Burny - Effective Curiosity
@burny_tech
On the quest to understand the fundamental mathematics of intelligence and of the universe with curiosity. http://burnyverse.com Upskilling @StanfordOnline
Hey! Follow me for explorations of intelligence, mathematics, science, engineering, technology, artificial intelligence, machine learning, physics, computer science, (not only computational) neuroscience, cognitive science, transhumanism, AI engineering, AI's benefits, risks,…
The in-context learner of the "beautiful @GoogleResearch paper" is a meta learner like @HochreiterSepp's 2001 meta LSTM [1] which learned by gradient descent (GD) a learning algorithm that outperformed GD - no test time weight changes! Since 1992, GD can learn learning algorithms…
Beautiful @GoogleResearch paper. LLMs can learn in context from examples in the prompt, can pick up new patterns while answering, yet their stored weights never change. That behavior looks impossible if learning always means gradient descent. The mechanisms through which this…
#A newly identified quantum state in hafnium pentatelluride demonstrates unique electron-hole pairing, potentially enabling radiation-resistant, self-charging computers for deep space applications. @physrevlett doi.org/g9vbdj phys.org/news/2025-07-p…
Donaldson+Scaduto proposed two conjectures on existence and uniqueness of special Lagrangians in Calabi-Yau 3-folds (extra dimensions in String Theory). With friends, I proved local versions of both in 2 papers. Fun fact: They are just accepted on the same day by G&T and IMRN!
The AI field is now split into (A) a "traditional" ml/dl domain, and (B) a "psycho-AI" domain where innovation requires an understanding of / intuition about the cognitive capabilities of pretrained models and how to prompt / fine-tune them. These two fields are IMO separated.
*hits unknown drug* What if we made sand think by melting it and striking it with lightning?
we discovered alien intelligence in sand and like 1% of the world cares lol
A psychedelic drug disorganizes cortical traveling waves. Groovy work. Complex slow waves in the human brain under 5-MeO-DMT doi.org/10.1016/j.celr… #neuroscience
The summer schools at Les Houches are a magnificent tradition. I was honored to lecture there in 2023, and my notes now are published as "Ambitions for theory in the physics of life." scipost.org/SciPostPhysLec…
Math, physics, astronomy, reality. Everything in the universe in one graph. A glimpse into the mind of "God." Source: pubs.aip.org/aapt/ajp/artic… A few of you asked for more info on this figure. The authors are showing masses, sizes, and relative densities of objects in our…
Reinforcement Learning of LLMs Latest tutorial taught by Prof. Ernest Ryu of UCLA , in spring of 2025 Link in comments 😎👇
in a fractal world, nothing is boring if you keep observing it deeper. and nothing complex, if you zoom out enough.
credit - @AcerFur It appears OpenAI are now routing all o3 requests in ChatGPT to GPT-5 ("zenith" in LMArena). It now gets extremely difficult mathematics questions o3 had a 0% success rate in correct/very close to correct and is significantly different stylistically to o3.
We have a very poor understanding of why deep neural networks like transformer models learn the parameters they learn. For example, in the paper below from 2013, the authors demonstrated that 5% of the weights of a trained deep neural network can be used to predict the values of…
Should you watch lectures or read textbooks? That's probably a preference. I prefer both at once. Lectures with lecture notes or books that the lectures are based on. Video transfers some information that only text doesn't transfer and is more engaging for me. And books have more…
The more you excercises when learning math, the more you will understand it, and the more you are able to follow the next lectures without getting lost. My rule of thumb is that if I can't explain it and write it down, if I can't prove it, if I can't apply it, etc., then I don't…
To first approximation, my mind has roughly 3 philosophers inside it: - Science and engineering philosopher, who wants every model to be implementable in our physical reality to be real, who wants models that predict physical reality empirically, and the degree of realness of a…
RL with verifiable rewards? Works great ✨ Realistic or non-verifiable tasks? Still a mess 📉 Reward models and AI judges? Fragile and inconsistent 💔 Our proposal? RL from Checklist Feedback 📜 arxiv.org/abs/2507.18624 👇
arxiv.org/abs/2507.18472 Another nice IKKT related paper this year. If spacetime can truly and fully emerge from random matrix models, either holographically or through some other duality, it will be a paradigm shift, also impacting how we think about quantum mechanics.