🔮 Parallel bets; peak CO2; AI & red-teaming; seeing pi++ #446
An insider perspective on AI and exponential technologies
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.
The latest from Exponential View
🧠 AI’s first flight — An early milestone in generalised intelligence
💡 My conversation with Dario Amodei, co-founder and CEO of Anthropic
We are hiring
We are looking for a strong generalist with deep interest in exponential technologies to join our team as a Research and Operations Associate. This is a perfect role for an early-career professional with 1-3 years of experience under their belt who wants a balance of intellectual problem-solving and operational execution.
Sunday chart: Who gets parallel bets?
The uncertainty of tech development means that the outcomes of R&D investments are never guaranteed. One strategy to help navigate this is making parallel bets, which provides optionality within uncertain environments. Matthew Ball illustrates how this approach has served Microsoft very well, from the advent of the PC to its current position in the AI market. Amazon, in contrast, placed all of its eggs in the Alexa basket — a single bet that did not pay off. The company was since forced to build up its AI capabilities again, almost from scratch.
The challenge, however, is that only the largest corporations have the resources to employ this strategy of parallel bets effectively. Companies like Google, through DeepMind, have pushed the frontiers in machine learning, and also ventured into governance research. Meta, despite its criticised focus on the metaverse, has built the depth of research capabilities to significantly advance AI.
The question then becomes: who else can employ this multi-faceted approach? The capital and appetite for risk involved make this difficult for smaller entities.
Why does this matter? The potential rewards for major breakthroughs in AI are enormous. If only a few giant corporations have the means to push the boundaries in this field, can the benefits of AI advancements be equitably distributed across society, and the world?
Key reads
Neural greatwork. A significant breakthrough in the neural networks domain has been achieved, addressing a longstanding debate about their capacity for human-level systematicity—a nuanced understanding of systems and concepts. This advancement came through a new training methodology known as meta-learning for compositionality, propelling neural networks to exhibit human-equivalent systematic generalisation. This approach not only nudges us closer to achieving Artificial General Intelligence (AGI) but also mirrors some of the foundational conceptual structures that make up human intelligence. (More technical paper is here.)
See also, an excellent discussion with DeepMind co-founder, Shane Legg, where he explores the missing pieces for building AGI.
Public eye. Red-teaming is one of the commonly used strategies to test system’s vulnerabilities, including AI’s. The idea is to get individuals to attack the AI system and identify the flaws. But this strategy is difficult to scale when it is within a company, and doesn’t offer any accountability for companies to disclose or combat any found vulnerabilities. Since generative AI’s take-off, companies have relied on unpaid online communities like those on Reddit to conduct such work — but there’s a lot more that we need to rethink when it comes to this important security practice: what red-teaming is necessary, at what stage of a product lifecycle, and should it be paid for and made in the public eye — these are some of the questions that need tackling. This concept aligns with the broader idea of public deliberation and scrutiny, evolving in ways even thinkers like philosopher Jürgen Habermas might not have imagined.
See also, the organisations attending the AI Safety Summit have been announced.
More passion, more energy. In 2003, the International Energy Agency published a report quantifying global energy investment needs: “the global financial system has the capacity to fund the required investment, but are the conditions right?”. Twenty years later, the IEA showcases the emergence of a new clean energy economy, led by solar and EVs. The pace of change is very fast, with more than $1bn (!) a day spent on solar deployment. The result — according to Carbon Brief — is that global CO2 could peak as soon as 2023. Beyond solar and EVs, some technologies still need a nudge from the public sector. A new analysis found that “Moving half of the funds spent on subsidies for fossil fuel heating to heat pumps could transition all European home heating to renewable methods by 2040.”
Community
😎 AI prompting workshop: Members are reaching out expressing interest in a prompting workshop to improve their use of AI. If you are among the interested to go deeper on generative AI and prompting with us in a dedicated workshop, please fill out this brief interest survey — this will help us craft our programme and get you where you need to be to take advantage of AI.
💭 Embodiment series: Join our members in a three-part series titled What Does it Mean to be Human in Dataspace? hosted on Zoom by Robbie Stamp, Chanuki Illushka Seresinhe and Luna Lacey. The first of the three events is scheduled for 8 November, 6-7pm GMT. Register here.
Market data
A study on French firms mandated to share excess profits with their employees reveals that it increased the labour share of firm income by 1.8%, reduced profit share by 1.4%, without impacting investment or productivity.
Daily active users on X have fallen by 16% since Musk took it over. Advertising revenue is down 54%.
In 2021 Google paid $26bn to be the default search engine on web browsers and mobile phones.
OpenAI is negotiating a deal potentially valuing it at $80 billion, almost tripling its worth in under six months, positioning it as San Francisco’s highest-valued startup.
Norwegian petrol and diesel new car sales have fallen from 100% in 2010 to only 4% so far in 2023.
Short morsels to appear smart at dinner parties
🧪 Destroying the forever chemicals.
🎻 Gig work is inspiring a new genre of songs.
☂️ Apple’s weather app just can’t seem to get it right.
🥧 Hearing and seeing pi as an irrational number. Stunning.
🔋 An illustrated history of electric vehicles, including a risky-looking electric pram.
🔬 Scientists have found out that an average person’s immune system consists of about 1.8 trillion cells (285714 times the number of sheep in Ireland), weighing around 1.2 kg (equivalent to 1.2 L of water). h/t Dr. Eric Topol
💫 Deflecting laser beams off nothing but air.
End note
It’s AI week - with the UK’s AI Safety Summit kicking off, bringing together many of the leading thinkers, technology firms, major governments and some randos to explore what governance mechanisms for AI’s highest risk areas could look like.
It’s a useful milestone to try to figure out what multilateral discussion agreement could look like, and it’s good to see that Chinese researchers are also represented. But we need to ensure that while this is a helpful first step, it cannot be seen to be the totality of what the public discussion around AI needs to be. I suspect this event will have a strong focus on security issues (the bioweapons risk) - and if it does, and creates a regime of observation and collaboration, could lead to further collaborations.
I’m intrigued to see how this first step plays out.
A
What you’re up to — community updates
A number of EV readers have been appointed to Antonio Guterres’ AI Advisory Group.
James Wise’s new book Start-Up Century is now available for order ahead of its release on November 23rd.
Gianni Giacomelli shared a case study about using genAI as a personal problem solver.
Nick Russ and his team at Axiom Space have announced a memorandum of understanding with the U.K. Space Agency for a U.K. space mission.
Laure Claire and Benoit Reillier are running the Platform Leaders conference on Nov 9th on the future of digital platforms. Register for free.
Share your updates with EV readers by telling us what you’re up to here.