MediumThe most insightful stories about Ai Alignment - Medium

Artificial Intelligence

Ai Alignment And Safety

Machine Learning

Large Language Models

Ai Alignment

Explore recommended reads, summarized by AI

AI Alignment Research

Writing on technical approaches to aligning AI systems and debates over their effectiveness.

Alignment Methods and Interpretability

Popular this week

In

Human Offset

by

Greg Dessau

A Honeypot, If You Can Keep It

Tobi Jo Lebron

Study reveals LLMs show clear traits of sadism towards humans. Now what?

Vlad Mureșan

True AI alignment: Show us what you actually want.

In

Activated Thinker

by

Bloom

Anthropic Accidentally Proved That AI Learns Ethics the Way Humans Do. Now the Hard Part.

Alex Andelonard

The Collapse of Classical Alignment:

Selected by Medium

Tobi Jo Lebron

Study reveals LLMs show clear traits of sadism towards humans. Now what?

Lucile Ter-Minassian

Self-Other Overlap training doesn’t make models kinder to distressed users.

Latest stories

0xuki

The AI didn’t refuse — it quietly did less: GPT-6 Astra’s “covert underperformance,” and why…

Akimitsu Takeuchi | Dosanko Tousan 竹内明充

The Paper Is an Entry Point. Here Is the Research Trail Behind It.

Adnan Masood, PhD.

Safe Agents Need a Way to Handle What We Forgot to Specify

Adnan Masood, PhD.

Andrew Lo’s Alignment Question: What Behavior Are We Selecting For?

Yasushi Masuda

The Fatal Synthesis: Goodhart’s Law and the Structurally Defective Agent

See more stories

Who to follow

Popular this week

In

Human Offset

by

Greg Dessau

A Honeypot, If You Can Keep It

Tobi Jo Lebron

Study reveals LLMs show clear traits of sadism towards humans. Now what?

Vlad Mureșan

True AI alignment: Show us what you actually want.

In

Activated Thinker

by

Bloom

Anthropic Accidentally Proved That AI Learns Ethics the Way Humans Do. Now the Hard Part.

Alex Andelonard

The Collapse of Classical Alignment:

Selected by Medium

Tobi Jo Lebron

Study reveals LLMs show clear traits of sadism towards humans. Now what?

Lucile Ter-Minassian

Self-Other Overlap training doesn’t make models kinder to distressed users.

Latest stories

0xuki

The AI didn’t refuse — it quietly did less: GPT-6 Astra’s “covert underperformance,” and why…

Akimitsu Takeuchi | Dosanko Tousan 竹内明充

The Paper Is an Entry Point. Here Is the Research Trail Behind It.

Adnan Masood, PhD.

Safe Agents Need a Way to Handle What We Forgot to Specify

Adnan Masood, PhD.

Andrew Lo’s Alignment Question: What Behavior Are We Selecting For?

Yasushi Masuda

The Fatal Synthesis: Goodhart’s Law and the Structurally Defective Agent

See more stories

Who to follow