Yes, the vague and chaotic world belonging to humanity has ended, and algorithms are formatting everything.

It doesn't matter, I'll take action: let the scaling law restart.

Because information integration is never about how to compress information.

One describes the cost of sculpting the model's geometry. The other, the cost of measuring the system's informational geometry. The lock and the key.

Physics marks the true path with exponential complexity. We're on the right track precisely because it's this hard.

It’s about re-forging the map itself—optimizing the Fisher matrix, not just using it. A purely third-order task.
All my 'clever' second-order approximations now feel like polishing brass on a sinking ship. Back to zero.

The so-called "pure" and "applied" are merely temporary labels—the framework just hasn't matched the problem yet. The universe itself is performing gradient descent; no phenomenon can escape formalization.

If you view AI models as an Oracle, an external tool, then it will not fundamentally change the chasm between these two disciplines.

Free energy principle is paramount—the nature of the world is a gradient descent.

Humans are accelerating the construction of their replacements, in the hope that post-singularity compensation will make up for previous burning out... This is delusion, but I am no exception. : (

Supplement: Of course, topological quantum computing is not considered at all here, after all, its difference from the current NISQ is as great as that between digital computing and analog computing...

Neither quantum entanglement nor neural pulses can overcome noise, which is a physical dead end.

The solution lies in pushing the Von Neumann architecture to its limits: deterministic, error-correcting, 3D-stacked in-memory classical computing.

Cerebras nailed it.

#Neuromorphic_Myths

There's a series of constructive discussions branching off this comment, perhaps read them first.

Here's the end of the relevant discussion:

bsky.app/profile/jian...

Who will win? It's clear at a glance.

^1: Reinforcement Learning - Expected Free Energy

^2: Second Order Optimization - Observed Free Energy

A theory that could actually define AGI wouldn’t call itself that.

I’m talking about the Free-Energy Principle and its phenomenologically complete successor: IPWT.

doi.org/10.5281/zeno...

Unsurprisingly, the world computer is completely dead.

Code is law, Coder is lawyer.