Vishwaa Sofat

emerging tech + china + industrial policy @harvard_law | prev: @ncuscr @schwarzmanorg @ucberkeley @stuyny

[Rush Doshi @RushDoshi]

The fact that Beijing has placed travel restrictions on the families of AI professionals is shocking.

Strongly suggests they are also unlikely to allow foreign investment abroad if it entails meaningful technology transfer.

[Rush Doshi @RushDoshi]

I’m usually not attacked for being naive on China. But I believe strongly in sending students, with appropriate protections, to China. I was one of these students.

I cannot see why our piece should be controversial, but realize there are some who believe all exchanges should be halted. I am not among them.

Here is our argument:

-Longstanding federal programs that encourage US students to learn about China — like Fulbright, CLS, and Boren — should be funded by the US government. We had similar exchanges with the Soviets, and of course with China, in the past.

- Given growing risks to students in China, those participating in these programs must be protected from harassment or threats of arbitrary detention, which are growing, and which are why these programs were difficult to renew in the past.

- But there is a way out. Summits provide an opportunity to achieve commitments that protect these programs and their students. Trump should (1) elevate these programs into leader-level deliverables, and (2) reach formal (or, more likely, informal) agreements to insulate them from politics. The former sends a signal not to touch these programs and the latter sustains it.

- As part of this effort, Min Zin and other scholars and students arbitrarily detained or exit banned by China must be released.

- We did not write this in the piece, but obviously these efforts should be accompanied by proper orientation and training for students before they visit China. That was also the case for these programs in the past. It is more criticism now.

We have done all this before. It has worked before too. Leader-level recognition of certain programs in China can provide protection.

The US desperately needs China expertise. After a surge among millennials, what expertise we did have faces a cliff as students lose interest in and opportunities for studying China and Chinese.

That is a problem.

[Kyle Chan @kyleichan]

This is one of the key points from my New York Times piece.

Calls to slowdown AI don’t make sense in China where few think a crisis is imminent.

[Julian Gewirtz @JulianGewirtz]

New from me: Some argue there is no point in pacing the AI frontier until China cooperates. But Beijing is unlikely to take American calls for restraint seriously unless Washington is willing to restrain itself.

If the United States wants effective AI diplomacy with China, what it does at home may matter as much as what it says at the negotiating table.

I’m delighted to partner with @ReadTransformer for my first-ever cross-posted Substack essay—a look at how Beijing is approaching U.S.-China AI diplomacy, and what Washington needs to do to make that diplomacy more credible and effective.

https://t.co/U4F8TYuEHr

[Sheena Chestnut Greitens @SheenaGreitens]

It’s a good idea to pay close attention to what current & former officials in China are saying, in order to understand their strategy and behavior.

Understanding is not endorsement.

[Mayor Zohran Kwame Mamdani @NYCMayor]

On Raksha Bandhan, a single thread symbolizes a powerful promise: to protect, support, and stand by one another.

Today I shared in the joy of this ceremony with elders at India Home, where South Asian seniors find community, care, and chosen family.

That spirit extends to our entire city. May we continue to stand by one another as we build a safer, and more affordable city for all.

[Arvind Narayanan @random_walker]

Something amazing is happening in the debate about text watermarking in Claude. This is my best attempt to make sense of the quickly evolving situation.

The key thing to keep in mind is that it is in fact possible to watermark LLM-generated text without degrading output quality (and in fact achieve a much stronger property, which is that the distribution of possible outputs is unchanged.) This is well established technically, and it has been implemented by Google / Gemini for over two years, and no one seemingly cared.

Admittedly, quality-preserving watermarking is one of those counterintuitive facts about probability, like that annoying Monty Hall problem (the one with two goats behind three doors). The theorem-understanding part of my brain has no problem with it, but the intuitive part of my brain is screaming that there must be some mistake.

I have no interest in re-litigating this. What I’m curious about is why, when Anthropic announced that they are rolling out watermarking that doesn’t degrade outputs, so many of their customers seem to have concluded that they are lying.

I think three things went wrong:

1. Anthropic’s rollout was terrible from a comms perspective. There was no blog post; just a quiet support page with the title “How Claude marks AI-generated content” that said strangely little about how Claude actually marks AI-generated content. And it had no explanation of why they’ve rolled it out worldwide even though it’s required by law only in the EU. No transparency about who gets access to the watermark verifier.

2. Generally low trust and high suspicion about Anthropic’s motivations, given their statements and actions over the last few months / years.

3. Unresolved questions about the right tradeoff between individual users’ freedoms and the collective benefits of pervasive watermarking. Unlike Pangram-style AI detection, most people didn’t even know this was a thing, and it’s always uncomfortable to find out that a thing exists at the same time that you find out it’s mandatory with no opt-out.

Evidently surprised by the strength of the backlash, Anthropic has been doing damage control, focusing on explaining how it works. But this only addresses the first point above. Once the narrative that they are lying took hold, people seem to be willing to reject anything they say about this topic, including the feasibility of distortion-free watermarking.

This is also a textbook example (I plan to use it in my classes!) to explain why the “tech policy moves slowly because politicians don’t understand tech” narrative is ridiculous. Tech policy does move slowly, but for the same reasons all policy moves slowly. Figuring out the right tradeoffs in almost any situation — and getting public buy-in — is genuinely hard and can never happen at the speed of tech. If tech policy were to move even half as fast as people constantly claim they want it to move, absolute chaos would result. This situation is a good example.

[Emily Feng 冯哲芸 @EmilyZFeng]

“The instant I stepped through its glass doors, it felt like I had been teleported back to China, not the country I knew, but one I imagine it can be.” Gorgeous essay from @yangyang_cheng as usual https://t.co/OJHUAWaFpy

[Kevin S. Xu @kevinsxu]

As of today's America, an AI model's output or chain of thought is not copyrighted nor treated as trade secrets.

So it is not IP.

You may *feel* like it is because of something you've read. But it isn't.

Thus, alleging distillation as "IP theft" is not a serious legal argument. It is a political and lobbying talking point masquerading as a legal argument.

The one pillar about America I respect and cherish the most is our independent, self-correcting legal system. It isn't perfect. It doesn't get everything right. But it employs a rigorous process that tries to be immune to lobbying, political pressure, and "feelings", in order to deliver impartial and fair judgments, if not immediately, at least eventually.

You may not *feel* like a rectangle should get an IP patent either. But it did (Apple for the iPhone, patent no. D618,677)

Laws are laws. Until they are, they aren't. Until they aren't, they are.

[李其 Lizzi @wstv_lizzi]

We wrote a while ago in our private newsletter about the different positioning of Beijing, Shanghai, Hangzhou, and Shenzhen as China’s four 4 AI innovation hubs just as @kyleichan said and the bureaucratic competition underpinning each ecosystem and among them. (should probably update that to include the emerging Hefei Wuhan nexus which is increasingly central through eg CXMT + YMTC, optoelectronics, the broader memory and hardware cluster etc)

It is also a good time to remind readers that AI competition is not only a U.S.-China issue. It is also a (fierce!) subnational competition within China itself. Cities and provinces are competing for talent, capital, good companies, and Beijing leadership attention! while Beijing leadership is running the horse race but also trying to coordinate these rival ecosystems toward one national goal. Lots of tension between local bureacratic competition and national mobilization (another defining feature of China’s AI push less discussed…)

Will probably write something public about it at some point…

[Arvind Narayanan @random_walker]

This is one of the most important questions about AI and one that I've thought about. In countries like India, much of the economy is informal — heavily reliant on tacit knowledge that's not written down anywhere, processes never codified, "I know a guy who knows a guy" networks, and pervasive corruption. This underground economy is trapped in a vicious cycle: low wages + extreme illegibility → no incentive for private actors to build AI for their use cases → everything stays informal + low productivity. I don't know if governments and/or NGOs can help break this cycle, but if they can, its impact would easily exceed that of rural electrification. India has successfully built and deployed digital infrastructure for legibility in the past — Aadhaar and UPI — so there is hope.

Two important and depressing readings on related points:

Dispatches from India https://t.co/kPJ3Av3kPQ by @arjun_ramani3. There's a lot in there and I won't try to summarize it, but it helps explain why this project is not just a matter of building technology. "Many of the basic facts needed to operate in Indian society are not recorded anywhere, but rather sit inside the heads of various information brokers who live off the rents from keeping that information private." In other words, codification will be actively resisted by rent seekers who are adapted to the current system.

AI could keep poor countries poor https://t.co/N4WJ9HK6gX by @deenamousa. Past waves of technology and globalization benefited poor countries because they got to export the products and services created through their cheap labor. AI probably won't have this effect — it substitutes cheap labor even more cheaply. (Implication: this makes it even more important for them to find ways to use AI to improve their local economies.)

[Kyle Chan @kyleichan]

Utterly self-defeating.

“The National Science Foundation (NSF) has decided to ban collaborations between every U.S. scientist it funds and nearly all Chinese research institutions and their employees.” https://t.co/pJLx3yMc4i

[Julian Gewirtz @JulianGewirtz]

My latest, with the great @danwwang, in the New York Times today:

The American A.I. chaos of recent weeks is profoundly self-defeating. With its actions against Anthropic, the U.S. government is skating close to its own Jack Ma moment.

We often think of A.I. as a race between the United States and China. Instead we are seeing the emergence of an even more acute form of competition, between the public power of governments and the private power of ambitious companies.

The challenges of regulating A.I. at the frontier are urgent and important. The U.S. government needs to strike a better balance between ambition and control, lest it irrevocably damage America’s long-term technological edge.

Read it here: https://t.co/zO700STUmK

[Eleanor Olcott @EleanorOlcott]

Truly wild how many people in Beijing, even living in the same neighbourhood, have no idea a plane crashed into the city’s tallest building on Friday. The censorship is incredibly effective.

[Vishwaa Sofat @VishwaaSofat]

@EleanorOlcott Some love for him in Kunming as well that I snapped walking around earlier today

[Grace Safarik @GraceSafarik_]

The torch is passed in NY-12.

[Vishwaa Sofat @VishwaaSofat]

@ghostrunnerblog Honestly surprising in terms of demographics and even more so the consistent flyering Bores had along StuyTown compared to Lasher

[Ian Landsman @IanLandsman]

Awesome shot by @DevirKaitlin

[Jalen Brunson @jalenbrunson1]

Vibes are immaculate right now

[Ryan Mead @OhRyanMead]

Jalen Brunson:

"We show up at 9 o'clock"

That's the most New York fucking line ever.