RSSAmplifier

Blog

Dimitri Lozeve's Blog

Recent posts

lozeve.comRSS feed ↗10 posts

Latest posts

Two Types of Bureaucracy: Enabling and Coercive

A summary of Adler and Borys's paper distinguishing enabling from coercive bureaucracy, and the features and factors that drive each form of formalization.

Johnson's Rule: scheduling items in two machines

Johnson's rule finds the optimal schedule for processing jobs through two machines. An explanation of the method, its intuition, and code in Python and BQN.

Overengineering Dice Rolling

A simple Dungeons and Dragons dice roller that grew into an over-engineered Rust tool with a parser, statistics, and a WebAssembly website.

Why I built Leibniz: in defense of Gantt charts and project scheduling

The origin story behind Leibniz, a project scheduling tool, and a defense of Gantt charts done well.

A Mastodon bot for interesting integer sequences

Building a Mastodon bot in Rust that posts a random interesting integer sequence from the OEIS every few hours to make its entries more discoverable.

The Bulgarian Solitaire

A discrete dynamical system on integer partitions that converges to triangular shapes, explained with Young diagrams and simulated in Python and BQN.

Language models are injective and hence invertible

How a recent paper proves Transformer language models are injective as functions from prompts to activations, and can be inverted to reconstruct inputs.

Reading notes: Kolmogorov-Arnold Networks

Reading notes on Kolmogorov-Arnold Networks, which replace MLP node weights with learnable spline activations on edges, grounded in a representation theorem.

Randomness and Uncertainty: from random noise to predictable oscillations via differential equations

Exploring how a differential equation can turn random noise into stable bistable oscillations, replicating Trefethen's experiment in Python with Diffrax.

High reliability organizations

Notes on high reliability organizations, the five traits that let nuclear plants and air traffic control manage risk, and their relevance to robust AI.