You're so close to revealing the question that begs to be asked. Holding code language constant, the language itself doesn't matter. That's correct. They wash out asymptomatically in complexity theory. Big O ignores coefficients because they don't change the shape of growth, just the slope, the eventual growth rate is the same class whether lin, quad, exp, etc. Almost<- all "wink" real systems will inherit the shape of the S-curve because eventually external constraints kick in. In intelligence systems these constraints manifest as a result of things like throughput capacity and number of users. The goal of maintaining balance in the inflection point of their journey along sigmoidal saturation, as can be seen, is a losing battle. Before the midpoint a wealth of super-linear growth. But positive feedback loops stagnate beyond that midpoint with the same additions barely moving the needle due to resource contention, latency, or bandwidth limits. Maintaining 3 maximums at K/2 (K being the carrying capacity) becomes unsustainable past the inflection point. Noise, pollution, friction, and data drag collapse any system's signal to noise ratio. This is where we are with current SOTA. In steps the BRUTE FORCE approach throwing compute at it's sigmoidal enemy like bare knuckle boxer trying to overpower entropy's oldest friend by sheer power alone. But we all know who wins this fight. Capacity inflation isn't efficiency innovation. You cannot change the shape of the curve, you can only stretch K outward. You can optimize all you want, every system eventually bends because of resource finiteness and the current undefeated champion of the universe, entropy. So, what's the question? If all ends inevitably the same, whether code or compute, then how do we ever move past the curve. Fortunately for you, if you were kind enough, or bored enough, or smart enough to make it this far, I'm happy to inform you, that the S curve isn't invincible. In a universe where everything is LLM shaped, this solution is just as cool and unique as you might already imagine it to be.
You're so close to revealing the question that begs to be asked. Holding code language constant, the language itself doesn't matter. That's correct. They wash out asymptomatically in complexity theory. Big O ignores coefficients because they don't change the shape of growth, just the slope, the eventual growth rate is the same class whether lin, quad, exp, etc. Almost<- all "wink" real systems will inherit the shape of the S-curve because eventually external constraints kick in. In intelligence systems these constraints manifest as a result of things like throughput capacity and number of users. The goal of maintaining balance in the inflection point of their journey along sigmoidal saturation, as can be seen, is a losing battle. Before the midpoint a wealth of super-linear growth. But positive feedback loops stagnate beyond that midpoint with the same additions barely moving the needle due to resource contention, latency, or bandwidth limits. Maintaining 3 maximums at K/2 (K being the carrying capacity) becomes unsustainable past the inflection point. Noise, pollution, friction, and data drag collapse any system's signal to noise ratio. This is where we are with current SOTA. In steps the BRUTE FORCE approach throwing compute at it's sigmoidal enemy like bare knuckle boxer trying to overpower entropy's oldest friend by sheer power alone. But we all know who wins this fight. Capacity inflation isn't efficiency innovation. You cannot change the shape of the curve, you can only stretch K outward. You can optimize all you want, every system eventually bends because of resource finiteness and the current undefeated champion of the universe, entropy. So, what's the question? If all ends inevitably the same, whether code or compute, then how do we ever move past the curve. Fortunately for you, if you were kind enough, or bored enough, or smart enough to make it this far, I'm happy to inform you, that the S curve isn't invincible. In a universe where everything is LLM shaped, this solution is just as cool and unique as you might already imagine it to be.
It's all sigmoida, all the way down.