📊 EV’s Charts of the Week #114
GPTs as GPTs, Mastodon rises, happiest countries++
Hi, I’m Azeem Azhar — I advise governments, some of the world’s largest firms, and investors on how to make sense of our exponential future. I wrote a book about the forces shaping this future — it’s a good place to start if you’re new to my work.
As I announced two weeks ago, the team and I are rethinking how we tell stories through this Wednesday Chart newsletter. We’re experimenting behind the scenes with a new way of researching and narrating the questions we care about.
Expect the new format to hit your inbox in a couple of weeks — if you’re a paying member you’ll be the first to benefit from our research. In the meantime, please enjoy an abbreviated edition of our Charts of the Week.
GPTs as GPTs
A new study, co-authored by OpenAI researchers, tried to measure the possible impact of Large Language Models (LLMs) on different jobs and industries1. They used exposure to a Generative Pre-trained Transformer (GPT) as a proxy for economic impact and applied this to a multitude of tasks. A task is considered exposed if access to a GPT could reduce its completion time by at least 50%. Human annotators judged whether tasks were exposed, which enabled the authors to estimate which jobs were most exposed, depending on which tasks they included. (The study also asked GPT-4 to estimate how likely it was that tasks could be automated.)
The study came out with two major findings.
First, the authors expect that 80% of the US workforce could see at least 10% of their tasks affected by GPT. In other words, most Americans could benefit at least a bit from these LLMs.
Second, 19% of American workers might have at least 50% of their jobs exposed to AI. One in five workers could do their job much, much faster. Or perhaps find downward wage pressure because of the arrival of this labour-saving technology. These workers tend to be in medium- to high-income jobs. This runs contrary to the long-held idea that AI first automates low-skill jobs, and tends to impact high-skill and creative jobs last.
When we compared the results to income distributions in the US, the point came across very clearly. While the median wage in the US is $56k a year, the median income for the degree educated is $84k per annum, right into the meat of the exposure. (When looking at our chart below, please note that we’ve extrapolated from the researchers’ wage data which was presented as a logarithm rather than absolute values. We’ve also got rid of the error bars for simplicity. Exposure, finally, is best interpreted as a relative measure rather than a precise scale.)
Of course, this is but one study. But it is a very helpful start.
The rise of the mammoth
Decentralised social media platform Mastodon hit the 10 million users milestone. No one is quite sure why, and a surge in bots can’t be ruled out (although Mastodon makes it hard for bots to join.)
Happiness is a warm… GDP per capita and social support
The annual World Happiness Report ranks countries by GDP per capita, social support, healthy life expectancy, freedom to make life choices, generosity, and freedom from corruption.
This type of ranking stays quite constant, mainly because they measure the health and trustworthiness of institutions and the freedom2 and well-being these institutions support. Afghanistan is last in the ranking: it also has amongst the weakest, least functional institutions.
Love our Charts newsletter? Forward the edition to your friends and colleagues — every recommendation helps us grow!
The original purpose is to figure out whether GPTs are “general purpose technologies.” As I explained in my book, AI as a general-purpose technology means it has a widespread impact across industries and society and intersects with other technologies to spur more innovation. The study’s authors concur with my view that it is. But the headliner is the analysis of the impact on tasks conducted by workers at different wage levels.
As with all international rankings, these results must be taken with a pinch of salt. Saudi women may disagree that their country is 30th globally, given their severe lack of freedom.
One of the most interesting potential impacts on wages from GPT will be whether or not is depresses wages / income for the a meaningful percentage of the select fields. This has played out previously in other fields where the diffusion of certain tools improves the overall experience for the end user / consumer but guts certain professions, particularly those where the bulk of the workers are individuals who I would argue are more subject to downward pressure as their profession gets commodified.
The report on happiness - Why is Taiwan called Taiwan Province of China?