Loved this piece and the idea of context windows for a lot of reasons, but three that Iʻve been thinking about recently:
Data vs Metadata:
In my limited circles, I made a reputation for myself as a "data guy" and yet have strayed from that practice a bit. Iʻve found myself telling people recently that "the thing about data" is that what tends to be important in interpreting or analyzing any given piece of data is actually the metadata (i.e. the context) which is much more difficult to acquire, store, transfer, etc. Data is essentially an extremely lossy compression of reality. So Iʻve been much more interested in processes of acquiring and transferring context -- those decompression algorithms that allow us to accurately reconstruct the reality data attempts to capture.
Design Patterns:
I hear a lot of people talking about complexity and a complex world -- rules and technologies that are sufficiently complex that no individual node within the system can comprehend them. I donʻt have a problem with this, but I do think itʻs perhaps a strategic error to design the structures and frameworks that undergird the complexity in divergent ways. If you spend a lot of time developing expertise in, say, the sport of Tennis, wouldnʻt it be great if you could transfer that expertise to another useful realm, say, Healthcare? Designing in a way where the context developed has universal value might not be a good recipe for competitive advantage, but would have huge value for problem solving and achievement. My conclusion for this is modeling our design practices after biological systems and other naturally occurring networks and ecosystems would give us the best opportunities to make the development of context transferable
Full Employment:
A friend of mine runs a waste diversion social enterprise that deconstructs buildings as an alternative to demolition (and then sells the materials as a second hand resource). They are generally able to achieve cost parity with demolition with two critical differences -- the jobs take much longer, but... they also employ more people. Itʻs sort of the opposite of machine powered human efficiency. Their enterprise gets a lot of funding from workforce development and economic development initiatives because of this job-creating characteristic.
As a new parent, Iʻm trying to "read ahead" a bit to see what Iʻm in for when my daughter is a teenager and something I learned from Ellen Gallinskyʻs "The breakthrough years" is that "contribution" is generally understood to be a psychological need (not just for teenagers but for all of us). If the goal of the economy is creating an efficient network of relationships to do arbitrary things, I think the AI question bears out one way. If the goal of the economy is creating an efficient network of relationships to fill a need for contribution, I think it plays out differently.
Like with farming, most demolition in my market is "conventional" rather than deconstruction, and fewer people are involved in the activity than at other points in history. But when I need to feel psychologically fulfilled, I find myself planting things, volunteering to do conservation work, remembering to water my houseplants
Thanks Ben! Insightful as always! I've come to view the full employment issue via the lens of returns to capital vs returns to labor -- the latter has been under pressure for a while but it seems clear that humanoid robotics will further accelerate that process; against that, I think we will eventually hit on some form of value-added redistribution, so that planting trees in a socially fulfilling way can be a useful and economic activity, rather than UBI which has all the problems of Utopian societies. I hope Hana doesn't live in a world where robots make everything and analog humans clean up the mess en masse!
Loved this piece and the idea of context windows for a lot of reasons, but three that Iʻve been thinking about recently:
Data vs Metadata:
In my limited circles, I made a reputation for myself as a "data guy" and yet have strayed from that practice a bit. Iʻve found myself telling people recently that "the thing about data" is that what tends to be important in interpreting or analyzing any given piece of data is actually the metadata (i.e. the context) which is much more difficult to acquire, store, transfer, etc. Data is essentially an extremely lossy compression of reality. So Iʻve been much more interested in processes of acquiring and transferring context -- those decompression algorithms that allow us to accurately reconstruct the reality data attempts to capture.
Design Patterns:
I hear a lot of people talking about complexity and a complex world -- rules and technologies that are sufficiently complex that no individual node within the system can comprehend them. I donʻt have a problem with this, but I do think itʻs perhaps a strategic error to design the structures and frameworks that undergird the complexity in divergent ways. If you spend a lot of time developing expertise in, say, the sport of Tennis, wouldnʻt it be great if you could transfer that expertise to another useful realm, say, Healthcare? Designing in a way where the context developed has universal value might not be a good recipe for competitive advantage, but would have huge value for problem solving and achievement. My conclusion for this is modeling our design practices after biological systems and other naturally occurring networks and ecosystems would give us the best opportunities to make the development of context transferable
Full Employment:
A friend of mine runs a waste diversion social enterprise that deconstructs buildings as an alternative to demolition (and then sells the materials as a second hand resource). They are generally able to achieve cost parity with demolition with two critical differences -- the jobs take much longer, but... they also employ more people. Itʻs sort of the opposite of machine powered human efficiency. Their enterprise gets a lot of funding from workforce development and economic development initiatives because of this job-creating characteristic.
As a new parent, Iʻm trying to "read ahead" a bit to see what Iʻm in for when my daughter is a teenager and something I learned from Ellen Gallinskyʻs "The breakthrough years" is that "contribution" is generally understood to be a psychological need (not just for teenagers but for all of us). If the goal of the economy is creating an efficient network of relationships to do arbitrary things, I think the AI question bears out one way. If the goal of the economy is creating an efficient network of relationships to fill a need for contribution, I think it plays out differently.
Like with farming, most demolition in my market is "conventional" rather than deconstruction, and fewer people are involved in the activity than at other points in history. But when I need to feel psychologically fulfilled, I find myself planting things, volunteering to do conservation work, remembering to water my houseplants
Thanks Ben! Insightful as always! I've come to view the full employment issue via the lens of returns to capital vs returns to labor -- the latter has been under pressure for a while but it seems clear that humanoid robotics will further accelerate that process; against that, I think we will eventually hit on some form of value-added redistribution, so that planting trees in a socially fulfilling way can be a useful and economic activity, rather than UBI which has all the problems of Utopian societies. I hope Hana doesn't live in a world where robots make everything and analog humans clean up the mess en masse!