LLMs confirm a hypothesis the tech world has had for decades. Their existence and rapid adoption signal a shift in cultural acceptance of a new kind of relationship that will shape life past this revolution. Their exponential improvement validates hopes for a more deeply integrated technological experience. But their irresponsible deployment indicates a hard road to cultivating a society that can humanely live and operate alongside this new innovation…
When engineers on the early Macintosh project imagined interaction with a computer, their first inclination was not towards the design of our current devices. The graphical interface was a compromise. It’s always been a stopgap of windows, folders, symbols, and a common design language that we’ve adjusted to over the decades. We often don’t realize it, but it’s not the most natural system for interacting with a tool. Scanning with our eyes, scrolling with our hands, clicking with curiosity. These are fine methods of completing tasks and consuming information, but our dream foresaw so much more.
They foresaw natural language processing. Science fiction has such a hold on our culture not just because it shows where we wish technology to go, but because it shows where technology will go. Jarvis in Iron Man, C3-PO in Star Wars, and HAL in 2001: a Space Odyssey. The dream of the futurist is to gain the ability to effortlessly converse with our devices. This device is now possible in the most general, empowered sense.
The model Siri and Alexa were built upon is antiquated with the invention of large language models. Voice assistants are dependent on voice commands, pre-programmed and reliant upon the user’s conformity to an unnatural language built by Apple and Amazon. “Hey Siri, turn on the lights,” is part of a set number of parameters the assistant is able to comprehend and act upon. If you ask something it hasn’t been given behaviors for, it replies with a simple admission of incompetence.
LLMs are not trained on code words tied to behaviors. Their power comes from predictive engines. Trained on language from the internet (that vast repository of human knowledge), they can decipher the queries put to it and respond with startling accuracy and uncanny prescience.
Already everyday people are realizing the dreams of Kubrick and Lucas by embedding LLMs into speech-to-text voice assistants to achieve the near-perfect synthesis of Siri and a chatbot. This new linguistic interface invites a whole new class of responsibilities, abilities, and formative shifts in the life of the social being…