Trevor Levin

@trevposts

Trying to help the world navigate the potential craziness of the 21st century.

DC usually, SF sometimes

2013年6月からXを利用しています

1,963 フォロー中

4,961 フォロワー

固定
Trevor Levin
@trevposts
11月9日
Having just spent the week in San Francisco, I feel like I've truly internalized the insanity of this situation for the first time. I think anyone who has seen me in the last few days can attest that I look like I've seen a ghost. The leading AI companies are planning to
Do Not Accept
@DONOTACCEPTai
11月9日
AI companies admit what they’re building might kill everyone. We don't have to agree to this. Add your name at donotaccept.ai to say: we do not accept your terms.
69
Trevor Levin
@trevposts
7時間
My main hope that recursive self-improvement -- AIs training the next AIs, way faster than humans have trained AIs -- is more than 6-24 months away is that the AIs will merely get astoundingly good at stuff you can measure but not "fuzzy stuff." Unfortunately, I think Dwarkesh's
Dwarkesh Patel
@dwarkesh_sp
12時間
Why do I think that AIs trained in a data center during the intelligence explosion will still be really good at fuzzy stuff (like, say, running a Fortune 500 company)? Two big reasons: 1. The generalization we're already seeing at <1/1000th of the scale of RL training that will
16
Trevor Levin
@trevposts
7時間
I agree with this. Jacob Coxon has had an enormous impact. One thing I'd add is it might be more impactful if you do this with a few others (especially at other companies).
Rob Wiblin
@robertwiblin
15時間
Frontier AI company staff must not quiet quit. If you've seen things that caused you to lose faith that your employer can do what it's doing safely - or is being honest and candid with the public - you 100% have a moral duty to loudly tell the world.
Trevor Levin
@trevposts
7時間
Brits should go to this!
PauseAI UK
@PauseAI_UK
11月12日
We are organising the largest AI safety march in history, and it’s happening right here in London. We want you to be there.
Trevor Levin
@trevposts
13時間
Lots of great ideas in here. Hard to pick just one to highlight, but a mandatory delay period (of say 6 months) between the completion of a model's training and the use of that model to do AI R&D seems like a great framework.
Rob Wiblin
@robertwiblin
13時間
The midterms gave us a shot at averting AI disaster. But even with a new Congress it won't be easy. Here's the best plan I've got for turning this ship around: 1:59 4 things Congress can do immediately 4:25 Why it's not enough 5:44 Why we must prepare for warning shots 7:28
1

Trevor Levin

Trevor Levinさんのポスト