The Deload Podcast interviews CEOs, founders, and builders of frontier technology companies that are transforming how we live. The mission of the podcast expands that of The Deload: To educate you about where the world is going to make smarter growth investments. Now with conversation.
If you haven’t subscribed, join more than 1,500 rational optimists:
Artificial Brains for AI
Gordon Wilson is the CEO and co-founder of Rain. Rain is building artificial brains — chips inspired by the human brain and designed specifically for AI applications. The goal is to make AI hardware that’s as efficient as the human brain for intelligent application.
My firm, Deepwater, is an investor in Rain.
Here are the key notes from my conversation with Gordon:
An artificial brain is an efficient learning machine. The brain is about 1,000,000x cheaper and more efficient than current AI infrastructure. Power, size, data efficiency — the brain wins on every metric.
If we want machines to learn like humans, we have to make training cheap enough that it can go to low power footprints. We can’t build C-3PO with cloud infrastructure. An autonomous agent in the real world needs to be able to learn in real time, just like a human.
When you go to the edge, that’s a fundamental bottleneck to enhancement. We have level 5 autonomous cars in San Francisco, but they have a mini datacenter in every car. That’s orders of magnitude away from being economically viable as a taxi service.
People consider training and inference to be separate problems with separate infrastructure. We only separate these two problems because of the cost differences. Even if you only focus on inference, you have to have some ability to train like fine tuning and model maintenance. The consequence of integrating training and inference is what learning really looks like. It’s adaptation and continuous learning from the sparse signals in the real world.
Rain uses a radical co-design approach that considers hardware, algorithms, and software as an integrated problem. When you look at the brain, there isn’t a separation between hardware, algorithm, and software. They are fundamentally inseparable. Today in AI, NVIDIA builds a GPU, then someone else builds a neural network and figures out how to train it. We need to find new ways to intertwine hardware and algorithms.
Today, GPUs and CPUs separate memory and processing as they are built on the traditional Von Neumann architecture. Rain is building a compute-in-memory solution where you do your processing exactly where you store your data. That’s how a synapse in the brain works. Synapses are memory units and processing units.
Rain plans on serving the edge opportunity first, then the data center. The company’s first chip is targeting a 100x improvement in efficiency — 10x reduction in power, 10x improvement in speed. The chip will be in the market by end of next year.
Learn more about Rain and Gordon:
Episode Time Stamps
1:30 Rain’s mission to build an artificial brain
3:53 The cost of training and inference for ChatGPT
7:58 The importance of doing training and inference on a single chip
10:41 Rain’s radical co-design approach to chipmaking
19:37 We need AI cheaper in data centers and on the edge
21:30 Where we are today in Rain’s journey
24:50 Customer needs for edge AI solutions
26:43 Probability of delivering on 100x+ improvements
Disclaimer: My views here do not constitute investment advice. They are for educational purposes only. My firm, Deepwater Asset Management, may hold positions in securities I write about. See our full disclaimer.
Intelligent Indices Update: ChatGPT vs the S&P 500
As we move to a world powered by AI running on Rain’s Artificial Brains, every industry will be transformed, including the stock indexing industry. There hasn’t been a major new stock index since the mid-1980s when the Nasdaq 100 and Russell 2000 launched. The next one is coming, and it’s going to be powered by AI.
The Intelligent SP Select is that index.
The Intelligent SP Select is a US large cap stock index that uses ChatGPT, Bard, and Claude to re-envision the S&P 500. The AI index committee selects and weights a large cap representation of the market that aims to offer superior exposure to large cap stocks.
Given that the Intelligent Indices choose to be underweight big tech relative to their legacy comparison indices, the Intelligent Indices should extend their long-term outperformance if we continue to see a reversion to the mean in tech valuations vs the rest of the market.
The future of passive investing is intelligent.