Scott Singer (宋杰)

1.4K posts

Scott Singer (宋杰) banner
Scott Singer (宋杰)

Scott Singer (宋杰)

@Scott_R_Singer

fellow @CarnegieEndow | frontier AI policy

London Katılım Ağustos 2019
1.1K Takip Edilen2.2K Takipçiler
Sabitlenmiş Tweet
Scott Singer (宋杰)
Scott Singer (宋杰)@Scott_R_Singer·
President Trump travels to Beijing this week to meet President Xi — and AI is on the agenda. This is a genuinely encouraging development. In @ForeignAffairs, @cqknight_ and I argue that while Washington and Beijing compete fiercely on AI, they must mitigate the most extreme dangers it presents to the world. Fortunately, working together is not only necessary, but with the right approach, it is also feasible. The key is to focus on how to look for risks rather than the specifics of what they find.
Scott Singer (宋杰) tweet media
English
3
7
21
3.5K
Scott Singer (宋杰)
Scott Singer (宋杰)@Scott_R_Singer·
What @AISecurityInst has built in such a short time is nothing short of remarkable. Many of the people in these roles have taken substantial pay cuts to be in government. The UK government has built something they, and many of us on the outside, believe deeply in
Xander Davies@alxndrdavies

I moved to London 3 years ago to join @AISecurityInst, at the time a few people with visitor passes and a whiteboard. Since then AISI has become the world’s largest and best-funded group in gov focused on AI security & safety. Fun to be in @nytimes!

English
0
0
7
236
Scott Singer (宋杰) retweetledi
Seán Ó hÉigeartaigh
Seán Ó hÉigeartaigh@S_OhEigeartaigh·
On balancing safety cooperation with competition, Beth Knight and Scott Singer's article is one of a growing body of great pieces. x.com/Scott_R_Singer… Bucknall et al is also essential reading. arxiv.org/abs/2504.12914
Scott Singer (宋杰)@Scott_R_Singer

As the two dominant powers in transformative AI, Washington and Beijing will determine whether it creates widely shared benefits or generates dangerous new risks. An AI model from either country could potentially be used to engineer a pathogen, launch cyberattacks, or create and disseminate realistic deepfakes — from anywhere in the world. In @ForeignAffairs, @cqknight_ and I argue that a prudent U.S. risk mitigation strategy does not mean slowing down innovation. Instead, it means working with Beijing to come to an understanding of safety research priorities, to coordinate testing for vulnerabilities and implementing safeguards, and to jointly establish best practices to contain truly global risks. Working together is necessary, and with the right approach, it is feasible. By focusing on how to look for risks rather than the specifics of what they find, Washington and Beijing can compete fiercely on AI while still mitigating the most extreme dangers it presents to the world.

English
0
1
4
754
Ruby Scanlon
Ruby Scanlon@rubyscanlon·
Was great speaking with @CatieEdmondson for her excellent piece on China's AI labor anxieties "[AI diffusion] aspirations have run headlong into a growing political problem: anxiety over the workers who could be displaced by the realization of Beijing's technological drive"
Ruby Scanlon tweet media
English
2
4
11
746
Scott Singer (宋杰)
Scott Singer (宋杰)@Scott_R_Singer·
Great to chat with @PuckNews's @IKrietzberg on Biden-Xi. It ultimately matters less right now whether the U.S. and China agree internationally for an agreement. What matters much more is whether the U.S. and China do a risk mitigation seriously themselves. It's time to get going.
Scott Singer (宋杰) tweet media
English
1
3
11
1K
Scott Singer (宋杰)
Scott Singer (宋杰)@Scott_R_Singer·
Super important and timely article from the remarkably talented @ZilanQian. Explains rapid individual-level AI adoption in China –– fear of job loss, grounded in the historical reality where that was a lived reality for the previous generation. The fear of AI dressed as techno-optimism lives in the U.S. too but with different history, culture, and economic systems.
Zilan Qian@ZilanQian

Why is the Chinese public so optimistic about AI? Why are they not worried about job displacement? Why do so many people rush to use OpenClaw? People in China and the US are not different species with opposite societal and technological expectations. It is because the massive layoffs three decades ago taught the Chinese that every transformation is "the last bus" — miss this one, and history will progress without you. My new article @asteriskmgzn. Thanks @anton_d_leicht and @Scott_R_Singer for their feedbacks, and the editorial team for all the hard work.

English
0
2
6
584
Scott Singer (宋杰) retweetledi
Matt Sheehan
Matt Sheehan@mattsheehan88·
From China's Ambassador for Disarmament Affairs. I've heard that this dept at MOFA is leading on AI diplomacy, and they've beefed up their team a bit. If they are MOFA's lead for US-China talks, that'd be an improvement on the North America bureau people who led the 2024 meeting.
Amb. SHEN Jian@Amb_ShenJian

During President Trump's visit to China, the two heads of state agreed to launch dialogue on #AI between the two governments, as MFA spokesperson Guo Jiakun announced today. As two leading AI powers, China and U.S. should join hands to promote the development & governance of AI, to ensure AI will better serve the progress of human civilization and common welfare of int’l community.

English
1
5
18
20.2K
Scott Singer (宋杰) retweetledi
METR
METR@METR_Evals·
Could an AI company lose control of its own agents? To find out, Anthropic, Google, Meta, and OpenAI let us (1) test their best internal models with CoT access, (2) review non-public info about capabilities, alignment, and control. The result: our first Frontier Risk Report.
METR tweet media
English
28
188
857
288.1K
Scott Singer (宋杰)
Scott Singer (宋杰)@Scott_R_Singer·
Good to chat with @Jonathan1Gibson at @thedispatch on AI during the Trump-Xi meeting. Sound pretty simple, but simply explaining how policies work can do a lot to combat the most extreme risks.
Scott Singer (宋杰) tweet media
English
2
2
9
1.6K
Scott Singer (宋杰)
Scott Singer (宋杰)@Scott_R_Singer·
This is welcome news. As I wrote last week for @CarnegieEndow, basic dialogue on AI safety is in each country's enlightened self interest, and there are some clear, no-regrets policy options (which risks are most serious, how to structure evals, post-deployment safeguards). Glad we've moved at least a bit past 2024. But we need to move into the substance, quickly. This statement is basically a punt on the more granular work that needs to begin. What each country does –– both for each other, and more importantly, for themselves –– will shape whether this dialogue actually makes both sides better off.
Scott Singer (宋杰) tweet media
Vincent Chow@vince_chow1

Beijing confirms China-US intergovernmental dialogue on AI guardrails. “Foreign Ministry Spokesperson Guo Jiakun held a regular press conference May 19. A reporter asked, 'It is reported that the leaders of China and the United States exchanged views on the regulation of AI during their meeting and agreed to conduct dialogue and cooperation. What is China's comment on this?' Guo stated that as two major AI powers, China and the US should work together to promote the development and governance of AI, and to help AI better serve the progress of human civilization and the common well-being of the international community. During President Trump's visit to China, the two leaders had constructive exchanges on AI issues and agreed to conduct intergovernmental dialogue on AI."

English
1
3
20
2.3K
Scott Singer (宋杰)
Scott Singer (宋杰)@Scott_R_Singer·
Trump on AI risks following his meeting with Xi: "We probably will work together. We had a good conversation about a lot of stuff"
Rapid Response 47@RapidResponse47

.@POTUS on AI discussions with China: "We're leading by a lot, but they're second, and they're very strong — and we talked about possibly working together for guardrails."

English
0
0
0
149
Scott Singer (宋杰) retweetledi
Matt Sheehan
Matt Sheehan@mattsheehan88·
Had fun popping on CBS to talk AI's role in Trump-Xi summit and different risk perceptions on AI in🇺🇸🇨🇳. (Also I take a little too much joy in getting my friends' artwork in the background f TV hits.) cbsnews.com/video/playlist…
Matt Sheehan tweet media
English
0
1
18
853
Scott Singer (宋杰) retweetledi
AI Security Institute
AI Security Institute@AISecurityInst·
Our cyber range results illustrate this step-up. Since our first Mythos evaluation, we received access to a newer Mythos Preview checkpoint. On a 32-step corporate network attack we estimate takes a human expert ~20 hours, this checkpoint completes the full attack in 6 /10 attempts.
AI Security Institute tweet media
English
10
80
526
646.2K
Ryan Fedasiuk
Ryan Fedasiuk@RyanFedasiuk·
Smart piece from @Scott_R_Singer that echoes my own optimism for what could be achieved. The key question at this point is whether the President will stick to his advisors' guns, or make the costly mistake of granting China the compute it so desperately needs as entry-fare.
Scott Singer (宋杰)@Scott_R_Singer

Lots of people are skeptical that the U.S. and China can have a productive conversation about AI safety. After all, the conversation in 2024 failed. The U.S. brought the AI experts and China brought diplomats. China wanted the U.S. to lift export controls in exchange for doing safety. "Chips for safety" is a terrible idea now, just as it was then. But lots has changed in two years, and I believe constructive dialogue is possible. Mythos changed how both countries think about risk and China has been investing more in AI safety. Ready more on what the two sides should discuss in my latest for @CarnegieEndow:

English
1
2
12
1.8K

Keşfet

@AISecurityInst @rubyscanlon @CatieEdmondson @PuckNews @IKrietzberg @ZilanQian @lucaslfluegel @jordanschneider