On the quest to understand the fundamental mathematics of intelligence and of the universe with curiosity.
Burny
Karma: 187
Current public Codex seems more capable than current public Claude Code
if theoretical physicists had discovered deep learning instead, they would call it something like Effective Non-Equilibrium Multiscale Hierarchical Non-Linear Coarse-Graining Spin-Glass Dynamics Renormalization Group Theory Ansatz (ENEMHNLCGSGDRGTA)
It’s quite surreal to look at my 2022 notes about AGI when almost no one in the mainstream knew about the concept and now it’s very much present in the mainstream
Good job
A lot of things in technology and engineering evolve quickly and much of the knowledge about it becomes outdated quickly as a result. But the knowledge that will forever remain relevant is the fundamental mathematics that governs that technology with the mathematics of the physical reality across all scales in which it all lives.
Starting to happen in software engineering, and AI engineering/research
Also I think a lot of people say a lot of things not as falsifiable factual claims, but as status signaling or values signaling or being part of ingroup signaling.
And if you want to do literally statistical physics on top of deep learning: https://deeplearningtheory.com
Dr. Jeff Beck is lovely transdisciplinary galaxybrain, mathematician turned computational neuroscientist doing physics+neuroscience-inspired machine learning
opus 4.5 is 4o for nerds https://x.com/justalexoki/status/2010027088217342095
“A mathematician builds the most explanatory model they can which they can still prove theorems about.”
″A physicist builds the simplest model they can which still explains the key phenomena.”
- Authors of Principles of Deep Learning Theory book
”An engineer builds the most realistic/accurate model they can which can still be computed within budget.”
- jez2718
How I think about alignment and ethics as a cooperation protocol software
“Claude Sonnet 4.5 was able to recognize many of our alignment evaluation environments as being tests of some kind, and would generally behave unusually well after making this observation.”
How do you rate the lowered sycophancy of GPT-5, relatively speaking?
According to Jan Leike, Claude Sonnet 4.5 It’s the most aligned frontier model yet https://x.com/janleike/status/1972731237480718734
I really like the definition of rationalist from https://www.lesswrong.com/posts/2Ee5DPBxowTTXZ6zf/rationalists-post-rationalists-and-rationalist-adjacents :
“A rationalist, in the sense of this particular community, is someone who is trying to build and update a unified probabilistic model of how the entire world works, and trying to use that model to make predictions and decisions.”
I recently started saying that I really love Effective Curiosity:
Maximizing the total understanding of reality by building models of as many physical phenomena as possible across as many scales of the universe as possible, that are as comprehensive, unified, simple, and empirically predictive as possible.
And I see it more as a direction. And I see it from a more collective intelligence perspective. I think modelling the whole world in fully unified way and in total accuracy is impossible, even with all of our science with all our technology, because we’re all finite limited agents with limited computational resources and time, limited modelling capability, and we get stuck in various models, from various perspectives, and so on. And all we have is approximations, that predict certain parts reality to a certain degree, but never fully all of reality in perfect accuracy in all it’s complexity. And we have a lot of blind spots. All models are wrong but some predictively approximate the extremely nuanced complexity of reality better than others.
And from all of this, intelligence and fundamental physics, which are subsets of this, are the most fascinating to me.
I like your definition of rationalism!
I recently started saying that I really love Effective Curiosity:
Maximizing the total understanding of reality by building models of as many physical phenomena as possible across as many scales of the universe as possible, that are as comprehensive, unified, simple, and empirically predictive as possible.And I see it more as a direction. And I see it from a more collective intelligence perspective. I think modelling the whole world in fully unified way and in total accuracy is impossible, even with all of our science with all our technology, because we’re all finite limited agents with limited computational resources and time, limited modelling capability, and we get stuck in various models, from various perspectives, and so on. And all we have is approximations, that predict certain parts reality to a certain degree, but never fully all of reality in perfect accuracy in all it’s complexity. And we have a lot of blind spots. All models are wrong but some predictively approximate the extremely nuanced complexity of reality better than others.
And from all of this, intelligence and fundamental physics, which are subsets of this, are the most fascinating to me.
Lovely podcast with Max Tegmark “How Physics Absorbed Artificial Intelligence & (Soon) Consciousness”
Description: “MIT physicist Max Tegmark argues AI now belongs inside physics, and that consciousness will be next. He separates intelligence (goal-achieving behavior) from consciousness (subjective experience), sketches falsifiable experiments using brain-reading tech and rigorous theories (e.g., IIT/φ), and shows how ideas like Hopfield energy landscapes make memory “feel” like physics. We get into mechanistic interpretability (sparse autoencoders), number representations that snap into clean geometry, why RLHF mostly aligns behavior (not goals), and the stakes as AI progress accelerates from “underhyped” to civilization-shaping. It’s a masterclass on where mind, math, and machines collide.”
Advanced version of Gemini with Deep Think officially achieves gold-medal standard at the International Mathematical Olympiad
Whaaat!?
Gemini 2.5 pro is way worse at IMO and got 30%, and DeepThink version gets gold??
But it’s more finetuned for IMOlike problems, but I bet the OpenAI’s model was too.
Both use “novel RL methods”.
Hmm, “access to a set of high-quality solutions to previous problems and general hints and tips on how to approach IMO problems”, seems like system prompt, as they claim no tool use like OpenAI.
Both models failed the 6th question which required more creativity
Deepmind’s solutions are more organized, more readable, more well written than OpenAI’s.
But OpenAI’s style is also more compressed to save tokens, so maybe going more out of human-like language into more out of distribution territory will be the future (Neuralese).
Did OpenAI and DeepMind somehow hack the methodology, or do these new general language models truly generalize more?
Often when I think, write, say, etc. something, I put it to all LLMs connected to the internet, and say something like:
“Fact-check every distinct factual claim in the following text. For each claim, search the web to verify it, then report whether it’s accurate, inaccurate, partially accurate, misleading, oversimplified, unreliable, unverifiable, or unfalsifiable. Note any important nuance or context that’s omitted. Be as scientific, rational, sceptical, etc., as possible, do not just agree. Quote the specific claims and provide your findings with sources you relied on:”
And how it often corrects me is great. But sometimes it’s wrong (supported by sources) and I have to correct it back or we have a discussion.
GPT-5.4 is the most correcting (thx bro) and Gemini and Grok are the most sycophantic fucks (I don’t believe you telling me I’m absolutely right all the time), and Claude is in the middle. I haven’t tested enough other models yet, like the Chinese and open models.