Design by Artis Briedis
Watching Erik Brynjolfsson’s presentation at Wharton (more on that experience here) helped me to recover my (already lost by then) confidence that even a technology layman can understand the basics of artificial intelligence. Finally, I had found someone who was not talking all the time about the nuts and bolts of the technology (far beyond my limited comprehension) but rather about the effects it will bring to the economy and the opportunities it will create for business.
Encouraged by this experience I started looking for Erik’s other videos. Here are my three takeaways after watching his introductory class to students at Stanford earlier this year which is available on YouTube:
Lesson #1
You probably need to be a trained economist these days to know that “GPT” stands not only for “generative pre-trained transformer”, as we have been led to understand, but has an entirely different meaning – that of “general purpose technology”.
To be considered for a general purpose the technology has to be:
- pervasive, or such that it affects a broad range of sectors and industries;
- able to be improved over time; and
- able to spawn complementary innovations, which is perhaps the most important feature.
The Watt steam engine became the first truly general-purpose technology in history that affected the global GDP and caused an exponential growth in living standards. Later on, electricity had a similar effect. Today the generative pre-trained transformer is already being called a GPT with the capital G.
Lesson #2
The economic impact of artificial intelligence is lagging behind the technology. Even if all the technological development suddenly stopped for 5-10-15 years there will still be business innovation and productivity growth.
Erik quotes Yann LeCun, a computer scientist specialising in machine learning and computational neuroscience, as having said that “a house cat has more common sense and understanding of the world than any LLM”. There seem to be other (better) technologies already under development but, irrespective of that, LLMs are still going to have trillions of dollars in economic impact.
Despite all the shortcomings, large language models are getting increasingly better – ChatGPT 3.5 scored better than 10% of humans taking the bar exam to qualify as lawyers; ChatGPT 4.0 already surpassed 90% of the exam takers.
Lesson #3
Technologies can be either substitutes or complements. Substitutes can perform the same function as already existing technologies do and can easily replace each other. As a result, they drive lower the value – if technology can substitute human labour, for instance, it drives down the value of human labour.
The complements, on the contrary, make the other parts more valuable – be it, for example, the right shoe which is a complement to your left shoe, or some software to your hardware. Historically most of the technologies have proven to be complementary.
Progress has not so far and is unlikely to force the labour out. The reason why we are paying more for labour today is because we have all the different machines and technologies.
And, as a final note, large language models become very efficient in learning how to deal with the problems that occur with high frequency. This is one of the reasons behind the slow progress with developing self-driving cars – there is a long long tail of issues that do not come up often but are critical for driving safety which need to be resolved.
For those willing to try and pick top lessons of their own here is the link to the video.
* from anything that you are reading, watching or hearing you can realistically expect to remember only a limited number of things. My solution is to pick just 3 items or ideas from any material. This number is non-negotiable. Even the most extraordinary experience gets compressed into 3 things to remember. This approach has worked well for me.
This note was first published on medium.com on 26 October 2024.
Aivars Jurcans has more than 20 years of corporate finance and investment banking experience. His services are currently available through Murinus Advisers. More of his writings can be found on his page Corporate Financier’s Notes.
