🔮 Pessimism; the AI reshuffle; synthesis setback; Microsoft eats Apple, Tolkien ++ #456
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 developments that I think you should know about in this newsletter.
Latest posts
If you’re not a subscriber, here’s what you missed recently:
I published the third part of my 2024 Horizon Scan to members this past week. Fitting for a Grand Finale, I cover some of the biggest questions on the horizon yet:
2024 elections and the epistemic quake (don’t blame AI),
The society of AI: From monocognitism to syncognitism,
Helpful humanoids: As humanoid robots improve, what to look out for,
Moving at human speed.
💥 Thanks to our sponsor Intercom, the complete customer service platform.
Sunday chart: Cautious New World
Economic sentiment in US media has undergone a decline since the 1970s — predicting economic performance. In a separate analysis, the FT found a declining trend in words relating to progress and future, and an increasing trend in words relating to caution, worry and risk in Western books. It seems that the West is becoming more negative and conservative in outlook.
These trends are of great concern. As John Burn-Murdoch puts it,
Language — particularly in books — doesn’t just describe the world as it is, it describes the world as it could be. Writing about how to progress to a better future can make that better future more likely. Writing about worries can create a worried world.
The increased usage of words relating to progress and the future started during the Enlightenment. There was an optimistic outlook, spurring groundbreaking advancements. Indeed, this cultural shift towards progress preceded economic acceleration. Should our culture’s progressive trajectory reverse, we are likely to encounter heightened resistance to change. Consequently, existing societal challenges, such as the reliance on fossil fuels, the rise of populism, and NIMBYism, might become further entrenched.
To fight this change, we need to deal with the underlying causes. They are not obvious. Perhaps it’s bad-new bias or the great stagnation. These are likely factors. But I think there might be a more pervasive cause. Over the same period that pessimism increased, so did social liberalism. While individual rights are crucial, the byproduct is a shifting burden of responsibility from society to the individual. In the exponential age, more individuals will struggle to keep up with the pace of change, resulting in their pessimism about the future. How might we address that?
Of course, this is just conjecture and is only one of many potential factors. I would love to hear your thoughts on these trends in the comments.
💥 Today’s edition is supported by Intercom.
Intercom is a complete, AI and human customer service platform—and the onlyplatform to combine a help desk, AI chatbot, and proactive support tools.
It also comes equipped with a fully featured inbox, ticketing, help centre, and messenger, making it not only complete—but also incredibly powerful.
We’ve got every single tool you need, all in one place, enhanced with AI, so you can give every single customer the fastest, most personalised experience imaginable.
Key reads
‘Offboarded’. The past year involved a flood of corporate jargon as over 260,000 employees were laid off at tech companies, a 60% jump from 2022. A lot of this was due to structural changes and economic reasons. But it’s clear that AI is being used in some cases. This week, Google laid off hundreds of workers across several divisions. According to The Information, the cuts in the Google ad sales division can be attributed to AI automating parts of the ad sales process. Duolingo also “offboarded” 10% of their contractors this week, citing efficiencies gained from AI-generated content as a factor. It appears the labour force disruption of AI has started. Unfortunately, technological disruption precedes employment creation in technological transitions. It might take a while before new roles enter the economy.
See also:
A Nature study found that using Google Search to evaluate misinformation news stories made participants trust those stories more.
Fox Corp has launched a new blockchain system to help media companies negotiate with AI model makers.
A humanoid robot learns to make coffee by watching people do it. (Although our resident Australian
notes, pod coffee isn’t real coffee, baristas are safe for now).An analysis of the dangers of the GPT builder revenue program. Hint: its incentive structure is similar to social media.
A false start. A few weeks ago, DeepMind released GNoME, which, according to their claims, identified 380,000 new stable materials. To synthesise these new materials, they enlisted the help of A-labs, an autonomous synthesis system, which helped synthesise 41 novel compounds. Or so they thought. According to a new preprint study analysing these synthesised compounds, none of the compounds were new materials. A significant issue was that GNoME incorrectly predicted new materials with neatly arranged atoms. In reality, when these materials were made in the lab, they turned out to be already known materials with atoms in a more random arrangement. The study recommends that future models like GNoME include the possibility of atoms being randomly arranged, known as compositional disorder, to better identify truly novel materials. This is a bump in the road for finding new materials, but the field is still making strong progress. In the end, this is just science at work. The feedback and ideas in this paper will improve future models.
See also:
Microsoft has discovered a material that may be able to produce batteries with 70% less lithium (assuming it is experimentally proven.)
The silicon swamp. When people think of failure in tech, they often think of Europe. Yet as The Economist highlights, this is a slight misperception, given the existence of ASML, the only manufacturer of extreme ultraviolet lithography equipment, an indispensable part of the semiconductor supply chain. ASML’s success is attributed to its extensive network of suppliers and technology partners, forming a European counterpart to Silicon Valley. It could provide an operating model, à l’Européenne, to be emulated by other companies and encouraged by governments.
Optimisation breakthrough. If Andrew Ng gives a standing ovation, you know it’s a big deal. Andrew praised the Direct Preference Optimization (DPO) paper by Rafael Rafailov et al. as a notable advancement in language model training, offering a simpler, more efficient alternative to traditional reinforcement learning from human feedback (RLHF). DPO innovatively combines the reward function and language model into one framework, streamlining the alignment of models with human preferences. However, this approach doesn’t overcome all risks in the model. As Lee et al. highlight, DPO can reduce toxic outputs, but they don’t remove underlying toxic capabilities, making it possible to revert the model back to its original, potentially harmful behaviour. Alignment still has a way to go.
See also: a recent paper by Anthropic describes how they were able to build agents capable of deception based on training LLMs of specific types of data. Worse, these agents eluded safety training.
Market data
It is estimated that in 2024, China’s public electric vehicle chargers will use more electricity than the entirety of Greece.
Natural disasters cost the world $250 billion last year.
First the rave caves, now the forests of Central Africa. The psychedelic Ibogaine was found to reduce symptoms of traumatic brain injury (such as PTSD and depression) by more than 80% on average.
A survey of 648 sales professionals found that they saved 2 hours and 15 minutes a day on average by using AI for tasks.
Microsoft overtakes Apple as the world’s most valuable company.
Trading firm Jane Street makes $5 million in revenue per employee. For comparison, Goldman Sachs makes a relatively minuscule $977k per worker.
Short morsels to appear smart at dinner parties
❗ A theory of the impact of gender norms on the use of exclamation points. via
💍 The complexities and demographics of Tolkien’s Middle-earth. h/t EV reader Catherine Lückhoff
🥑 The perfect webpage: how the web underwent a transformation to cater to Google's search algorithms.
🤖 By age four, kids begin to prioritise the accuracy of information over whether it's delivered by a human or robot.
💥 The four “AI stack wars” via
🫀 First-of-its-kind heart transplant on a child allows the organ to grow with him.
End note
I gave a lightning talk at the DLD Conference in Munich on Thursday, covering the past, present and future of the exponential transition. You can watch it here.
Reminder, I’ll be in Davos participating in the Annual Meeting of the World Economic Forum. Do try me again if you are around, although the calendar is quite tight now.
Cheers,
A
What you’re up to — community updates
Congrats to
on publishing her first book, Not the End of the World!Ludwig Siegele is taking charge of The Economist’s AI initiatives.
Rafael Kaufmann co-authored a Climate Collective report on AI for nature and climate action.
- reflects on the past year and poses 12 questions exploring the future of democracy.
- delves into the intricacies of human intelligence.
Rodolfo Rosini’s company, Vaire Computing has been listed as “one to watch” by Sifted in their 2024 semiconductors briefing.
Jeff Jarvis reflects on his testimony in the Senate about AI and journalism.
- highlights the best books he read in 2023.
Share your updates with EV readers by telling us what you’re up to here.
First two helpful ideas that come to my mind: (1) balancing the individualism of social liberalism with an adjustment toward collectivism (ideally beyond anthropocentrism) and (2) focusing less on optimism about progress for society first and more on optimism about movements that would shift power away from the extreme inequality that makes pessimism a logical response.
There’s definitely future shock happening, generating its attendant fear of being able to cope with more. The cure is, I believe, simple in theory and incredibly difficult in practice. If populations trusted that their elected representatives would ensure everyone would get a fair shake, whatever the impacts, fear would dampen considerably.