Lots of stuff in the pipeline, but to start with here's a series of 6 new talks from Evan Hubinger (author of Risks From Learned Optimization, the Mesa-Optimizers paper), going through a load of interesting AGI Safety Ideas.
2023-05-10 15:12:29 +0000 UTC
View Post
I made a new video with Computerphile, about ChatGPT!
https://www.youtube.com/watch?v=viJt_DXTfwA
Language Models, Simulacra, Reinforcement Learning from Human Feedback, Scaling and Inverse Scaling, ...
New main channel video! I'm still doing these!
This one's about how it's surprisingly hard to make AI systems that consistently tell the truth, and the safety problem that poses.
https://youtu.be/w65p_IIp6JY
...
Michael Trazzi interviewed me for his podcast The Inside View! We talked about what you'd expect I suppose, YouTube, AI, Stampy, all that good stuff. He also gave me a t-shirt?
2022-08-19 22:25:12 +0000 UTC View Post
You might notice I don't talk about specific AI takeover scenarios, with hacking, bribery, nanotechnology etc. Why is that?
2022-08-15 00:57:12 +0000 UTC View Post
I got a new camera for filming low-friction videos, and took it to the zoo! It takes a while to get it set up, testing footage at 9:20
I'M NOT DEAD! I did get COVID though. But I'm ok now.
I'm still working on the next main channel video, which turned out to be much harder than I expected. In the meantime here is a short video about current events (which aren't even that current now that I've got the video made). This vide...
2022-07-07 15:18:13 +0000 UTC View Post
This is a second talk following up the one I posted here: https://www.patreon.com/posts/62888854
This is somehow even lower production value than the last one, since the camera was stuck in a weird mode where it...
I'm at a research retreat right now with a few independent alignment researchers! The other day John Wentworth presented a talk giving his take on the alignment problem, and we filmed it just to have a record. I took photos of the flipchart after and edited them in. So the result is long, has rel...
2022-02-22 06:06:55 +0000 UTC View Post
https://youtu.be/HYtJdflujjc
A quick short to let people know that The Alignment Research Center is offering prizes of up to $50k for proposals for their Eliciting Latent Knowledge problem!
It's a very interesting report, ...
I get frustrated sometimes. A long time ago I made this video rant about it, but the production quality was far too low to publish it so I only released it to patrons. I've now (largely thanks to urging from Nick and Nicole) gone back and re-made it properly, and this is that.
Some friends of mine are running a (free) bootcamp "to bring people interested in AI Alignment up-to-speed with the state of modern ML engineering", and they thought my Patreon would be a quality source of candidates. This is a great way to get started learning the skills to do actual AI Safety w...
2021-11-14 21:03:35 +0000 UTC View PostWho here is on TikTok? I'm experimenting with it a bit, here's an attempt at a video about misalignment and GitHub Copilot!
My impression so far is:
- 3 minutes is a little limiting, but not bad at all
- the on-device editing tools are no good
- but, the shorter format has lower ex...
I got a new audio recorder and microphone, which works unlike any other I've used!
Thank you for your support :)
https://www.youtube.com/watch?v=sMYNjttnY4w
Some people ran real versions of the thought experiments from the 'Mesa-Optimisers' videos! What they found won't shock you (if you've been paying attention)
Also UPCOMING E...
2021-10-08 20:00:03 +0000 UTC View Post
I spent the last 3 weeks or so in the USA, doing various things which have given me a lot of video fuel but no actual videos, so here's a remaster of an existing second channel video:
A while back I gave a talk at "AI and Politics" in London, which I think was a decent introduction to AI S...
The previous video explained why it's *possible* for trained models to end up with the wrong goals, even when we specify the goals perfectly. This video explains why it's *likely*.
https://youtu.be/IeWljQw3UgQ
Also ...
If you're good at video editing, especially using open source software like Kdenlive and Blender, consider applying, to help me make more videos more quickly! :)
http://www.nonlinearfund.org/videoeditor.html 2021-03-27 18:45:57 +0000 UTC View Post
This "Alignment" thing turns out to be even harder than we thought.
https://youtu.be/bJLcIBixGj8
Also UPCOMING EVENT TODAY, we will have an experimental Live Watch Party on the Discord ...
2021-02-13 20:03:36 +0000 UTC View Post
How do you get an AI system that does better than a human would, without doing anything a human wouldn't?
A follow-up to "Maximizers and Satisficers": https://youtu.be/Ao4jwLwT36M
The Paper: 2020-12-05 15:48:51 +0000 UTC View Post
My camera died in the middle of shooting the new video! So I had to get a new one - this is that.
Thanks for making it possible for me to just buy new equipment when I need it!
https://youtu.be/JgmHftE4Bs0
I'm starting a Discord Server! It's been running for a little bit for patrons at the 'Advisor', 'Hero' and 'Personal thank you in a video' support levels, and now I'm opening it up a little wider, to include the 'Thank you in videos' tier.
I made a video explaining the idea of the Discord, ...
2020-09-22 16:10:01 +0000 UTC View Post
AI might create enormous amounts of wealth, but how is it going to be distributed?
There's also a patreon-exclusive video of the full conversation I had with the author of this paper, Cul...
2020-07-03 14:56:34 +0000 UTC View PostJust thought I should let patrons know that we've had a go at recording Computerphile remotely, talking about OpenAI's new GPT-3 model.
I wanted to include a photo but there's really nothing to show since it's just me at my regular desk with Sean on the phone :/
Anyway hopefully that ...
2020-06-17 10:40:58 +0000 UTC View Post
Why do some ignore AI Safety? Let's look at 10 reasons people give (adapted from Stuart Russell's list).
https://www.youtube.com/watch?v=9i1WlcCudpU
Enjoy!
2020-05-31 15:54:50 +0000 UTC View Post
Answering the other half of the questions you asked: Moral philosophy, biological intelligence enhancement, nukes, regulating compute clusters, lethal autonomous weapons, etc!
https://www.youtube.com/watch?v=0m...
AI systems do what you say, and it's hard to say exactly what you mean.
Let's look at a list of real life examples of specification gaming!
2020-04-25 20:27:15 +0000 UTC View Post
I shot a video, so I had to cut my hair first. For some reason I filmed it, and now you can watch that. This is what we have become. This is The New Normal.
Seriously though this is just me cutting my hair, feel free to skip it if that sounds like it's not your preferred kind of weird.
<... 2020-04-25 20:25:26 +0000 UTC View Post
You asked, I answered :)
I ended up shooting a lot of footage for this, so this video is part 1 of 2.
Let me know if you want to see more Q+A videos in future!
2020-03-10 12:19:16 +0000 UTC View PostIn case any of you are as well, look me up on the networking app
2020-02-21 14:53:10 +0000 UTC View Post