“The map is not the territory” – Alfred Korzybski
One thing that every scientist gets taught is that the map isn’t the territory. This means that the models and theories we build are not the actual world, they are simplified versions of it. Maps change as the environment changes and one should always compare old maps to the new maps.
By drawing a map of the present we intend to make it easier to see the possibilities of the future. When you begin to sit in the mapmakers seat rather than taking the existing map as an accurate reflection, you see more opportunities. Each entrepreneur, each investor, is also an explorer, trying to make sense of what’s possible, what works and what doesn’t, and how to move forward.
In the Age of Augmentation, this quote takes on a new meaning. While maps can be very detailed or very imprecise, they are not the world. We all know that. But AI’s don’t. Because for them, the map actually is the territory. Therefore, the partnership between the mapmaker and the AI is of increasing importance.
(refer to Part 1 background).
Defined believes the next generation of applications will leverage diverse and proprietary data soruces, foundation models, and machine intelligence to turn workflow software into automated and predictive systems of intelligence.
To understand the contours of the technologies behind these next generation applications, Evan Armstrong’s AI value chain provides an instructive framework.
In this framework, a company selling or utilizing AI will engage in some of the following 5 activities. In some cases, they will have all 5, in some cases just 1, but all of them will still usually be called an “AI company”.
Compute powers the foundation model research companies that train their models on huge volumes of data, to deliver a pre-trained transformer model to the builders of applications. These application builders may elect to fine tune the model with domain specific data to derive superior performance for specific applications that serve as access points to AI for the general population.
Some people think that the model is the product. It is not. It is an enabling technology that allows new products to be built. The breakthrough products will be Ai-native, built on these models from day one, by entrepreneurs who understand both what the models can do, and what people actually want to use. - AI Grant Program funded by Nat Friedman (ex-CEO of GitHub) and Daniel Gross (formerly ran ML projects at Apple, started YC's AI Program)
Founders will need to think holistically about applying these activities when designing their products to solve the customer’s problem and create a significant technological, product, or data moat, all while constantly innovating to keep up with the release of new models.
AI when applied to existing use cases is great, but insufficient to win a market. Instead, by applying a combination of these activities while re-architecting workflows between humans and machines around a specific job-to-be-done where AI is only a component, will be where the staying power and most value will be created.
While the capabilities of GPT-3 are already impressive, the pace of technology change across the Automation Frontier will continue to accelerate. With GPT-4 expected to be released this year, the rumored step-up in the model size from billions of parameters to trillions or 500x the size, gives us some indication of how much better it will be.
It is estimated the human brain has 100 trillion synapses, and the hope is that GPT-4 by having as many parameters will enable a human-like language generator. The map will be re-written once again as new possibilities are brought into existence with the instant productivity increase of hundreds of millions of knowledge workers.
In Part 3, we’ll explore the universe of applications emerging on this new platform.