🔮 Accelerating workers; prompt engineering; SpaceX, hackers, solar++ #420
Hi, I’m Azeem Azhar. As a global expert on exponential technologies, I advise governments, some of the world’s largest firms, and investors on how to make sense of our exponential future. Every Sunday, I share my view on developments that I think you should know about in this newsletter.
In today’s edition:
AI cuts training times and can help poor-performing workers,
Geopolitical considerations: semiconductors and yuan vs USD,
Solar PV boom, we’re here for it.
Sunday chart: I know kung fu
In a new study, Brynjolfsson, and Raymond and Li show that generative AI significantly impacts labour productivity in certain jobs. The research reveals that generative AI can boost productivity by 14% and enhance lower-performing employees’ performance.
The technology accelerates the learning curve, enabling workers to gain six months of experience within just two months.1
This study is one of the first trying to grapple with the effects of the generative AI wave on labour and economies. What makes this one particularly important, is that it has a high external validity — it studies people in their natural environment, rather than in an experimental setting.
It also further confirms what the study of Noy and Zhang suggested in the beginning of March: AI has the biggest positive impact on the lowest-performing employees. This doesn’t, however, mean that high-performing employees become dispensable.
So what could it mean? First, the highest-performing employees are important to establish the knowledge that AI can then use to enhance poorer performers. Second, organisations must adapt their performance measurement, compensation structures, and workforce management in response to AI's influence. Middle managers may need to reconsider their roles as AI takes on more knowledge transfer tasks.
For a deeper understanding of these findings, listen to my discussion with study authors Erik Brynjolfsson and Lindsey Raymond.
Key reads
Europe’s linchpin. The Dutch company ASML has become a key player in the $580 billion semiconductor industry, as it makes the indispensable high-end semiconductor lithography equipment. This thrust the company at the centre of the US-China geopolitical rivalry.
With the company’s high-end machines churning out chips that can also go into state-of-the-art weapons and artificial intelligence devices, ASML is effectively being treated as critical infrastructure for US national security and has become a target of industrial espionage for China.
The Dutch government stopped ASML from selling advanced machines to China in 2019, and has now tightened export restrictions on older machines. While ASML says this has no impact on its growth for now, its CEO believes that the restrictions will encourage China to develop its own chip equipment. Further reading: Stanford University Professor Steve Blank’s roadmap on how to reorganise US Congress to deter China.
Lawyers at the starting line. The field of legal AI is experiencing a surge in investment and attention, as numerous law firms demonstrate a growing interest in the technology. Harvey, a cutting-edge legal AI startup, already boasts a waiting list of over 15,000 eager law firms. In just a few short months, multiple prominent firms have embraced AI solutions at a remarkable rate. Considering the legal profession’s historically slow adoption of new technology, this is but another signal of how important a competitive advantage AI has become. While GPT-4 passed the bar exam with flying colours last month, this study shows that the large language model still struggles with accounting.
Promptly learn prompting. Microsoft published a detailed guide to prompt engineering techniques. It includes system messages to prime the model with useful context, few-shot learning to teach new tasks, and priming the output. Learning how to prompt is absolutely crucial for every person. First, AI could become our interface with the internet, and perhaps much of the physical world. Second, most AI will rely on the “human in the loop” to provide instructions. Prompting an AI with skill is how you get the most out of the technology, and do so safely.
This study shows an extreme example of why the style of prompting matters: anxiety-inducing prompts lead to the AI showing a lot more harmful bias. (See also, good analysis of what Google’s centralisation of AI products under DeepMind might mean.)
Market data
US pension funds account for 15% of the LP base in German ventures (€4.7b), compared to less than 1% by German pension funds (€94M).
The yuan (48.4%) has overtaken the USD (46.7%) to become the most widely-used currency for cross-border transactions in China for the first time.
Web3 funding dropped 82% YoY.
Men (31%) consume podcasts weekly more often than women (19%) in the UK.
Short morsels to appear smart at dinner parties
🛰️ SpaceX sent out the world’s first 5G satellite, and it only weighs 10kg (22lbs).
🌞 Solar PV passed the massive 1TW globally installed capacity milestone in 2022.
🩻 Dive into this interactive visualisation of 21 million biomedical scientific papers.
🧬 Ancient human genome count surpasses 10,000.
🌫️ The fascinating history of mind-altering substances throughout the 19th century.
🎼 Your native language impacts how you perceive music.
🦹 Hacker group names are becoming increasingly absurd.
End note
Got a few minutes spare? I do recommend the second half of
’s analysis of modularity. Modularity is a key design pattern for new technologies and worth understanding.Have a great week!
A
What you’re up to — community updates
Bill Gross’ Heliogen goes public on NYSE, congratulations!
My friend Craig Cohon continues his 2,600-mile trek to offset his carbon footprint.
Ussal Sahbaz has published a report analyzing Turkey’s digital policies after the May 14th elections, exploring potential shifts if the opposition wins after 22 years of Erdogan’s government.
Ralph Perrine launches a new podcast about change leadership.
Share your updates with EV readers by telling us what you’re up to here.
We generated over 1 million Q&A using various models and we learned that prompting doesn’t change the results substantially. It also doesn’t fix flaws well. Ethan Mollick recently wrote an article concurring. I think prompting will become a very basic skill like search. Both are iterative to a degree. One has a fancy name.
Is the link to the law article accessible? Get taken to Notion but no access.