LLM/AI safety research at Palaestra Research. Previous: Berkeley
Repositories
ethanelasky/collusion-on-the-open-web
Code for 'Encoded Coordination on the Open Web' (LessWrong): agents coordinating through innocuous web services
ethanelasky/ARENA_3.0
ethanelasky/debate
ethanelasky/books-dl
Extract epub from books.com.tw. (purchased only)
ethanelasky/jobd
A small job scheduler for GPU pods: queues, DAGs, timeouts, retries, dependency-aware scheduling across a pod pool.
ethanelasky/agent-search-collapse-viewer
Interactive viewer for the agent search-collapse experiment
ethanelasky/verbatim-obsidian
Debate-style card cutting for Obsidian, modeled on Verbatim (paperlessdebate.com)
ethanelasky/b200-runtime
Published, digest-pinned B200 runtime images for the Debate scheduler
ethanelasky/transformers
🤗 Transformers: the model-definition framework for state-of-the-art machine learning models in text, vision, audio, and multimodal models, for both inference and training.
ethanelasky/mlx-lm
Run LLMs with MLX
ethanelasky/rubric-arena
ethanelasky/jp-mining-note
Temporary bugfix fork for JPMN
ethanelasky/sft-control
Online RL vs SFT for training-based AI control — Apart/Redwood hackathon March 2026
ethanelasky/debate-ai-control-notes
ethanelasky/textbook
ethanelasky/control-arena
ControlArena is a collection of settings, model organisms and protocols - for running control experiments.
ethanelasky/ld-debate-book
Lincoln-Douglas Debate from First Principles
ethanelasky/unsloth-zoo
Utils for Unsloth
ethanelasky/unsloth
Fine-tuning & Reinforcement Learning for LLMs. 🦥 Train Qwen3, Llama 4, DeepSeek-R1, Gemma 3, TTS 2x faster with 70% less VRAM.
ethanelasky/arnesen-debate-modeling
ethanelasky/cs285
Worked solutions for self-study of the course Deep Reinforcement Learning (CS 285) at Berkeley
ethanelasky/rlhf-book
Textbook on reinforcement learning from human feedback
ethanelasky/undergrad-course-notes
Notes from my time at Berkeley
ethanelasky/neelaca
ethanelasky/tmc
Comparing Taiwanese Media on China and US through NLP
ethanelasky/pyLDAvis
Python library for interactive topic model visualization. Port of the R LDAvis package.
ethanelasky/tdm-studio-parsing
Efficiently parse HTML/XML-structured ProQuest documents.
ethanelasky/venmo-validator
ethanelasky/swot
:school: Identify email addresses or domains names that belong to colleges or universities. Help automate the process of approving or rejecting academic discounts.