Software and Computer Engineering

Why Large Language Models Are a Big Deal

If we look back at the last few revolutions that shaped the world in the last century, all of those can be pinned down to an evolution of the paradigm of how work is done.

The disruption of the Industrial Revolution

During the onset of the Industrial Revolution, the shift was from artisanal work to mass-produced goods. The assembly line and the economies of scale allowed a never-before-seen access to consumer goods, unthought-of during previous generations.

While social inequality persisted -- or was arguably magnified -- the general consumer would eventually light up their home during the night, preserve food for longer in a refrigerator, and watch live television. All of this was impossible to achieve in scale in a pre-industrial era.

The rise of the knowledge work

Then, more recently, in the revolution we've experienced in the last few decades, the shift was from the physical work to the intellectual work (although the latter didn't necessarily replace the former).

There was a growth in the knowledge work, as coined by Peter Drucker, where specialists would offer their expert insight and that, despite being intangible, would be valuable and also a form of work.

Many of the high-paid occupations today consist almost entirely of knowledge work: physicians, engineers, lawyers, managers, to name a few. Their entire productive output is knowledge that, when applied to a given situation, can be useful and valuable to other people.

The digital printing press

On that note, the advent of the printing press had a multiplying effect on knowledge work. A single author could now carefully craft a book for a few years, and then the product of their work could benefit thousands or millions at a relatively low marginal cost (cost for each additional unit of a book).

That effect was further amplified on the Digital Revolution we are immersed in, as computer software became the product of the most valuable companies in the world. Computer software, while requiring a large amount of expert work to manufacture, can be designed once and infinitely replicated, benefiting an unbounded number of individuals and businesses.

Where do Large Language Models fit into this

The big shift afforded by Large Language Models (LLMs), such as Chat GPT, is that it proved that, for the first time in history, not only the product of knowledge work could be infinitely replicated, but the production of knowledge work could be commoditized.

LLMs enable the future where expert agents can be trained in a given domain at a great cost, but after that one-time investment, can mass perform knowledge work, servicing thousands and millions of individuals in a context-aware, specific manner, that was only achievable before with a human expert.

While a person is fundamentally constrained in how much work it can handle, how much time it can dedicate, how long they take to learn and become proficient in a given field, LLMs have no such limitations.

Why this is unprecedented change

Models can be, just as software, infinitely replicated. The only constraint being the capital allocated to produce and run the underlying physical hardware. They can run 24/7 and still deliver the exact same level of performance, unlike their human counterparts.

LLMs can always get their knowledge from a replicable repository, while people are born without knowing much, and require learning and acquiring experience first-hand.

Large Language Models will bring an unprecedented disruption on how work is done -- just like the Industrial Revolution and the Digital Revolution did. Knowledge work in particular will be the most affected. The next few years and decades will consist of economical and societal restructuring to accommodate this new reality.