🔮 AI’s education lesson; virtual power plants; open-source; brains, the microbiome, Snoop Dogg++ #421
Your insider's guide to AI and exponential technologies
Hi, I’m Azeem Azhar. As an expert on AI and 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 insider view on developments that I think you should know about.
In today’s edition:
LLMs shake education firms;
Automation revives the age-old battle between labour and capital;
The power of virtual powerplants.
📢 We’re looking for cool products to share with our audience. If you’ve got a product or service you think we’d love complete this form and we’ll get in touch to discuss collaboration opportunities.
Sunday chart: Easy as AI, b, c
This week, shares in EdTech firm Chegg plummeted by over 40% as students turned to ChatGPT rather than signing up for the firm’s offerings. Other educational players saw their shares drop in sympathy.
Chegg’s difficulties mask the positive story: the possibility of widening access to education globally. More than a quarter-of-a-billion children are not in primary or secondary education worldwide. The cost of education resources and access to teachers are two of the barriers. But the prospect that anyone with a phone could access a personalised tutor could be good news.
Generative AI models, like ChatGPT, show immense promise as learning assistants, offering students personalised tutoring experiences. The next step involves endowing these learning assistants with memory and fine-tuning them to focus on facilitating learning rather than simply repeating answers. Khan Academy’s AI assistant demo provides an example of what this enhanced technology might look like.
Weekly commentary: The future of open-source AI
All software markets are a balance between open- and closed-source. Where will AI end up?
It all started with leaked Google documents shared by an anonymous insider to SemiAnalysis who claims that open-source large language models pose a significant challenge to Google, as well as to OpenAI. The insider argues that neither Google nor OpenAI has an economic barrier, or “moat,” that allows them to generate sustained long-term value.
In the last few days, I’ve spoken to Emad Mostaque from StabilityAI, Yann LeCun from Meta, and several other people who are developing these things. In addition, I’ve been thinking about open-source and public goods for more than two decades. Here’s my take, open for paying members.
🐙 The open-source future of artificial intelligence
Watch now (12 min) | In today’s commentary I delve into the question of open-source versus closed-source models for AI, and how this will shape the future of the internet. ⭐️ As usual, I recommend watching the the video! ⭐️ There is more in it than the essay alone. It all started with leaked Google documents shared by an anonymous insider to
Key reads
In this multi-layered reflection on automation and labour, my friend Hari Kunzru argues that midcentury visions of automation as a way to free people from drudgery now seem quaint and even sinister.
[T]he sudden explosion of generative AI models [...] appears to be a new incursion into the eroding territory of the human. Until recently, we’ve assumed that only low-status tasks would be assigned to machines. The automation of creative work has revealed to a whole new class of people that they are not immune to Schumpeterian creative destruction.
(See also: The Writers Guild of America is on strike. It is a case of labour against, not so much technology, as capital. Streaming, and chatbots, change the balance of power between workers and firms. Expect more. Also, Ted Chiang in the New Yorker: could AI serve to accelerate efficiencies within firms, favouring businesses over employees? Does that have to be the case?)
Key read for founders this week: Investor Gordon Ritter suggests that specialised software with deep data and narrow context will create the most value in the generative AI era. Startups should focus on building in specific domains with narrow context or choose to major in generative AI’s emerging technical capabilities and hunt for a function or vertical problem that benefits from their insights.
Virtual powerplants, where one strings together the battery storage capacity in thousands of EVs to create medium-scale electricity storage, turn out to be wonderful. According to Brattle Group, the annual cost of providing 400MW of adequacy runs to $43m with gas, $29m with traditional battery farms; and only $2m with a virtual powerplant (VPP). As you all know, I write about VPPs in my book. The insight here is that our electric vehicles store 60-80kWh, while American homes need 10kWh a day. The home can easily “borrow” when the sun isn’t shining. Aggregating across thousands of cars and homes allows for a more nuanced distributed storage system. Anyhow, it’s going to be huge. (h/t: Jigar Shah.)
Market data
According to Russian government figures, 10% of their tech workforce left Russia in 2022 (likely an underestimate).
70% of technical innovations can be attributed, either directly or indirectly, to advanced materials.
IBM’s CEO states that 30% of the company’s non-customer-facing roles could be replaced by AI and automation over five years.
Compared to the last decade, 22% more AI researchers believe progress will stem from better algorithms (totalling 53%), while 19% fewer strongly agree it will result from increased compute (totalling 40%).
Only 12% of SDH goals are on track to meet the 2030 deadline. See also, we’re in the decade when renewables sales race up the S curve via EV member Nat Bullard.
OpenAI’s revenues will exceed $200m this year, up from $28m last year.
Short morsels to appear smart at dinner parties
🤖 🎶 Snoop Dogg on AI existential risk. And Grimes pioneers a new model for artists working with AI.
🧠 A new study captures ‘conscious-like’ brain activity in comatose patients during the dying process, providing new insights into death. And on the other spectrum of life, the first-ever brain surgery has been performed on a foetus.
💉 A contraceptive vaccine could revolutionise birth control by preventing pregnancy without hormonal side effects.
🪙 The Sleeping Beauty Problem has been dividing mathematicians since 2000.
End note
Could I have chosen better lyrics for the headline and graph this week? I would love to know. Put your suggestions in the comments…
In the meantime, we are running some of my all-time favourite episodes of the Exponential View podcast again. This week, my picks include my 2020 conversation with Sam Altman, and my conversation with Demis Hassabis from the same year.
And finally, our team is hosting an AI Praxis session for annual members of Exponential View. This will be an opportunity for our community to showcase how you utilise AI tools and prompting, and for those who haven’t run experiments yet - to learn. We are looking for members who want to share their experiments with others in this session — if you’re one of them, fill out this form.
A
What you’re up to — community updates
Rumman Chowdhury is launching the largest-ever public red-teaming exercise involving the major LLM providers. The White House has got behind it too.
The short film, Diwali, starring Abhinav Chugh, won a prize at the Printemps Carougeois short-film competition.
Sam Butler-Sloss has a fantastic presentation on the energy transition.
Ramsay Brown was interviewed for The Monocle Daily on Joe Biden’s meeting with AI CEOs.
Gianni Giacomelli on GPTs and business process industrialisation.
Rafael Kaufmann was on the ReFi DAO podcast to talk about his startup Digital Gaia’s work building an open network of AI copilots for regeneration.
Share your updates with EV readers by telling us what you’re up to here.