🔮 SpaceX vs Boeing; Telegram; AI Doom; battery crash; lab-grown chicken, social biome & papercuts ++ #489
An insider perspective on AI and exponential technologies
Hi, it’s Azeem. Welcome to our Sunday newsletter where I help you ask the right questions about technology and the near future. For more, see Top posts | Book | Follow me on Notes
Ideas of the week
A telegram about Telegram. Pavel Durov’s arrest earlier in the week exemplifies the exponential gap at play. As I wrote in my telegram about Telegram on Saturday, tech platforms are facing growing challenges operating in a global, borderless digital space while navigating the constraints of national laws and regulations. Platforms have to recognise their dual roles as communication utilities and public squares – a position that requires a greater level of responsibility.
Space oddity. SpaceX is a quintessential Exponential Age company. It has a focused ambition, iterates quickly and as a result, generates substantial learning effects. And now it’s coming to the rescue of two American astronauts stranded in space by Boeing’s space jalopy, the Starliner. This is a test of evolutionary fitness. Boeing epitomises the end of the late industrial era. It is burdened by bureaucracy; it relies on regulatory and political capture; and it has taken management theories like outsourcing past the extreme to the absurd. The firm has dispersed R&D and manufacturing across a global network of suppliers – a strategy that yielded quality issues and, at times, had fatal consequences. In contrast, SpaceX embodies the vertically integrated model of the Exponential Age, with in-house production that delivers rapid innovation and dramatic cost reductions in space launches. Since its founding in 2002, SpaceX has brought down the cost of space launches into low Earth orbit by 90%. Ten years ago, Boeing received a $4.2 billion contract from NASA to develop crewed launch vehicles with zero success. The same year, SpaceX got a $2.6 billion contract and has completed nine crewed launches so far. The tenth will be the rescue of the astronauts stranded by Boeing.
This evolutionary prowess extends beyond aerospace. In the automotive industry, we see Chinese multinational BYD applying vertical integration to produce electric vehicles at a fraction of the costs. In defence, Y Combinator just announced that it’s funding its first weapons startup on a mission to build low-cost cruise missiles and deliver on the needs of the US Department of Defense “in a form factor that’s 10x smaller and 10x cheaper”. It shows that SpaceX has opened the space for startups to change the paradigm in the bureaucratic and politicised practice of government technology procurement. For an exceptional essay on how mismanagement has destroyed American companies and why Silicon Valley seems to be the only place where firms can deliver on the promise of breakthrough improvements, I recommend Marko Jukic’s argument.
History repeats itself (hopefully). When Haber and Bosch first synthesised ammonia from air in the early 20th century, they changed history—creating fertiliser to support life, and explosives and chemical gas for the opposite. Getting there was intimidatingly difficult. The first plant in Oppau, Germany, produced 7,000 tonnes of ammonia per year. Today, global production exceeds 123 million tonnes annually – a 15,000 times the capacity of the Oppau facility. Cameron Roberts and Gregory Nemet, suggest in a new paper that this history could inform the scaling of Direct Air Capture1. Like ammonia synthesis, DAC is an energy-intensive process manipulating atmospheric gases. Currently, experimental facilities capture less than 10,000 tons of CO2 worldwide, but Roberts and Nemet argue that reaching gigaton scale by 2050 is possible if countries prioritise DAC as they once did with nitrogen synthesis for food and military needs. This would require strong government backing, reliable demand and national prioritisation. Those dismissing DAC’s potential should heed ammonia’s lesson: when the stakes are high, human ingenuity can scale mountains – or capture atmospheres.
The evolving Rumsfeld matrix of AI. AI is creating new frontiers for us in real time. An AI diffusion model can now simulate the first-person shooter game Doom without pre-programmed game mechanics – real-time, at over 20 frames per second. It wasn’t known this was possible until now. In Rumsfeld terms, it was a “known unknown”. Normally, game developers would define the game rules, physics and assets upfront, in order to establish how everything in the game behaves. However, in this case, a model created a playable game just by presenting each frame one after another sequentially. It paves the way for AI-assisted game design, where developers could prototype complex games through ordinary language. More profoundly, it hints at a future of infinitely variable, emergent game worlds that adapt in real time to player actions. But, this isn’t just about games. It could lead to a new class of AI-driven interactive environments to revolutionise training simulations, virtual prototyping and how we design and interact with complex software systems.
Data
Battery prices have fallen by more than half in just 18 months.
Lab-grown chicken’s estimated cost of $6.2 per pound is approaching that of its embryonic counterpart at $4.1 per pound
As of late 2017, Google’s ad technology could see or access 84% of the global web display ad market, excluding China.
In Q2, China became ASML's top market, accounting for 49% of lithography system sales.The Dutch government is planning on preventing ASML from maintaining its deep ultraviolet lithography machines in China
The new Cerebras AI chip can run Llama 3.1 8B at 1700 token/s, or about 400 times faster than you can read.
AI model LTM-2-Mini has a 100 million token context window – 10 million lines of code or 750 novels. Heading towards Eric Schmidt’s notion of the infinite context window.
75% of the latest Y Combinator start-up cohort are working on AI products. And OpenAI has been valued at over $100 billion.
Short morsels to appear smart at dinner parties
👀 Nearly half of all AI safety researchers at OpenAI have left the company.
🐤 Every country in the world has a lower fertility rate now than in 1950.
❤️🔥 Our social circles may have influence on our gut microbiome.
🥸 North Korea registered 13 submarines with the International Maritime Organisation for the first time.
☀️ Agri-PV FTW. Putting solar panels over berry farms increases berry production instead of harming it.
😝 The history of coder vs. programmer dichotomy shows that it’s a myth.
☝️ Even small increases in family income through cash transfers might help reduce some forms of child abuse and neglect.
🌎 NASA is building the first telescope designed for planetary defence.
📰 You didn’t know you needed to know this, but here it goes: paper types ranked by likelihood of paper cuts.
In focus
This week, I wrote an essay revealing why I changed my mind on self-driving cars. I was previously pessimistic about the timelines and positive impact AVs would have on urban transport. But with growing new evidence, I’m changing my position.
DAC uses solvents or sorbents to capture carbon dioxide from the atmosphere.
Here’s a few links to an award winning short film we made about Haber back in 2008. Enjoy!
https://en.m.wikipedia.org/wiki/Haber_(film)
https://youtu.be/znXhaRnQ26I
I'm stunned by AI generated and playable version of Doom without any pre-programmed rules. This is insane! It's like AI is learning to create entire worlds on its own. Imagine the possibilities for gaming, simulations, and even software design. What do you think this means for the future?