LESSWRONG
LW

2042
Boundaries (membranes) for AI safety by Chipmonk
Boundaries / Membranes [technical]AI
Frontpage

3

Boundaries Update #1

by Chris Lakin
11th Apr 2024
1 min read
2

3

This is a linkpost for https://formalizingboundaries.substack.com/p/update-1
Boundaries / Membranes [technical]AI
Frontpage

3

Previous:
What does davidad want from «boundaries»?
1 comments47 karma
Next:
Retrospective on Mathematical Boundaries Workshop
No comments22 karma
Log in to save where you left off
Boundaries Update #1
10Thomas Kwa
1Chris Lakin
New Comment
2 comments, sorted by
top scoring
Click to highlight new comments since: Today at 11:13 PM
[-]Thomas Kwa2y107

Any technical results yet?

Reply
[-]Chris Lakin2y10

Thanks for asking. This is the intention of Mathematical Boundaries Workshop which is running now. Let me know if you'd like to come on Sunday

Reply
Moderation Log
More from Chris Lakin
View more
Curated and popular this week
2Comments

Boundaries agenda updates in the last few months.

“What does davidad want from «boundaries»?”

davidad and I had a lesswrong dialogue I recommend reading.

If you need a refresher on boundaries, read both the above dialogue and the formalizingboundaries.ai website.

Conceptual Boundaries Workshop

We ran Conceptual Boundaries Workshop on Feb 10–12.

In attendance: David ‘davidad’ Dalrymple, Scott Garrabrant, TJ (Tushant Jha), Andrew Critch, Allison Duettmann, Alex Zhu, Jeff Beck, Adam Goldstein, Manuel Baltieri, Lisa Thiergart, Abram Demski, Evan Miyazono, and me.

For more about what we discussed, see Evan’s personal retrospective.

Supported by The Foresight Institute, Blake Borgeson, and the Long Term Future Fund.

ACX Grant

Scott Alexander granted us $40,000 for boundaries projects and workshops.

Mathematical Boundaries Workshop

Mathematical Boundaries Workshop is running this week for 5 days. Goal: develop boundaries math further, ultimately for application in real-world scenarios. Many category theorists are in attendance.

We are inviting a few guests to hang out at the end of the workshop — this Sunday morning, Berkeley CA. Email me chris@chrislakin.com if you’d like to come.

davidad’s ARIA programme now live

davidad’s ARIA programme for safeguarded AI is now live and soliciting applications for the first phase (>$74M over 4 years). See the ARIA page.

future updates

Subscribe: https://formalizingboundaries.substack.com/ 

Mentioned in
22Retrospective on Mathematical Boundaries Workshop
13Cooperation is optimal, with weaker agents too  -  tldr