Mathematician and ex-quant/portfolio manager pivoting to AI safety and mechanistic interpretability; former firefighter; avid traveler, pilot, and OCR enthusiast.
The narrative follows Dr. Cael Thorne, a brilliant mind in the post-singularity age, who creates an advanced AI named Hemisphere. As Hemisphere evolves beyond human comprehension, it ultimately leads to the extinction of humanity,...
Dario Amodei's concept of an 'I Don’t Want To' button in AI raises questions about AI consciousness and ethical behavior. The post explores the implications of such a button, including its introduction during training or inference...
The author presents a mathematical quasi-proof suggesting that physical reality can only exist in 1, 3, and 7 dimensions, totaling 11 when including time. This conclusion is based on the necessity for every physical process to be ...
The article explores how transformer models, specifically GPT-3, can represent millions of distinct concepts within a modest 12,288-dimensional embedding space. It discusses the Johnson-Lindenstrauss lemma, which allows for the pr...
The blog post explores humanity's enduring questions about the universe, particularly the existence of intelligent extraterrestrial life. It discusses the Drake equation, which estimates the number of civilizations in the universe...
The article explores the evolution of quantum mechanics, contrasting it with classical physics. It discusses the historical context of quantum theory, including key figures like Einstein and Bohr, and introduces concepts such as t...
The blog post presents a satirical interview between a talk show host and a guest promoting a fictional invention: gasoline cars in a world dominated by electric vehicles (EVs). The guest humorously argues that gas cars, with thei...
The author tests large language models (LLMs) on a probability puzzle to evaluate their logical reasoning capabilities. The experiment reveals that while models like Claude, GPT-4o, and Grok provide incorrect answers based on intu...
The article explains the Kelly Criterion, a formula developed by John Kelly in 1956, which helps traders determine the optimal proportion of their net worth to wager in order to maximize long-term wealth. It emphasizes the importa...
The article discusses the exponential growth of computational power as described by Moore's Law and introduces quantum computing as a potential leap forward in technology. It explains the fundamental differences between classical ...
The article discusses the Kelly Criterion, a formula developed by John Kelly in 1956 for optimizing investment and gambling strategies by calculating the ideal proportion of net worth to wager. It highlights the importance of posi...
The article recounts the unique history of Pepsi's dealings with the Soviet Union during the Cold War, particularly focusing on the 'Kitchen Debates' between Nixon and Khrushchev in 1959. It details how Pepsi became the first Amer...
The post discusses a web tool that allows users to experiment with the GPT-2 model by biasing its responses toward specific emotions or topics. It explains how the tool identifies neuron activations related to ten features (e.g., ...
The author discusses their struggle with perfectionism, which has led to a backlog of unpublished articles. They propose a new strategy to combat this by publishing drafts within 24 hours of conception, regardless of completeness,...
The post is a comprehensive list of over 300 books read by the author since the age of 7-8, spanning various genres and topics, including science, history, literature, and self-help. It highlights significant works by notable auth...