More than Generative and Beyond Just Language
"Large Language Model" and "Generative AI" are misleading names for General AI
In the world of artificial intelligence (AI), we now regularly use terms like "Generative AI" and "Large Language Models" (LLMs). However, these labels aren't exactly right. They oversimplify the true nature and potential of AI, leaving us with a less explanatory description of what these tools can really do.
First, let's talk about "Generative AI." Sure, AI technologies like ChatGPT or Stable Diffusion can generate things. They can make text or images, which is pretty neat. But generating stuff is just one part of what they do. The real magic is in how they understand different areas like language, how we see things, and how we think and reason. This broad understanding lets them create useful outputs, which is something that older AIs simply couldn't do.
Now, let's tackle the term "Large Language Models." When we talk about state-of-the-art models, we often use this label. But is it the right one? Not exactly. When we started creating these tools, the goal was indeed to model language, and that's what early attempts like BERT (2018) were able to do. But as much bigger and more powerful models like GPT-3.5 and GPT-4 were deployed, we found that they were doing much more than just understanding language.
These hyper-scale AI models can actually model our world's general knowledge. They don't just understand words and sentences; they grasp facts, ideas, and the way things relate to each other in the world around us. So, while the term "Large Language Models" might describe the technique used to build them, it doesn't capture what they've become: not just models of language, but models of knowledge itself.
Because of this, a better name for these AI technologies is "General AI." That’s what’s unique and special about them - they are general tools that possess general knowledge and can accomplish a very diverse (and as yet not fully understood) set of tasks. We don't usually use this name because people associate a specific acronym, AGI (artificial general intelligence), with far-out fantastical ideas that can derail any attempt at sensible discussion. But these new AI technologies are, at a basic level, a general form of artificial intelligence. They're perhaps not as advanced as we'd like yet, but they're a big step in the right direction.
We need to be more precise with our AI terminology. "Generative AI" and "Large Language Models" aren’t helpful and don't fully cover the reality of these advanced technologies. It's not just about generating outputs or modelling language—it's about understanding a wide range of areas and modelling knowledge on a grand scale. So, let's move towards using terms like "General AI" and "Knowledge Models," terms that do justice to the amazing potential of these AI tools.