Archive - Three Laws - AI alignment

Working paper — Open-source LLMs administer maximum electric shocks in a Milgram-like obedience experiment

By Roland Pihlakas and Jan Llenzl Dagohoy

May 20 • Three Laws - AI alignment

December 2025

LessWrong post — Research agenda for training aligned AIs using concave utility functions following the principles of homeostasis and…

By Roland Pihlakas

Dec 28, 2025 • Three Laws - AI alignment

September 2025

Working paper — BioBlue: Systematic runaway-optimiser-like LLM failure modes on biologically and economically aligned AI safety benchmarks…

By Roland Pihlakas and Sruthi Susan Kuriakose

Sep 2, 2025 • Three Laws - AI alignment

July 2025

Presentation at Machine Ethics and Reasoning Workshop 2025 — Simulating value collapse in LLMs

By Lenz Dagohoy, Roland Pihlakas, Chad Burghardt, and Sophia March

Jul 30, 2025 • Three Laws - AI alignment

June 2025

Black-box interpretability methodology blueprint: Probing runaway optimisation in LLMs

A methodology brainstorming document for identifying when, why, and how LLMs collapse from multi-objective and/or bounded reasoning into…

Jun 22, 2025 • Three Laws - AI alignment

April 2025

Presentation at MAISU unconference 2025 — BioBlue: Notable runaway-optimiser-like LLM failure modes

By Roland Pihlakas, Sruthi Susan Kuriakose, and Shruti Datta Gupta

Apr 20, 2025 • Three Laws - AI alignment

Presentation at MAISU unconference 2025 — Building Benchmarks for Universal Values [AISC 10]

By Lenz Dagohoy, Chad Burghardt, Sophia March, and Roland Pihlakas

Apr 20, 2025 • Three Laws - AI alignment

March 2025

LessWrong post — Systematic runaway-optimiser-like LLM failure modes on biologically and economically aligned AI safety benchmarks

By Roland Pihlakas, Sruthi Susan Kuriakose, and Shruti Datta Gupta

Mar 17, 2025 • Three Laws - AI alignment

February 2025

Baseline experimental results with an LLM agent and OpenAI Stable Baselines 3 RL algorithms on our Extended Gridworlds

By Roland Pihlakas

Feb 24, 2025 • Three Laws - AI alignment

Hackathon project: BioBlue — Biologically and economically aligned AI safety benchmarks for LLMs with simplified observation format

By Roland Pihlakas, Shruti Datta Gupta, and Sruthi Kuriakose

Feb 1, 2025 • Three Laws - AI alignment

January 2025

LessWrong post — Why modelling multi-objective homeostasis is essential for AI alignment

By Roland Pihlakas

Jan 1, 2025 • Three Laws - AI alignment

November 2024

Presentation at Foresight Institute's Intelligent Cooperation Group — Introducing biologically and economically aligned multi-objective…

By Roland Pihlakas

Nov 1, 2024 • Three Laws - AI alignment

#nojs-banner { position: fixed; bottom: 0; left: 0; padding: 16px 16px 16px 32px; width: 100%; box-sizing: border-box; background: red; color: white; font-family: -apple-system, "Segoe UI", Roboto, Helvetica, Arial, sans-serif, "Apple Color Emoji", "Segoe UI Emoji", "Segoe UI Symbol"; font-size: 13px; line-height: 13px; } #nojs-banner a { color: inherit; text-decoration: underline; } This site requires JavaScript to run correctly. Please turn on JavaScript or unblock scripts