bloom - evaluate any behavior immediately  🌸🌱

HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusal

Sparse Autoencoder for Mechanistic Interpretability

Using sparse coding to find distributed representations used by neural networks.

Mechanistic Interpretability Visualizations using React

One of the Discord bots ever designed. Now incorporates a calls/channel link system, fortunes, deletion, dice, arbitrary code execution and reminders.

A lightweight (350MB) Lisp interpreter in Malbolge Unshackled, often dubbed the hardest turing complete programming language.

A Dependently Typed Functional Programming Language

A core concatenative programming language with variables and first-rank polymorphic type inference

Python pattern matching like functional languages.

Swedway (esolang)