Skip to content
Longterm Wiki
Back

Paul Christiano's AI Alignment Blog

web
ai-alignment.com·ai-alignment.com/

Paul Christiano is one of the most cited researchers in technical AI alignment; this blog contains seminal posts that introduced iterated amplification and AI safety via debate, which are now central research directions in the field.

Metadata

Importance: 82/100homepage

Summary

Personal research blog by Paul Christiano, a leading AI safety researcher, covering foundational concepts in scalable oversight, iterated amplification, AI safety via debate, and related technical alignment approaches. The blog has been highly influential in shaping modern alignment research directions at organizations like ARC and Anthropic.

Key Points

  • •Primary source for iterated amplification and debate as scalable oversight mechanisms
  • •Introduces and develops theoretical frameworks for aligning powerful AI systems
  • •Covers topics including eliciting latent knowledge, myopic training, and AI-assisted evaluation
  • •Influential in shaping research agendas at ARC Evals, Anthropic, and the broader alignment community
  • •Serves as a working document repository for Christiano's evolving safety research ideas

Cited by 1 page

PageTypeQuality
Paul ChristianoPerson39.0

Cached Content Preview

HTTP 200Fetched Mar 20, 202656 KB
[Sitemap](https://ai-alignment.com/sitemap/sitemap.xml)

[Open in app](https://play.google.com/store/apps/details?id=com.medium.reader&referrer=utm_source%3DmobileNavBar&source=---publication_layout_nav-----------------------------------------)

Sign up

[Sign in](https://medium.com/m/signin?operation=login&redirect=https%3A%2F%2Fai-alignment.com%2F&source=collection_home---publication_layout_nav-----------------------global_nav------------------)

[Medium Logo](https://medium.com/?source=---publication_layout_nav-----------------------------------------)

Get app

[Write](https://medium.com/m/signin?operation=register&redirect=https%3A%2F%2Fmedium.com%2Fnew-story&source=---publication_layout_nav-----------------------new_post_topnav------------------)

[Search](https://medium.com/search?source=---publication_layout_nav-----------------------------------------)

Sign up

[Sign in](https://medium.com/m/signin?operation=login&redirect=https%3A%2F%2Fai-alignment.com%2F&source=collection_home---publication_layout_nav-----------------------global_nav------------------)

![](https://miro.medium.com/v2/resize:fill:64:64/1*dmbNkD5D-u45r44go_cf0g.png)

[![AI Alignment](https://miro.medium.com/v2/resize:fill:128:128/1*N56Qc5-aHTcfGff0scntKQ.png)](https://ai-alignment.com/?source=collection_home_page----624d886c4aa4----------------------------------------)

[![AI Alignment](https://miro.medium.com/v2/resize:fill:160:160/1*N56Qc5-aHTcfGff0scntKQ.png)](https://ai-alignment.com/?source=collection_home_page----624d886c4aa4----------------------------------------)

## AI Alignment

[3.4K followers](https://ai-alignment.com/followers?source=collection_home_page----624d886c4aa4----------------------------------------)·

![Paul Christiano](https://miro.medium.com/v2/resize:fill:40:40/1*BNjZCuQuRfIgcXCBMipuBw.jpeg)

1editor

Follow

[**My research methodology** \\
**I explain why I focus on the “worst” case when doing theoretical alignment research.**](https://ai-alignment.com/my-research-methodology-b94f2751cb2c?source=collection_home_page----624d886c4aa4-----0-----------------------------------)

[![Paul Christiano](https://miro.medium.com/v2/resize:fill:20:20/1*BNjZCuQuRfIgcXCBMipuBw.jpeg)](https://paulfchristiano.medium.com/?source=collection_home_page----624d886c4aa4-----0-----------------------------------)

[Paul Christiano](https://paulfchristiano.medium.com/?source=collection_home_page----624d886c4aa4-----0-----------------------------------)

·Mar 22, 2021·18 min read

[![Paul Christiano](https://miro.medium.com/v2/resize:fill:20:20/1*BNjZCuQuRfIgcXCBMipuBw.jpeg)](https://paulfchristiano.medium.com/?source=collection_home_page----624d886c4aa4-----0-----------------------------------)

[Paul Christiano](https://paulfchristiano.medium.com/?source=collection_home_page----624d886c4aa4-----0-----------------------------------)

·Mar 22, 2021·18 min read

[**An unaligned benchmark** \\
**What an unaligned AI might look like, how it could go wrong, and how we cou

... (truncated, 56 KB total)
Resource ID: c47be710c3b15e51 | Stable ID: ZDNmM2Q3ND