(づ•ᴥ•)づ┬─┬
-
Matrix Calculus (for ML and beyond) -- notes.
-
Solving the inverse problem to find a hidden object
-
Greedy Coordinate Gradient
-
Compact transformers can't learn all sequences.
-
Translating stories to games with language models
-
Scale and Direction: Understanding Homogeneous Functions
-
Model Merging and the Geometry of Optimization
-
Some Linear Algebra from Lewis Signaling Games
-
“Reasoning with Sampling” — Notes on Karan & Du (2025)
-
A few notes on directed information
-
Fun with stochastic activations
-
Trying out the GEPA prompt optimizer with chess puzzles
-
Adding Facts and Watching LLMs Snap to Rationality
-
horizon-alpha and horizon-beta also betting, gpt-oss-20B is thinking first.
-
Directed graphs from papers with OpenAI o3
-
Using ChatGPT Agent to find a coffee carafe with a particular diameter
-
Heuristics when LLMs play a betting game
-
Turns out general agents contain world models.
-
Revisiting a Root-Finding Idea from Undergrad
-
LLM agents and chess puzzles
-
Further adventures with LLM curve-guessing
-
GPT-2's Anisotropic Antics
-
LLMs and Dynamic Cheatsheets: Learning from Experience
-
The Smallest Weight Punches Above Its Weight (in Expectation)
-
A brief analysis of automerger data, feat. SLERP and DARE-TIES LLM merging
-
Yet another self-attention tutorial
-
Rewriting the Kullback-Leibler with an integral transform