🔮 Tech’s right shift; predicting complexity; conservatism failure; longevity, synthetic blood & robots ++ #483

An insider’s guide to AI and exponential technologies

Ideas of the week

Smol is beautiful. Back in my horizon scan in January, I said “[s]maller models will be in demand: they are cheaper to run and can operate on a wider array of environments than mega-models.” Now with the release of GPT-4o mini, this argument is even stronger. It is the best-performing cheap model on the market – nearly 30 times cheaper than GPT-4o, with roughly the same performance as the original GPT-4. Since GPT-3’s May 2020 release, prices have dropped an astounding 99%. The lower prices will make running multiple AI agent systems, which iteratively refine tasks through multi-step workflows, much more affordable. As Andrew Ng highlighted, these systems can lead to significant performance gains.

Tech leans right. Musk, Andreessen and Horowitz have confirmed their move to the right, with substantial donations to Trump’s campaign. This practice of buying political influence is common across US democracy despite its perverse impacts:it is a form of rent-seeking, after all. (And no wonder such donations are strictly limited, if not banned, in other advanced democracies.) As

Noah Smith

notes, the shift reflects tech’s maturation – as growth slows, taxes and regulation have become pressing concerns. Regulation is necessary after the failure of two decades of self-policing, but those in the industry moving to the right are growth-centred rather than human-centred in their regulatory ethos. See also: Om Malik argues regulators should focus on big tech’s control of digital identities and data, rather than acquisitions.

Cracking complex systems. A new deep learning model can forecast tipping points in systems with hundreds of interacting parts – a leap beyond previous methods which were limited to simple scenarios. The algorithm accurately predicted sudden shifts in three complex systems: coordinated rhythmic behaviour, ecosystem resources, and brain activity patterns. When tested on African vegetation patterns, it anticipated critical rainfall thresholds long before transitions occurred, with low prediction errors. This is an important step forward in our ability to foresee and potentially prevent catastrophic changes in everything from climate systems to financial markets. Is it a case of data uber alles?

Has conservatism failed? Jon Askonas argues it has, offering a fresh perspective on rapid societal changes by emphasising technology’s role over political ideologies. Askonas points out that while conservatives have long championed the protection of tradition, they’ve largely viewed technological progress as an unstoppable force rather than something that can be guided or shaped. This passive stance meant that conservatism failed to actively guide or restrain technological changes that disrupt traditional ways of life. As a result, technology, more so than progressive policies, eroded traditions. This will only worsen as we head deeper into the Exponential Age, where increasingly faster technological transitions will accelerate societal shifts. (For another exceptional perspective on the existential questions facing liberal conservatism, do read this essay by Edmund Fawcett, with whom I used to work.)

---

Data

Many used EVs in the UK are now equal to, or lower in price than, petrol equivalents.

China and India accounted for nearly 50% of aggregate global productivity growth from 1997 to 2022. 

Each additional robot per 1000 workers decreased the average lifetime earning potential in local US labour markets by $3,900 between 2004-2008.

Amazon says it has matched 100% of its global operations’ electricity use with renewable energy in 2023 – 7 years ahead of schedule. However, this was mostly via energy certificate purchases (aka smoke and mirrors).

Cloud computing cut US energy costs by an estimated $2.8 to $12.6 billion in 2017.

67% of surveyed participants attributed some possibility of consciousness to ChatGPT, while 33% believed it was definitely not conscious.

In the Roman Empire, it took 40 years of labour to make a tonne of copper. Now it only takes 21 days.

---

Short morsels to appear smart at dinner parties

😭 Chinese text-to-image model Kling ‘reunites’ users with their deceased relatives through animated photography.

👴🏼 Gerontocracy is not new. 

👀 Most investors are so risk-averse that they fail to gain from their successes. 

⚒️ An interesting read on the chilling effect the anti-communist crusade of the 1940s had on Hollywood and American society. 

🩸 Why is synthetic blood not a thing yet? 

🇨🇳 Connections and family wealth are now seen as the primary paths to success in China. There’s a shift from 2014 when it was ability and talent.

👵 Research confirms that inflammation blocking is pivotal for longevity. 

💡 A short history of the “lightbulb conspiracy”, an industry-wide planned obsolescence in the 1920s. 

---

In focus

Here’s a preview of my Saturday column. I examine a different way to think about universal basic income in the Exponential Age.