Intelligence
Will we hit the marginal returns of intelligence?
Today’s issue feels hard. I really don’t feel inspired. Last week has been busy. Let me start from the one thing that has the greatest potential impact: intelligence.
🧠 Will we hit the marginal returns of intelligence?
Anthropic is among the most interesting companies in the world. Currently, I’d struggle to name someone else more relevant, more ambitious, or with a more impressive track record. If Anthropic is the one company you should be watching in right now, then this is the one post you should be reading this week to imagine our future.
“Machines of Loving Grace” is an exercise of rigorous optimistic reasoning by its CEO Dario Amodei. It’s well-balanced prose about How AI Could Transform the World for the Better. It’s main quality is clarity. Good execution can only follow limpid thinking.
The questions Dario is answering are pragmatic. From which fields is the acceleration most likely to come from? For which discoveries, the returns to intelligence are higher?
He delves into biology, neuroscience and economics to show us what we could expect five years after “Powerful AI” is reached. Dario offers an optimistic account that is not overly unrealistic. A good starting point for dreamers and doomers alike.
💸 An Esoteric Crypto Investment Memo
BitTensor is a decentralised network that coordinates incentives to build open-source machine-intelligence across a series of task-specific subnets (see a list on TaoPill).
BitTensor ($TAO) is borne out of over-regulation of intelligence. At scale, BitTensor has the potential to become the most liquid AI asset in the world, the highest-paying “employer” of AI talent in the world, the largest computing cluster in the world, and - eventually - the most powerful intelligence in the world. (Investment Memo by EV3, Aug 2024).
Every day, it distributes around $ 3 million (at current prices) to developers and operators as a reward solving specific tasks. Think protein folding simulations, or deepfake detection. Its mechanism is simple: a class of nodes, called validators, query and check the miners’ output against a work product definition set out by a subnet owner, who has “staked” (immobilised) substantial financial resources in pursuit of a specific subnet goal.
If it sounds vague, it’s because it’s meant to be. It's designed to be as general as possible. It acts as a meta layer to coordinate incentives. If in the world of human intelligence, we have capital, labour and regulations to coordinate large-scale collective action, here they have re-invented and codified the game.
It’s currently commanding a circulating $ 3.4 billion valuation ($ 10 bn fully-diluted). About a fourth of the last reported Anthropic funding round valuation (~ $ 40 bn). We believe it has ample room to grow. Staking yields between 10 - 15% APR (declining).
You may ask: where do its cash flows come from? They don’t. It’s more like a platform capturing the value orchestrated across buyers and sellers. It’s a marketplace for intelligence. It’s a captive ecosystem that makes possible for a. subnet owners to solve hard AI problems, for b. tech-savvy AI miners to compete for the solution rewards. In the middle sit validators who extract value from both sides, making the exchange trustless and uncensorable.
You could think it as a response to OpenAI and Anthropic from the unbound crypto crowds. It’s a far west that eschews politically correct regulations.
It’s an effort to radically democratise machine intelligence by pooling resources together very tightly. Its strength is allocating resources internally with the most stringent market mechanisms.
It’s difficult to imagine any single entity competing with that level of internal efficiency. Networks are the future of large-scale cooperation.
Will the seven levels of management corporation become a thing of the past before the end of the XXI century?
Until next week,
David