š® The Sunday edition #507: The intelligence explosion; TikTok & chips; immortality & new religions ++
An insider's view on AI and exponential technologies
Hi, Azeem here.
āIt just keeps going up... and up... and upā. Thatās how famed AI blogger Gwern describes the remarkable acceleration in AI capabilities. Todayās edition examines this potential intelligence explosion alongside the UKās ambitious new AI strategy ā a 10-year plan that could redefine how nations approach AI. Itās the best government plan Iāve seen so far.
This and more to get into in todayās weekend edition of Exponential View!
Are we on the edge of an intelligence explosion?
In my essay earlier this month, I argued that we should take Sam Altman and other industry insidersā claims about imminent, transformative AI capabilities by 2025 quite seriously. The evidence is there in massive infrastructure investments and in expertsā increasingly compressed timelines. AI analyst Gwern highlights a powerful data flywheel effect taking shape, where each new model spawns data for training an even better one:
Every problem that an o1 solves is now a training data point for an o3 (eg. any o1 session which finally stumbles into the right answer can be refined to drop the dead ends and produce a clean transcript to train a more refined intuition).
He also notes that:Ā
If youāre wondering why OAers are suddenly weirdly, almost euphorically, optimistic on Twitter, watching the improvement from the original 4o model to o3 (and wherever it is now!) may be why. Itās like watching the AlphaGo Elo curves: it just keeps going upā¦ and upā¦ and upā¦
There may be a sense that theyāve ābroken outā, and have finally crossed the last threshold of criticality, from merely cutting-edge AI work which everyone else will replicate in a few years, to takeoff ā cracked intelligence to the point of being recursively self-improving and where o4 or o5 will be able to automate AI R&D and finish off the rest.
So what to make of this? Unlike Gwern, I wouldnāt use the term ābroken outā. It is too laden with allusions to the containment problem. Instead, I see AI as a tool that can improve the creation of the next generation of AI tools ā possibly to a point where the self-reinforcing flywheel truly speeds up.
God save the MachineĀ
We now have a high-level plan for AI by the UK ā one that, rather than merely fearing the technology (EU) or fixating on Beijing (US), seeks to shape its trajectory. The UK government is placing a large bet on āenabling and embracingā AI, with a decade-long roadmap for infrastructure. This includes a proposed 20-fold expansion of state-backed computing power and āAI Growth Zonesā to anchor new data centres. Just as significant is the National Data Libraryās pledge to release at least five āhigh-impactā datasets to spur AI development.Ā
There are many more excellent initiatives, including making it much easier to build both the electricity systems and the infrastructure required for data centres and doubling down on industrial centres of excellence for this new technology.Ā
The plan, put together by Matt Clifford, a long-time Exponential View friend, is a comprehensive statement of intent about the importance of industrial renewal and the attention a new general-purpose technology requires to prepare society for its deployment. Of course, the benefits will depend on the exact nature of implementation, but as far as government plans go ā this is the best Iāve seen so far!
A two-front confrontation
In the past few days, Americaās technological and cultural stand-off with China has matured. Weāre amid another milestone in techno-geopolitical fragmentation. The tensions have been apparent for years but the speed with which theyāre now playing out is accelerating.