Discover more from Exponential View by Azeem Azhar
🔮 Time-space compression; Falcon soars; rebel microchips; skincare for mummies ++ #439
Your insider guide to AI and exponential technologies
Our new TV series is out. In the first episode, I speak with the Prime Minister of Estonia at length about her country’s post-independence journey, building resiliency through technology, and nurturing an entrepreneurial spirit for a global market. Watch on YouTube, and when you do — hit the like button. If you are in the US, the show airs live on the Bloomberg TV channel at 10pm EST on Wednesdays.
🤳 Thanks to our sponsor Try It on AI. Get Business Executive AI Headshots for the price of two lattes.
Sunday chart: Pressed for time
In his 1989 book, The Condition of Postmodernity, David Harvey introduced the idea of “time-space compression” to describe how advancing technology, communication, and globalisation would result in tangible changes in our societies. In turn, this would have subjective psychological effects of disorientation and alienation, amongst others. Future shock, in the terminology of Heidi and Alvin Toffler. Harvey and the Tofflers mirror some of my arguments in Exponential, although I emphasised the institutional framing of the issue.
The past six months have felt like a study in time-space compression, further and faster than the year before or the year before that. This is what we should expect in the Exponential Age. Let’s look at some of the most recent indicators.
Open-source LLMs are now only a few months behind proprietary ones as measured by benchmarks (which themselves have their limits). The UAE-based Technology Innovation Institute has just this week launched Falcon 180B, the most powerful open-source LLM to date. Equally, Huawei has been able to produce a powerful, modern smartphone with advanced connectivity despite the robust action taken against it by America and the enforcement of strong technology controls towards China.
And it’s only going to continue. As the McKinsey analysis shows, estimates for the time until software systems can reach human-level performance have fallen dramatically. Over the past year and a quarter, the consensus, as evaluated on Metacalculus, for the first weakly general AI has come forward from 2042 to 2027.
As AI systems get better, expect even more time-space compression. The dynamics today are driven by familiar themes: the sharing of method and practice; global access to data, compute, and talent; and the traditional jostling of commercial and strategic competition. The systems being developed are still tools, yet to be turned systematically towards their own improvement.
Dealing with this compression is a big, complex challenge. We’re moving quickly but late to tackle what we might consider the institutional scaffolding for a new technology. EV reader, Ian Hogarth, is doing impressive work in assembling experts and framing challenges for the UK’s foundation model task force. (See community notes below.)
These are the mechanics of our system: just one aspect of what Harvey and the Tofflers explored (and the one I emphasised in my book). Another lens considers our subjective psychological response to this process (and it will be a persistent process, not a singular destination). Toffler suggested we’d need “islands of stability”, like familiar spaces, rituals or values. Harvey proposed, inter alia, the importance of local attachments and finding some way of accepting the ephemerality of the process.
My sense is that the frameworks, tools or ways of being we need to psychologically deal with this environment also require the kind of global and national initiatives that the technocratic safety engineering calls for. There don’t seem to be many structured efforts on what I suspect is an important piece of work. Psychologies that are driven by fear, instability and disorientation do not make for happy, stable societies.
What do you think? Comments below are welcome.
Phoney restrictions. The recent release of Huawei’s Mate 60 Pro smartphone featuring a 7-nanometer Kirin 9000s chip, fabricated by China’s top chipmaker SMIC, raises sharp questions about the efficacy of U.S. chip sanctions. The industry in China seems to be advancing fast despite the restrictions. Both the Kirin chip and the 5G RF chips would “mark a major milestone and breakthrough for China’s semiconductor industry, a disruptive move on the global stage,” according to one analysis.
As I argued in this week’s video essay, China’s rise as a technological powerhouse, coupled with its involvement in international institutions, could infuse global governance mechanisms with its domestic values. How the rest of the world responds to the country’s emerging power is not straightforward: as the Huawei case shows, diplomatically acceptable barricades are leaky. And it can create collateral damage: Qualcomm, the American company which used to sell chips to Huawei stands to lose substantially because of the new Kirin 9000s chip.
Foreign or made of bits? AI detectors frequently misidentify writing by international students as AI-generated, revealing a systemic bias against non-native English speakers. A study by Liang et al. found that across seven AI detection tools, human-written text from a native speaker was incorrectly detected around 5% of the time, against 61.3% for non-native speakers. The detectors are often designed to flag text with predictable word choices and simpler sentences, characteristics commonly found in the writing of non-native speakers. While there are emerging ideas for watermarking AI-generated text and images to ensure trust, these methods are better at confirming AI authorship than disproving it. The question is, can we govern AI and protect trust online without the ability to reliably tell AI-generated material from human compositions?
See also: Researchers successfully used AI feedback to align LLMs to human preferences in an effort to solve the scalability problem of reinforcement learning from human feedback (RLHF).
A tale of two CEOs. Time magazine recently published its list of the 100 most influential people in AI in 2023. Two of them, Sam Altman and Elon Musk, were defining characters in the founding of OpenAI in 2015. The two men have now parted ways because of disagreements over the direction the company was taking. Nevertheless, their philosophy remains a driving force in Silicon Valley: if you’re worried about a technology, do your best to create it so that you can somehow inject your view of safety into its design.
See also: a profile of Meta’s Yann Lecun explores why he doesn’t believe the “doomer narrative.”
We published two posts for our Premium members this week:
I delve into how to govern technology in an increasingly fragmented economic environment. As is probably clear from this week’s newsletter, these questions have been on my mind a lot.’s Promptpack showcases how to use ChatGPT to analyse your marketing funnel. It is very practical.
Today’s edition is supported by Try it on AI: Get Business Executive AI Headshots from the comfort of your home.
Exponential View MD,, decided to test Try it on AI herself. She submitted fresh selfies, selected the Business Studio style and within an hour had over 100 portraits to choose from. Marija picked the Premium Package, which includes 2 human edits that ensure the final portraits selected are perfect.
Exponential View readers get 15% off using EVAI15 on checkout for all packages. Code valid for the next 48 hours only, so don’t miss your home studio shoot.
Brussels city centre has managed to reduce car traffic by 25% while increasing bicycle traffic by 36% compared to last year.
In August, battery electric vehicles accounted for 20.1% of the market share for new vehicle sales in the UK, up from 14.5% during the same period last year.
Since the beginning of the year, Tesla has significantly reduced the prices of its Model S and Model X, by 29% and 34% respectively, bringing them down to $74,990 and $79,990. This adjustment has made the Model X eligible for the US IRA subsidy.
Denmark’s economy grew 1.7%, on the strength of demand for GLP-1 agonists. If you exclude its pharma sector, the economy contracted. Weight loss drugs contributed to the ballooning economy. How’s that for poetic?
Short morsels to appear smart at dinner parties
👃 AI can predict smells from molecular structures.
🐣 Scientists have reproduced a model of a human embryo from stem cells.
🥼 GLP-1 agonist, semaglutide, appears to be able to reverse type-1 diabetes in certain patients.
©️Microsoft announced the Copilot Copyright Commitment, the assurance that customers could use the output of Copilot AI without fear of copyright infringement.
🌚 Welcome to the new moon race.
🚗 BMW is getting help from China for its first electric Mini Cooper platform.
🏺 You may want to opt out of ancient Egyptian mummy skincare.
The summer is over for those of us in the Northern Hemisphere and now work begins again. What better time to tap into Exponential View to understand how the coming months and years will affect your life and work!
If you love EV already, take a moment to share our work with your network — if you use the referral code below, you will also accrue points towards rewards. Win-win!
P.S. Our team will be in Berlin for the Falling Walls Science Summit in early November, one of our favourite annual technology and science events. Exponential View readers get an exclusive 20% off on tickets with code exp_view23. Register here.
What you’re up to — community updates
At least six readers of EV were named among the 100 most influential individuals in AI 2023 by TIME. Congrats!
Ian Hogarth is making great progress as he leads the UK’s Frontier AI Taskforce.
Robbie Stamp and Chanuki Seresinhe are giving a talk named “The AI Goosebumps Club: Thought experiments about humans and artificial intelligence” at CogX in London on Sept 12th, 16:00-16:45, Magazine building.
Michael Liebreich wrote about how getting to net zero will be harder — and easier — than we think.
François Candelon warns that we should be realistic about generative AI’s economic impact.
Share your updates with EV readers by telling us what you’re up to here.