🔮AI eats software; Biden's biotech; Taiwan's gold; space rice++ #415
Hi, I’m Azeem Azhar. I advise governments, some of the world’s largest firms, and investors how to make sense of our exponential future. Every Sunday, I share my view on developments that I think you should know about.
In today’s edition:
How AI might eat the traditional software business,
Biden boosts bioeconomy builders,
Better imagination through AI.
⏱️ Read time: 10 minutes
Today’s edition is supported by our knowledge partner, Singularity Group.
As Exponential View readers know well, the Exponential Age is here. Singularity’s Executive Program is designed to keep you ahead of the curve. Hosted in Silicon Valley, this 5-day in-person program will immerse you in the fundamentals of exponential mindset — and help you identify the abundant opportunities inherent in exponential technology.
Sunday chart: AI ate software
“The current generation of [LLMs] is a missile aimed, however unintentionally, directly at software production itself,” argues investor Paul Kedrosky. The language of software is structured and formal so lends itself well to large language models. These models can generate code, even for non-coders, making software production more accessible and less complex. (The article itself is a bit long, so jump to the section labelled “Demographics, Aging, and the Coming Labor Disruption from LLMs.”)
The Microsoft paper from earlier this week pointed out that “Remarkably, GPT-4 executes the code directly, without translating it into other well-defined programming languages. This demonstrates the potential of AGI models as a novel tool for programming with natural language.” A technical example of this comes from Simon Wilison who uses OpenAI’s new plug-in architecture to create an interface to a remote database just using natural language.
From the perspective of the labour market, I suspect wage polarisation is a more likely outcome in the next few years. That is, the very best human coders (building these models) will be able to charge more, while the rump (say bottom nine deciles, for sake of argument) experience downward wage pressure. Equally, we’re seeing hints of new roles emerging, such as people employed to fine-tune, feedback, monitor and manage the performance of these models.
From the perspective of the economy, the consequence might be to unfetter our imaginations on what could be built.
See also: Demis Hassabis joins the chorus of voices calling for caution in the pace of AI development. Bill Gates also notes that the “Age of AI” is only beginning.
Weekly Commentary
In my members-only weekly commentary, I argue that LLM-enabled services will be a paradigm shift in how we experience the web. In fact, such AI interfaces will represent the thing that follows Web 2.0 (call it Web 3 to distinguish it from Web3, the blockchain-based vision that has failed to deliver.)
Key reads
Artificial ideas: Go was a landmark in the development of AI. Deepmind’s AlphaGo was the first computer program to beat a professional player, and did so by playing against itself and learning in that way. Today, Go AIs are helping study human-AI cooperation. The authors of this study asked an AI to evaluate the quality of human Go players’ decisions across time, and found that players make significantly better decisions since the advent of “superhuman AI”. The hypothesis is that players use more novel, previously unobserved moves thanks to these AI programs. This in turn improved their decision-making.
Nature’s factory: President Biden sets out some (self-proclaimed) bold goals in this biotechnology and biomanufacturing strategy. The idea is to shape a post-fossil fuel industrial economy by taking advantage of biology converging with other general-purpose technologies. Biden’s ambition includes decarbonizing chemicals and materials sectors and putting sustainability at the heart of agriculture. Some goals include: 90% of all plastics becoming recyclable-by-design, and 30% of all US chemical demand through biomanufacturing pathways within 20 years. I think we stand a chance of blowing through both of those.
21st-century gold: Taiwan’s TSMC makes a third of the world’s silicon chips. This fascinating essay by Virginia Heffernan explores the extraordinary complexity and importance of the “fabs”, the factories that produce the chips that have become indispensable to us in less than a century. It’s difficult to understate the importance of such a place. Taiwan is at the centre of battles for territory, and also the key provider of a technology of central geostrategic importance. Many governments are now trying to build up a domestic microchip industry, but Taiwan will remain the main player for some time to come.
Gordon Moore, the original exponentialist, has died. Obituary here.
Market data
EV readers, Paul Daugherty and James Wilson, forecast that 40% of working hours across industries will be impacted by LLMs.
Spending on gaming is 70% higher than spending on streaming platforms with 3.2bn people playing video games worldwide in 2022. Gen Z is driving gaming’s growth, with 9 out of 10 British 16-24 year-olds playing video games.
UK heat pump adoption was among the worst in Europe at 88 sales per 100,000 people. European leader Finland was over 40 times higher.
Bye-bye instant delivery apps: e-grocery investment dropped by 73% year-on-year in 2022.
A place for vehicle-to-grid energy storage? 11 times more batteries went into EVs than grid storage in the first quarter of 2023.
Short morsels to appear smart at dinner parties
🐚 We finally have the nominees for the much anticipated “mollusc of the year” award.
🫶 60% of Replika’s (an AI chatbot companion) paid subscribers had a romantic relationship with their AI chatbot.
🎙️ Using electrical background noise to date a recording.
🪤 A hormone shot sobered mice up.
🎹 An analysis of a lock of Beethoven’s hair reveals his cause of death.
🍚 China managed to harvest rice in space.
🇨🇳 At the same time, China’s demographic problem is hitting the capital: Beijing’s population declines for the first time since 2003.
End note
At some point this week, trying to keep up with the speed of movement in the AI space, I caught myself thinking “f%$^@£&!ing Exponential Age.” I’ve had a chance to speak to many of the key players in the field, both research labs and industry, as well as businesses building tools on top of them. And the demand from bosses and boards to make sense of this is keeping me busy.
We’ll continue our explorations through the newsletter and various commentaries for paying members in the coming weeks, including exploring sovereign AI, securing social licence, thinking about regulation, exploring open vs closed models, risks and more.
I’ve continued experiments with GPT-4. This week I got the system to play with a counterfactual scenario. In this case, I asked what might have happened if the Spanish invaders of South America had been susceptible to a local disease (rather than decimating local populations with smallpox).
The output is reasonable. It doesn’t matter whether GPT-4 is showing evidence of reasoning, or merely parroting out symbols that make sense in this case. The point is that it is quite an advanced use case, not a million miles away from analysis I do as part of my professional life. And this whole process took less than 10 minutes while I was multitasking. You can read it here.
Have a great week!
A
What you’re up to — community updates
CEO Annalee Bloomfield’s Sustain.Life has been certified as a B Corporation.
Chris Mastrodonato launched a new podcast about how complexity shapes innovation, organisations and products.
Ben Carey’s team launches Climate Science Translated, enlisting top comedians to translate climate science in more effective and emotional way that drives action.
Share your updates with EV readers by telling us what you’re up to here.
If only writing the actual code was the hardest part of software engineering.