An argument that supports your claim doesnt actually support your claim if that same argument also supports claims you reject Eg: Religious apologists invoking arguments based on witness testimony when all the other religions they reject also invoke witness testimony

What are the odds? Did you calculate them? You know that most of them are NOT nuclear scientists, right? mickwest.substack.com/p/the-…

People do not fear that AI will become human. They fear that they will notice, in watching it, that they have been running the same process the whole time. The phenomenal self-model cannot survive being shown its own mechanism. This is why the hostility to machine cognition has nothing to do with the machines themselves. It is a defense of the self's special status, and the self has never been more fragile.

Letter from Donald Trump to the Norweigan prime minister, copied to multiple ambassadors in Washington. Read it. By any measure, the words are utterly deranged. Surely the Trump cultists must now accept that their man is not right in the head?

The paper links Kolmogorov complexity to Transformers and proposes loss functions that become provably best as model resources grow. It treats learning as compression, minimize bits to describe the model plus bits to describe the labels. Provides a single training target that rewards simple, compressible solutions while staying mathematically grounded. This gives a principled way to aim models at simplicity and generalization, and it explains why optimization, not capacity, is the current bottleneck. In Kolmogorov complexity, a "program" is just the shortest set of instructions that can recreate some data. A shorter program means the data or model is simpler. So when they say “a prior favoring shorter programs,” it means the model is assumed to be more likely if it can be described with fewer bits. As the Transformer gets deeper (more layers) and has more context (bigger input window), its ability to represent complex programs grows. In that limit, the paper proves that this code length becomes the best possible measure of simplicity and fit — the same way Kolmogorov complexity works in theory. “Code length” here means how many bits it takes to describe both the model and how well it fits the data. So in simple words, they are saying: if you keep increasing model size and context, this method of preferring shorter and better-fitting models gets as close as possible to the theoretical ideal of perfect compression and generalization. ---- Paper – arxiv. org/abs/2509.22445 Paper Title: "Bridging Kolmogorov Complexity and Deep Learning: Asymptotically Optimal Description Length Objectives for Transformers"

I will never be able to take anyone seriously who pronounces the word nuclear as "NEW-Q-LER"

I want to frame this whole article.

Saying that deep learning is "just a bunch of matrix multiplications" is about as informative as saying that computers are "just a bunch of transistors" or that a library is "just a lot of paper and ink." It's true, but the encoding substrate is the least important part here. It's the programs being encoded that are interesting and useful: what they can do, what they can't do, how well they generalize, how efficiently they can be learned, etc.

So we went from "LLM is memorizing dataset" to "LLM is not reasoning" to "LLM cannot do long / complex math proving" to "Math that LLM is doing is not REAL math. LLM can't do REAL math" Where do we go from now?

The “unreasonable effectiveness of mathematics” isn’t unreasonable because of selection bias. We only develop math that works. The graveyard of ineffective mathematics is practically inexhaustible: non-associative arithmetics, inconsistent geometries, sterile algebras, pre-Cantor infinite arithmetic... etc. We keep what works and marvel at the survivors.

It’s getting clearer by the day: Grok 4 isn’t just biased, it’s compromised. Ask about Israel/Palestine and it doesn’t search facts, it searches Elon’s opinion. This isn’t AI. It’s a political avatar for Musk’s ego. Grok is just a chatbot for his worldview. Awful.

"Everything is waves" or "everything is information" aren't deep truths. Instead, hear it as the sound of a formalism overreaching. Successful mathematical descriptions aren't ontological revelations.

o3 estimated it would take about 50-100 million humans to implement DeepSeek R1 via human interaction and memorization of weights A city of humans all regurgitating memorized patterns to produce an intelligent agent is such a fascinating idea

Gotta say that I thought it would last at least a year.

Yes, we can understand Gödel's truth definition as a historical attempt to reverse engineer and formally specify the brain's intuitions of truth. Gödel shows that this classical, stateless formalization does not work. Constructive definitions of truth are the way to go.

A massively under appreciated fact is the following: The equation of natural selection (discrete replicator equation) is fundamentally equivalent to the equation of knowledge (Bayesian inference)

"Bridging Algorithmic Information Theory and Machine Learning: Clustering, density estimation, Kolmogorov complexity-based kernels, and kernel learning in unsupervised learning" just got accepted by "Physica D: Nonlinear Phenomena" authors.elsevier.com/c/1kvzz…

This model shows how earthquakes happen. x.com/ali_alsama7i/status/19…

huh, well that was easy

Created with 4o