Skip to main content

New AXRP with Samuel Albanie!

In this episode, I chat with Samuel Albanie about the Google DeepMind paper he co-authored called "An Approach to Technical AGI Safety and Security". It covers the assumptions made by the approach, as well as the types of mitigations it outlines.

Video
Transcript

Interesting asymmetry: it seems a lot more common to put an old tune to new words than the reverse.

I finally have a short and clearly-not-tracking-you link for my anonymous feedback form! If you want to give me feedback you can do so via w-r.me/feedback

If you want, you can verify that it doesn't track you or anything by looking at the corresponding public repo: github.com/benwr/w-r.me/blob/m…

I made a hacky link shortener this way for work reasons, and then realized it could work really well for the rare occasion like this, when I want to have a short link with no tracking.


Feedback form for Ben Weinstein-Raun

Google Docs
in reply to Ben Weinstein-Raun

Ben Weinstein-Raun
Ben Weinstein-Raun
•
(It might still be possible in principle that actually this is somehow served from some other repo - I don't know what happens if one tries to do a github pages site deploy with a CNAME that doesn't match the actual deploy URL; I hope it visibly fails but it might not)
in reply to Ben Weinstein-Raun

Ben Weinstein-Raun
Ben Weinstein-Raun
•
You can at least see that some kind of check happens in deployment that references the correct url: github.com/benwr/w-r.me/action…

What's an example where it actually makes sense to build your own agent? I see tons of tutorials floating around recently, but it's hard for me to imagine a case where I wouldn't just e.g. build an MCP for Claude Code instead. What am I missing?

PSA: You can use a GitHub Pages site as a personal link shortener. Plus you can use it to solve the "why should I trust that this link shortener isn't tracking me" problem, by making the backing repo public.

New AXRP with Peter Salib!

In this episode, I talk with Peter Salib about his paper "AI Rights for Human Safety", arguing that giving AIs the right to contract, hold property, and sue people will reduce the risk of their trying to attack humanity and take over. He also tells me how law reviews work, in the face of my incredulity.

Video
Transcript

in reply to Daniel Filan

Ben Weinstein-Raun
Ben Weinstein-Raun
•
Your thumbnail game is getting better
in reply to Ben Weinstein-Raun

Daniel Filan
Daniel Filan
•
Thanks! I tried making the text bigger and adding more question marks.
in reply to Daniel Filan

Daniel Filan
Daniel Filan
•
oh also because he called in his face is bigger and more front-on than if he were in person and I had a camera on him.
This entry was edited (1 week ago)
in reply to Ben Weinstein-Raun

Daniel Filan
Daniel Filan
•
I will say tho that this is not performing well so far compared to my other videos.
in reply to Daniel Filan

Daniel Filan
Daniel Filan
•
actually, it's not performing as well view-wise, but it is performing quite well in terms of cumulative time people have spent watching it. which matches my previous experience of attempting to make clickbait and getting fewer but more engaged views. maybe the 'clickbait' stuff is actually just a good description of what's happening in the interview?
⇧