🔮AI’s world models, Europe’s risky four, BYD vs Tesla, murderous students & vegetarianism++ #443
Your insider guide to 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.
Latest posts
If you’re not a subscriber, here’s what you missed recently:
The cow and the academics: How to interpret disagreements about generative AI,
The promise of open-source generative AI, with Stability AI’s Emad Mostaque.
🎨 Thanks to our sponsor, Masterworks, an investing platform that enables everyday people to invest in multimillion dollar paintings by artists like Banksy and Basquiat.
Sunday chart: World models?
How does an LLM work? Simple answer: it predicts the next token according to an input, based on its training and reinforcement learning. But is there more to it than that? Somewhere along this learning process, could the LLM be building some kind of understanding of the world, a “world model”? This concept of a world model is currently the subject of hot debate across the AI community. (I talked about the many disagreements in generative AI and how to deal with them in Friday’s essay.)
The latest salvo comes from MIT researchers Wes Gurnee and Max Tegmark, who recently published a paper arguing that
modern LLMs acquire structured knowledge about fundamental dimensions such as space and time, supporting the view that they learn not merely superficial statistics, but literal world models.
LLMs, the authors argue, have a world view because they appear to have a structured notion of space. There is a constant, stable structure of relationships underneath the statistical relationships.
But is this proof of a world model? And how could we tell? First, it depends on how you define a world model. The idea that intelligent systems (animals included) must have an internal model of how the world works is not a recent one. Nvidia’s Jim Fan offered a good summary of how our understanding of world models in AI has developed over recent years, one key aspect being that they capture causality and intuitive physics. An LLM would have to show more than predictive capacity, an actual causal understanding and ability to simulate the physical world. However, an understanding of, say, geography wouldn’t support an understanding of the physical laws and causality of the world.
explains that this appearance of having a structure to the knowledge doesn’t mean you have an understanding of the knowledge sufficient for it to be called a “model”. In fact, he shows that much more primitive AIs contain correlations about physical space.A lot of the internal machinations of our LLM helpers remain a mystery, but this kind of research, and most importantly the discussions it provokes, brings us closer to understanding them.
See also: A study has found that LLMs provide useful feedback on research manuscripts, in some cases better than human feedback. Plus, the first wearable AI device was unveiled by Humane at the Coperni Paris fashion show.
🎨Today’s edition is supported by Masterworks.
Monet’s painting sold for $8 million… and everyday investors profited
When the painting by master Claude Monet (you may have heard of him) was bought for $6.8 million and sold for a cool $8 million just 631 days later, investors in shares of the offering received their share of the net proceeds.
All thanks to Masterworks, the award-winning platform for investing in blue-chip art. To date, every one of its 16 sales out of its portfolio has returned a profit to investors. With three recent sales, Masterworks investors realised net annualised returns of 17.6%, 21.5% and 35%.
Shares of every offering are limited, but Exponential View readers can skip the waitlist with this exclusive link.
Key reads
Eleven seconds of a fatal Tesla crash. An exceptional piece of investigative reporting pulls apart the 11 seconds of a fatal Tesla crash. It shows how many factors need to be controlled for it to realise the promise of self-driving. In this case, the driver set the car at 25% above the speed limit; a truck edged slightly past the junction and the vehicle itself did not see the truck (!). Tesla’s autopilot is, according to its lawyer, “just fancy cruise control.” Progress made by Britain’s Wayve is going under-reported — they are generating fully synthetic worlds, a “world model”, it says, for the vehicles to train on.
Europe’s risky 4. The European Commission declared the following four technologies as “highly likely to present the most sensitive and immediate risks related to technology security and technology leakage”: (1) advanced semiconductor technologies, (2) AI technologies, (3) quantum technologies, and (4) biotechnologies. Many states recognise their dual-use nature: there’s an imperative to figure out how to capitalise on these technologies—and Europe largely does so horribly. As Peter Thiel points out, of the seventeen $100bn corporations founded since 1990, eleven were in the U.S. and 6 in China. A risk-first mindset doesn’t leave one in control of the outcomes of the technologies as they scale up. The wording of the European Commission’s communiqué simply doesn’t grasp that, positioning Europe likely as a technology taker once more.
See also: The U.S. National Security Agency is starting an AI security centre.
Meta’s money cow. Meta is expanding its AI tools to all advertisers, now helping all marketers create images and texts faster. The company ad business is working well — contrary to X. Meta’s digital ad revenue increased 12% in Q2, which drove its share price to an 18-month high. Using generative AI for digital marketing looks like one of the fastest, easiest ways to increase profits from the technology: the faster, and therefore cheaper, it is to make an ad, the more ads can be rolled out, the more sales for the company. Happy advertisers, happy platform.
I would recommend reading Maria Vesserman’s article on how to use AI throughout all of the activities of a SaaS company. It’s great inspiration, and shows how deeply AI could be infused in business processes - and what it’s not so good at.
EV battle royale. As we wrote in our monthly retrospective, the global EV market is heating up. In Q3 2023, Chinese car maker BYD sold 431,603 fully electric vehicles, up 23% from Q2, and only about 3000 cars less than Tesla. The gap had never been so small. Tesla will need to remain agile to keep its lead. BYD established its dominance in the Chinese market by selling affordable EVs, which appealed to a wide base of customers. Earlier this year, the company added two luxury cars to its offering, breaking into the $100,000+ market. It’s wonderful to see how the formerly unpopular, clunky and un-sexy EVs are now becoming very popular, although a Chinese dominance in yet another market that might be cause for concern.
Market data
The price of Chinese-made LFP and NMC lithium-ion batteries has dropped below $100 per kilowatt-hour, a key threshold for affordability.
Around 80% of Bloomberg’s clients predict that the oil and gas industry will outperform other energy sectors over the next 12 to 24 months.
On September 30th, rooftop solar panels provided enough electricity to meet all of South Australia’s demand during the early afternoon, peaking at 101%, with total renewable energy generation reaching 114%. This surpasses the previous record of 99.2% set just a week earlier,
Mental health issues cost UK businesses £6.9 billion in lost productivity due to long-term illness in the year to August.
Short morsels to appear smart at dinner parties
🐱 Cats brainlessly purr, quite literally.
🧶 The online knitting community revolts against knitting.com (and capitalism).
🪐 Remote places on Earth serve to simulate life on other planets.
🗡️ Oxford was the mediaeval homicide capital of England because of… its murderous students.
🏗️ Recreating the world’s most useful petrochemical without fossil fuels.
🥗 The likelihood of being a vegetarian may be influenced by genes.
Community events
🇨🇭 We are holding a meet-up for Exponential View members in Geneva next week, on Tue, 10 October 2023. Register here.
😎 We are hosting another AI in Praxis event on 26 October to share the best practices for using AI in daily work and life. The event is open only for members of EV with an annual subscription. If you would like to present your use case on the day, please fill out this form.
End note
I’m sharing a few things I’ve read/am reading right now.
I finished reading Tolstoy’s A Confession, which is quite the most remarkable, personal and introspective work. It’s a surprisingly honest book, quite direct, perhaps not as layered as Tolstoy’s famous tomes. But I strongly recommend it.
In keeping with the Russian lit scene, I’m also reading The Master and Margarita. Also recommended, not least because it is the inspiration for one of my favourite tracks, Sympathy for the Devil.
On the non-fiction side, James Ashton’s The Everything Blueprint on the ARM architecture is a very fun, journalistic history of this key chip architecture.
Finally, I put together an upbeat ambient playlist which runs for about five hours. You can pick it up from Tidal or Spotify. Enjoy!
Cheers,
A
What you’re up to — community updates
Chloé Ipert published a paper on how to classify blockchain ventures.
Rumman Chowdhury discusses truth and AI in an episode of AI IRL, a Bloomberg Original.
Zoltan Patai and his team at Flawless, an operations observability startup, raised $2.2M in seed funding.
Felix Zeltner discussed AI, the future of work, Europe and the next Internet with Amy Webb.
- and his team at Deep Forest Sciences published an introductory post about building no-code AI platform Chiron for small molecule drug discovery.
Share your updates with EV readers by telling us what you’re up to here.