@undo_hubris

PhD student at @UofT working on AI alignment theory. Heavily tattooed. Research Lead at https://xitter.cf/t.co/8LhzJnSE2C. Blog: https://xitter.cf/t.co/ivZ9BGOoOt

Joined January 2018
  • Tweets1.5K
  • Following1K
  • Followers1.2K
97Media
I'm excited to announce Principled Agents, a research nonprofit! Our aim is to develop principled solutions for core problems in AI alignment, ensuring the goals we give AGI are beneficial and safe. We're currently hiring and running a monthly discussion group, see below.
9
8
1
93
5,073
Rubi Hudson retweeted
Thomas Kwa
@Kwathomas0Oct 12
Proposed terminology for groups of agents: - Team: ≤105 agents. (size of college football roster) - Fleet: 105–6,939 agents. (largest fleet in history was 6,939 ships, for the D-Day invasion) - Flock: 6,940–3.7 billion agents. (A passenger pigeon flock over Ontario in 1860 had ~3.7 billion birds.) - Swarm / Collective: more than 3.7 billion, up to 12.5 trillion agents. (The largest recorded insect swarm was ~12.5 trillion Rocky Mountain locusts. As independent literary validation, the Borg Collective also canonically contains trillions of drones.) Beyond 12.5 trillion, the insect precedent runs out and we will need a new term. This is well-grounded in human experience and literature and is the only reasonable choice of terminology. thank you for coming to my ted talk
Ethan Mollick
@emollickJan 26
Lets not call groups of agents "swarms" - it is both terrifying (maybe the point?) & not a useful analogy. Groups of agents should be called teams or organizations. It both describes how to structure them and also how to use them. Don't let the weird AI folk naming win again!
12
15
2
218
10,113
There are two ways to make a useful AI product: more reasoning, and less reasoning
Joe Weisenthal
@TheStalwartSep 28
HSBC's head of AI and Data Science Mark Macdonald put out a note on Jev. Not much in it that will be new for the folks here who already care, but notable that this kind of model is getting Wall Street attention
8
496
Rubi Hudson retweeted
Nan Ma@mananselfSep 22
The U.S. flag packs 50 stars into a rectangle. I found a packing with stars 95% larger in diameter. It has central symmetry and is our best so far, though not proven optimal. I don't know if an optimal packing has central symmetry. AI-assisted search.
1
1
14
520
Rubi Hudson retweeted
🚀 Rocket@rocketalignmentSep 21
Just wait til antitrust law has to deal with acausal trade between AIs
1
3
22
848
The Simulation Hypothesis says the future will reassemble the past
1
4
126
This is an fantastic paper! My new #1 recommendation for mechanism designers interested in AI safety. I know Andrew has been thinking about alignment-related issues for years, and this work reflects that deep consideration.
Andrew Koh
@andrewjkohSep 2
We develop a mechanism design framework for AI alignment and control: arxiv.org/abs/2609.01595 It’s largely conceptual but we offer stylized applications to failure modes (sandbagging, alignment faking), safety practice (scalable oversight, peer prediction), and a way to think about the value of alignment, interpretability, capability, and control.
1
2
19
2,046
When the US standardized stop signs in the 1920s, everyone already agreed red was the right color ...after all, "red" was "stop" for traffic signals. The problem was that durable reflective red paint didn't exist... all the reds would fade in the sun. By the late 1940s, fade-resistant porcelain enamels and reflective sheeting finally made red feasible.
1
1
49
2,095
makes u think xitter.cf/StefanFSchubert/status…
Stefan Schubert
@StefanFSchubertAug 18
The borrowing to finance AI investment is contributing to rising interest rates on government debt, the FT reports Cf @BasilHalperin
4
5
51
4,531
This has gotta be one of the cutest AMF distributions - someone from Sumbe, Cameroon moved to the US. Later, when they went to visit family, they caught malaria. So they and their coworkers at PACE Analytical funded bednets for the whole village. againstmalaria.com/Distribut…

Against Malaria

Against Malaria. People from all over the world raising money to help combat malaria

againstmalaria.com
3
14
669
I think Leopold is irresponsible for brushing off misalignment risk without making a serious argument against it, but I would absolutely invest if he started a new hedge fund at current prices and promised to cap out at 3x leverage
1
15
1,695
I don't understand how an AI can forecast 50% more accurately than me. When I forecast 50%, I'm perfectly calibrated.
2
9
727
I'll be presenting my paper, Corrigibility Transformation, at the first ICML poster session today! Swing by to chat about what kinds of goals are safe to give powerful AI systems.
15
522
Rubi Hudson retweeted
Jesse Smith
@JesseTayRiverJun 3
The EA forum is basically tied with the Fine Homebuilding forum in uses of the term "load-bearing". 2995 for the EA forum vs 3029 for Fine Homebuilding
7
20
4
212
7,233
Every TV show should have one episode called "Future Imperfect" about an evil future and one episode called "Fearful Symmetry" about evil clones TBH.
1
6
19
1,340
not getting in the first 0.1c starship because i don’t want to be 0.2cmogged in 30 years
2
18
584
Rubi Hudson retweeted
Chana
@ChanaMessingerMay 13
The optimal number of things that are actually zero is nonzero
2
2
46
1,932
Claude Coat
🤖 Made with AI
116
118
62
2,865
195,062
The Researcher and Research Assistant roles start as remote, contract positions, but we hope to make full-time offers with in-person options. The details and application forms are at principledagents.org/hiring, please share with anyone you think could be a good fit.
2
1
16
1,646
Our discussion group meets monthly to discuss new research that ties together incentive theory and AI alignment. You can sign up for the mailing list at principledagents.org/discuss…, and suggest presenters (including yourself!) by emailing [email protected]. (3/3)
9
513
Load more