🔮 AI, trust, & lawyers' future; China's chip success; face swap, robot swarms, obesity drugs ++ #460

An insider’s guide to AI and exponential technologies

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. Every Sunday, I share my view on AI and other exponential technologies in this newsletter. I am also on LinkedIn, Threads, and Substack Notes.

---

Latest posts

If you’re not a subscriber, here’s what you missed recently:

• 🔮 The brilliant, complicated simplicity of ChatGPT

• 🔮 Existential risks’ existential problem

• 🔮 Opening the AI black box with Richard Socher

---

Sunday chart: Highway to hell

The last 12 months saw the planet’s average temperature exceed 1.5ºC above pre-industrial levels. A multi-year average, rather than a single year, is what scientists pay attention to, and these past months have likely been buttressed by the warming effect of El Nino. The yearly increase is a wake-up call — as are record sea surface temperatures and the continuing collapse of Antarctic ice.

A new study¹ found that there was evidence of warming in the oceans in the mid-1860s, 80 years earlier than originally thought; it also suggests that the warming is 0.5°C higher than the IPCC estimates — although, even if warming did start before 1850, it doesn’t necessarily mean that climate impacts would come any quicker. If this study resets our baseline for what pre-industrial temperature levels are, it could put us closer to the 2ºC limit than previously expected. It implies that the opportunity to limit global temperatures to 1.5°C, the 2015 Paris Agreement limit, has passed — and the mean land temperature may exceed 2.5°C by 2035 if we keep going with business as usual. The scientists suggest that the realistic goal we must work towards is keeping the combined land and ocean temperatures below 2°C.

---

Key reads

Where the chips may fall. Chinese chipmaker SMIC, with Huawei, is crafting production lines in Shanghai to develop 5nm processors, which shows that China’s semiconductor industry is making progress despite international restrictions. Although the 5nm chip is a generation behind the state-of-the-art 3nm, it is good enough “to upgrade Huawei’s new flagship handset and data centre chips”. The US has been putting increasing restrictions on China’s access to advanced semiconductors and chip-making equipment in an attempt to stop it from using powerful chips for its military, and becoming a cutting-edge tech and AI superpower.

As I wrote in November 2020:

[US] pressure is likely to spur Chinese activity in this sector… I think it’s highly unlikely that the US can stop China developing a semiconductor industry. The knowledge is out there and there isn’t some scarce natural resource (like yellowcake) that can be controlled.

Currently, the production lines are worse than those of TSMC, with yields a third lower and costs about 50% higher for a similar 5nm node. But if the alternative is not accessing the technology, perhaps that is a price worth paying.  

China's chip opportunity

See also:

• $7 trillion: Sam Altman’s new plan for a chip industry that can support the demands of AI.

• A new stamping chipmaking technique by Canon could challenge ASML’s domination on lithography machines.

A stamp of trust. OpenAI is integrating new watermarking features into DALL-E 3 images, including invisible metadata and a visible CR symbol. Meta, too, is watermarking images generated through its AI tool, but it struggles to identify third-party AI-generated content. Authentication and watermarking is part of the Biden administration’s standards for AI safety and security, and government agencies are expected to set a good example and give proof of authenticity in their comms with citizens.

Watermarking has limitations, however. University of Maryland researchers have shown that there is currently no reliable watermarking for AI images, “we broke all of them”. We need to rely on the Swiss cheese model of risk management rather than a silver bullet. 

Legal ease. Junior lawyers and legal outsourcers were pitted against ChatGPT in a new study. The conclusion:

Advanced models match or exceed human accuracy in determining legal issues. In terms of speed, LLMs complete reviews in mere seconds, eclipsing the hours required by their human counterparts. Cost-wise, LLMs operate at a fraction of the price, offering a staggering 99.97% reduction in cost over traditional methods [aka humans].

Needless to say, this could herald a seismic shift in the legal industry. The AI models were wildly cheaper and more than ten times faster. A fully-blown piece of legal tech could incorporate multiple layers of LLMs with traditional techniques and likely reduce the error rate to below the junior lawyer level — while retaining a two-order-of-magnitude cost advantage.

See also:

---

Market data

A major rise in unemployment, similar to that during the Great Recession, could lead to a 2.3% decrease in the average death rate.

Face swap attacks on ID verification systems rose 704% in 2023.

Total internet bandwidth is now 1,217 Tbps, growing at a 28% annual rate. In 1990, the total volume of traffic on the internet was roughly one terabyte per month. Today’s network could transfer that volume of data in 0.007 seconds.

Apple’s secretive self-driving car project, known as Project Titan, has dramatically increased its testing in 2023, driving four times more miles than the previous year.

In 2022, Alphabet spent $6 billion buying up companies. In 2023, that number dropped over 91%, with no major deals struck.

---

Short morsels to appear smart at dinner parties

🐝 One person can supervise a swarm of more than 100 robots.

🎂 “It might be nice in the future to get some ads going to offset the cost of the servers.” The first articles written about Facebook 20 years ago, when it was first created. 

🐜  World’s first transgenic ants unlock mysteries of insect social behaviour. 

🧥 The multi-decade work of animating Gogol’s “The Overcoat” by two Russian filmmakers. 

💊 Scientists have found evidence that obesity drugs could also tame inflammation. See my discussion with Dr Eric Topol on the evolution of temperance drugs.   

🖨️ Rapid 3D printing with liquid metal. 

🌪️ Hurricanes are becoming so intense that we might need a new way of rating them. 

---

Partner with Exponential View

Sponsoring Exponential View is the best way to reach a global audience of executives, founders, investors and policy-makers — while supporting your favourite newsletter. Get in touch to explore how we can work together.

---

End note

I know that Google launched Gemini on Thursday. People in the UK are meant to have access to Gemini Advanced (previously known as Ultra), but neither my Workspace nor Gmail account could access them. So I have no feedback to offer on it, yet.

Meanwhile, I put together a short musical experiment. Logarhythm Alpha is an ambient electronica set with an exponential decay structure. Each track is about 90% the length of the previous track, reflecting the time compression of the Exponential Age.

Enjoy,

Azeem

---

What you’re up to — community updates

• I will join EV member Greg Shove on 22 February for a virtual fireside chat. We’ll aim to answer: where is money in AI?

• Christina Nesheva joined the advisory board of the newly launched Appolica Venture studio.

• Richard Fallon has become the new chairman of Manchester Angels, an angel network. 

• Abishek Gupta writes about tackling anthropomorphisation in generative AI. 

• Vikram Mansharamani has published his new book The Making of a Generalist: An Independent Thinker Finds Unconventional Success in an Uncertain World.

Share your updates with EV readers by telling us what you’re up to here.

1. The study analysed sea sponges as a proxy. Sea sponges live about 300 years, and studying them provides a record that can reach further back than other sources of sea temperature data.