise terminology helps researchers, developers, policymakers, and the public have more informed discussions about AI development and governance.
When artificial intelligence models like ChatGPT and Claude first gained widespread attention, they were commonly called “Large Language Models” or LLMs. While this term accurately described their initial focus on processing and generating human language, it has become increasingly inadequate as these systems have evolved to handle much more complex tasks.
Today’s frontier AI models can analyze images, solve mathematical problems, write and debug computer code, and engage in structured logical reasoning. They can break down complex problems into steps, integrate information from multiple sources, and even interact with external tools and databases. This expansion of capabilities goes far beyond simple language processing, making the term “Large Language Model” somewhat misleading.
A more accurate term that has gained traction in the academic community is “Foundation Model,” introduced by Stanford’s Institute for Human-Centered Artificial Intelligence. This term better captures how these systems serve as a foundation for many different tasks and can work with various types of information, not just text. It also reflects how they can be adapted and fine-tuned for specific purposes while building upon a common base of capabilities.
Another relevant term is “Large Multimodal Model” (LMM), which specifically acknowledges these systems’ ability to work with multiple types of input and output – text, images, and sometimes even audio or video. This term has become particularly relevant as more AI models incorporate the ability to understand and generate various forms of media.
The shift in terminology isn’t just about being technically accurate – it’s about helping people better understand what these systems can and cannot do. When we call something a “language model,” it suggests its primary function is processing and generating text. This can lead to underestimating the system’s problem-solving capabilities while perhaps overestimating its actual understanding of language. These models don’t just predict the next word in a sentence; they engage in forms of pattern recognition and structured reasoning across multiple domains.
However, it’s worth noting that no single term has yet emerged as the perfect replacement. While “Foundation Model” and “Large Multimodal Model” each capture important aspects of these systems, the AI community continues to debate the most appropriate terminology. This ongoing discussion reflects both the rapid evolution of the technology and the challenge of accurately describing systems with such broad and complex capabilities.
