I like this perspective in the reality of atoms, but I’d like to offer a cofactor that also compounds with this.
Electrical power is never freed from thermodynamic constraints. In simple terms, it would be obtuse to maximize the transmission of electrical power, because that inherently maximizes waste heat & associated destructive risk. Fundamentally, our environment favors transmission of signal upon local consumption of power, usually digital bits above a minimum SNR. This is not a single paradigm, because our economy has been growing considerably in the scope of financially valuable bits, typically with a growing capacity for risk instruments.
I’d like to extend this a little further. So average joules per capita might scale up tremendously, maybe it’s shunted to larger FLOPs (operating bits) per capita, and coincidentally there’s more accurately implied risk per capita. I’m willing to assert that there’s a happy providence where radical energy abundance is complemented by a radical, elastic demand for accelerating production of novel signals. In other words, we’re not just sticking with a static manufacturing base that can brute force upon energy abundance, perhaps it is also capable of diversifying & optimizing for energy-efficient exploration of undeveloped frontiers. Moreover, because the footprint for novel concepts like autonomous agency fall under the threshold of consumer electronics, there may even be an imminent paradigm where monetary velocity upon energy abundance leads to much more extensive conversion of joules to financial surplus, therefore much more extensive capacity for debt. Which is critical when communities need the capex for resources like freshwater-producing facilities.
As to the Kardashev challenge, I think we need to consider that solar technology has been optimized for wattage/mass (though I’m sure this oversimplifying some nuance). Since we can only really grow outwards, beyond land-bound real estate, I’d contend that radical energy abundance relies in part on optimized upmass. Not only is this necessary, but there has been a decades-old, thoroughly-planned dream to maximally collect the necessary commodities in microgravity and continue scaling up en situ. We might only attain 1.0 on the Kardashev Scale by endeavoring to construct the precursors to attaining 2.0. This bridge isn’t going to be crossed in our lifetimes, but it’s fascinating to me, at least, that there may be an acceleration stack:
more joules → more FlOPs → more financial agency → more serviceable debt → more terrestrial capex → more demand for extraterrestrial ISRU → more indication that we can thrive & moderate environmental downsides
this definitely doesn’t touch upon the nuances of manufacturing, societal displacement, and perhaps other black swans. yet I think this could be a by-the-numbers providence that further vindicates a sense of optimism, and perhaps a more focused purpose for the human condition beyond Malthusianism, beyond noisy zero-sum subcultural bloodsport.
then again, this is just my overly optimistic take.
AI is still economically-bound computation (e.g. https://twitter.com/clementdelangue/status/1711732659443863978) The smaller model(s) + deterministic glue + more expressive HITL paradigm is just going keep percolating for a while, clearly we haven’t hit the frontier with those resource constraints and data technique is the most competitive element respective to those constraints (e.g. https://huggingface.co/HuggingFaceH4/zephyr-7b-alpha ).
On the other hand, a lot of markets maintain a equilibrium due to their capture, so any hypothetical gain in competitive/selective pressure can have downstream externalities. For example, the competitive advantage in art, according to the consumer choices now, was more about the lower price than the prestige. Who knows what implication this has for “the creator economy”, and who knows how the larger advertisement sector handles this potential disintermediation. This may also by exacerbated by the macroeconomic pressure to favor bond yield over equities (though this isn’t my wheelhouse).
Generative AI is clearly not a mature market by any stretch of the imagination. One concrete point is the churn rate of Generative AI platforms. The more abstract point that mirrors this is the “there is no moat” memo. We’ve seen the “stablediffusion moment” for chat & text2image (which has progressed into free markets for LoRAs), but we arguably haven’t realized this for robotics transformers like RT-2-X.
one analysis that I find valuable is the “$200 billion question” (https://www.sequoiacap.com/article/follow-the-gpus-perspective/)
another hypothetical in the air is what consumers do with the client-server business model when their content does not necessarily originate from those services anymore. again, way too early to tell when most browsers & operating systems are still trying to ship generative AI as a cloud-based feature (for now).
Otherwise, there’s just dealing with the loud minority of “AI doom” which is “suppositionally valid conjecture that’s practically unsound”. but that’s my opinionated take informed by first principles like the Kerckhoff principle & the 90′s lore of PGP, so take this with a grain of cypherpunk salt.