dan/Getty
Are large language models (LLMs) our new operating systems? If so, they are changing the definition of what we consider to be software.
Also: 8 ways to write better ChatGPT prompts - and get the results you want faster
Several analogies are used to describe the impact of fast-evolving AI technologies, such as utilities, time-sharing systems, and operating systems. Andrej Karpathy, co-founder of OpenAI and former senior director of AI at Tesla, believes that an operating system is the most apt analogy. In this scenario, LLMs are essentially new kinds of computers that orchestrate memory and compute for problem-solving activities.
LLMs are complex software ecosystems, Karpathy explained in a talk at a recent startup conference in San Francisco. There are many parallels between these ecosystems and the operating systems of yore. Tellingly, LLM environments "have a few closed-source providers like Windows or Mac OS, with an open-source alternative like Linux," Karpathy related. "The Llama ecosystem is a close approximation to something like Linux."
Also: How to turn AI into your own research assistant with this free Google tool
Utility and timesharing computing are analogies that can be applied to LLMs because they are ubiquitous and their build requires high capital. "We're in this 1960s-ish era, where LLM compute is still very expensive for this new kind of computer," Karpathy explained. "That forces the LLMs to be centralized in the cloud, and we're just thin clients that interact with it over the network."
This cost and complexity mean that, in a sense, a "personal computing revolution" hasn't happened yet with LLMs, since "it's just not economical and doesn't make sense," he added.
Unlike current operating systems, a common graphical user interface has not been developed for LLMs. "Whenever I talk to ChatGPT or some LLM directly in text, I feel like I'm talking to an operating system through a terminal," said Karpathy. "A GUI hasn't yet really been invented in a general way. Should ChatGPT have a GUI different than just text bubbles? Certainly, some apps have a GUI, but there's no GUI across all the tasks."
However, despite these challenges, progress is rapid. Karpathy suggested that we are entering the era of "Software 3.0." While the development of programs in Software 1.0 involved coding into a system, and Software 2.0 was based on neural nets, Software 3.0 uses prompts "in our native language of English."
... continue reading