China-US AI Tensions and AI Safety Summit Controversy

0:00Today on the AI Breakdown, we're looking at new Biden administration policy making it even harder for China to access AI chips. Before that on the brief, New York City gets an AI policy, Anthropics Claude is available in more countries, and the chair of the SEC says AI causing a financial crisis is unavoidable. The AI Breakdown is a daily podcast and video about the most important news and discussions in AI. Go to Breakdown.network for more information about our YouTube channel, our Discord, and our newsletter. Welcome back to the AI Breakdown news you need in around five minutes. We start today on the product side of things where Anthropic has announced that their premier LLM, Claude, is now available to users in 95 countries.

0:47Now, Claude is probably best known at this point for its much longer 100k token context window. That gives people the ability to analyze much larger documents than you can, for example, with ChatGPT, which has been a really valuable use case for writers, for data analysts, and just other people that deal with longer documents. These 95 countries now have access to both the free Claude as well as Claude Pro. Now, interestingly, Anthropic also released the results of a recent experiment where they curated an AI constitution based on the opinions of around a thousand Americans and then trained a model against it using their approach to what they call constitutional AI.

1:23Now, whereas other LLMs use strategies like RLHF or reinforcement learning from human feedback as a way to ensure that AI responses are aligned with human values, Anthropic is trying to premiere something that they call constitutional AI. Effectively, they are trying to train their models on an underlying set of principles so that it can use those principles to reason about how it should approach different queries and situations. If it works, it's a much more scalable solution to some of the questions of AI safety and risk. Now, Anthropic's main version of constitutional AI takes principles from things like the U.S.

1:56Constitution, as well as other important historical human documents that represent foundational civilizational principles. It seems like for this study, they wanted to compare that to the opinions of a randomly sampled group that represented the U.S. in terms of age, gender, income, and geography. While they said a perfectly representative sample was impossible, they tried to have it match basic demographic statistics in the U.S. They write, We used polis to ask our public to deliberate on the normative values they would like AI to abide by, and then used those opinions to curate a new AI constitution.

2:25We found that the public constitution overlapped with the anthropic written constitution around 50 % of the time. One of the key differences, they said, the public constitution focused more on objectivity and impartiality, and placed a greater emphasis on accessibility. Now when they trained a model to adhere to the principles in that collectively designed constitution, they found that, quote, the collectively designed public model to be slightly less biased and equally as capable as the standard Anthropic model. Now, it feels pretty clear that what they're exploring is the right way to give people ownership and representation in the way that we design these systems.

2:57One of the messy, important things about democracy is that people have voice. When it comes to decisions made by big tech companies, obviously people do not have voice. So I like seeing that Anthropic is at least doing those experiments to see what it looks like when you actually just ask people directly what principles AI should follow. Now, moving over to the realm of AI and finance, SEC Chair Gary Gensler, who has made enemies of just about every financial establishment at this point, from crypto to traditional finance and beyond, is now really, really focused on AI risks. In an interview with the Financial Times, he said that without swift intervention, it was quote, nearly unavoidable that AI would trigger a financial crisis within a decade.

3:34Gensler said, it's frankly a hard challenge. It's a hard financial stability issue to address because most of our regulation is about individual institutions, individual banks, individual money market funds, individual brokers. It's just in the nature of what we do. And this is about a horizontal where many institutions might be relying on the same underlying base model or underlying data aggregator. So basically, it sounds like Gensler's concern is that all of these high-frequency trading firms and other investors are going to be using the same models to make the same decisions, which creates the potential for everyone moving in the same direction all at once, which could cause serious market dislocations.

4:08He's worried, in other words, about herd behavior. Now, as much as Gensler has been a thorn in the side of many market participants, it does seem like this is at least worth an exploration and not an out-of-hand dismissal. Although what the right answer is, even he is not suggesting at this time. Here's a new one. You may remember Praz as the Fuji's rapper, or you may remember him as someone who was just convicted for money laundering, illegal lobbying, and campaign finance violations in April. Well, he has now requested a new trial alleging that his former attorney used AI for the closing arguments.

4:38The motion filed with a federal judge on Monday alleges that his former lawyer, quote, used an experimental artificial intelligence program to draft the closing argument, ignoring the best arguments and conflating the charge schemes, and then he publicly boasted that the AI program turned hours or days of legal work into seconds. Going further, they noted that the attorney in question has a stake in the company that was used to actually write the closing argument, which the new lawyers contend is a clear conflict of interest. Pretty fascinating argument and will be interesting to see how the federal judge responds to it.

5:08Over in New York City, Mayor Eric Adams is touting what he's calling a first-of-its-kind plan for responsible artificial intelligence use in the New York City government. Now, the phrase is fairly full of jargon, but basically it sounds like what they're trying to accomplish is the development of frameworks that allow city agencies to use AI tools in their work, but while avoiding the challenges of bias, hallucination, and other issues that come with using AI currently. Adams said, While artificial intelligence presents a once-in-a-generation opportunity to more effectively deliver for New Yorkers, we must be clear-eyed about the potential pitfalls and associated risks these technologies present.

5:43I am proud to introduce a plan that will strike a critical balance in the global AI conversation, one that will empower city agencies to deploy technologies that can improve lives while protecting against those that can do harm. Now, unfortunately for Mayor Eric Adams, most of the coverage around this announcement so far has been focused on the fact that there are now robocalls using his voice, speaking in language that he doesn't speak. As The Verge writes, the calls have already triggered alarm over whether the mayor is misleading people. The city reports that there have been quote, thousands of robocalls in Spanish, more than 250 in Yiddish, 160 in Mandarin, 89 in Cantonese, and 23 in Haitian Creole.

6:18Said the executive director of the Surveillance Technology Oversight Project, or STOP, this is deeply unethical, especially on the taxpayer's dime. Using AI to convince New Yorkers that he speaks languages that he doesn't is deeply Orwellian. Now, of course, this is a real Rorschach test. And to some extent, people's reaction probably has as much to do with what they think of Adams as what they think of the principle underneath. On the one hand, it's very clear why there would be concerns about a politician passing themselves off as speaking a bunch of languages that they don't speak. At the same time, one of the great promises of AI is breaking down these linguistic barriers.

6:49Defending himself, Adams said, I got a city to run, and I have to be able to speak to people in languages they understand. Over in the world of lawsuits, best-selling author John Grisham has joined the crowd suing OpenAI and others for wrongly training their models on copyrighted works. And Grisham added the juicy little line, For 30 years I've been sued by everyone else for slander, defamation, copyright, whatever, so it's my turn. Indeed, Grisham used very serious language, saying that the threat from AI cannot be truly appreciated. Finally, one little follow-up to Marc Andreessen's new Techno Optimist manifesto that I read to you guys a couple of days ago.

7:21As I told you, I expected this to generate a lot of conversation, controversy, critique, and we are starting to see that. Very well-respected venture capital writer Dan Primack responded to the piece on Axios saying Mark Andreessen's AI manifesto hurts his own cause. And basically the point that Primack makes is that by choosing not just to invoke what's powerful and good about technology, but that by explicitly calling out a quote literal enemies' list of bad ideas, including sustainability, ESG, stakeholder capitalism, and more, in his view, Andreessen had lost the plot. Primack writes, Right now we're watching what happens when risk management and tech ethics are ignored for the sake of unlimited growth, as told via sworn testimony in the fraud trial of FTX founder Sam Bankman-Fried.

8:05What do you think? I know this has generated a lot of response. If you're interested in talking about the manifesto, come join us on the AI Breakdown Discord. You can find it at bit.ly slash AI Breakdown. I'm super interested in your thoughts. That, however, is going to do it for today's AI Breakdown Brief. I'll be back soon with the main AI breakdown. Are you interested in how two top-of-mind trends, AI and crypto, can work together? If so, I have the perfect podcast recommendation for you. Web3 with A16Z Crypto, the chart-topping show brought to you by venture firm Andreessen Horowitz. Web3 with A16Z Crypto is your definitive resource for the future of the internet, whether you're already building in these spaces or simply curious about what's next.

8:48If you need a place to start, they recently released an excellent episode with Stanford cryptography professor Dan Bonet and former Google Xer Ali Yahya in conversation with host Sonal Choksi about the intersection of AI and crypto. From fighting deepfakes and proving humanity to large language models like ChatGPT, they cover it all. I highly recommend checking it out, especially if you'd like to learn more about how AI and crypto will impact our everyday lives. Beyond Crypto and AI, this show is for creators seeking more ways to truly own their work, for business leaders trying to prepare for the future today, and for innovators exploring trending tech topics.

9:21So go ahead, listen to Web3 with A16Z Crypto wherever you get your podcasts. Welcome back to the AI Breakdown. Today, we are talking about a policy update in AI land, which is very integral to one of the key geopolitical tensions that's shaping the nature of artificial intelligence and how it gets rolled out across the world. That is, of course, the battle between the US and China, and in particular, the Biden administration's attempt to keep advanced semiconductors and AI chips away from the PRC. So today we are going to talk about the new restrictions, how they close loopholes, as well as look at some other implications of this China-US battle and how it's playing out in other parts of the world.

10:03But first, let's do a little bit of background. It has been about a year since the Biden administration first started imposing export restrictions on semiconductors and chip-making equipment to China. On October 7, 2022, the White House announced a set of what the New York Times called sweeping new limits on the sale of semiconductor technology to China. Now, the restrictions came technically from the Commerce Department, but they were clearly not just focused on Chinese private industry. Indeed, a much more specific focus was an attempt to slow down Chinese military development, who, again, as the New York Times put it, used supercomputing to model nuclear blasts, guide hypersonic weapons, and establish advanced networks for surveilling dissidents and minorities.

10:41Basically, these rules made it so that companies couldn't supply advanced chips or chip-making equipment unless they received a special license. The U.S. government said at the time as well that most of those licenses would be denied. However, there were some workarounds. One of them was that Chinese company affiliates that were located in other countries didn't face those restrictions, and in general, bilateral trade relationships through countries especially in the Middle East were also seen as an area of chip leakage. Earlier this year, we saw additional actions from the White House to try to shore up some of these holes.

11:10In August, it was reported that the administration had blocked, had asked companies including NVIDIA and AMD to stop selling chips to certain parts of the Middle East due to concerns that those chips would ultimately end up with China. In that same month, President Biden also issued an executive order that banned investment in certain Chinese sectors. Basically, the executive order created two categories of investments, those that will be prohibited entirely and those that will require notification. Where those lines are drawn in areas including quantum, semiconductor, and AI industries is subject to a rulemaking process being run by the Treasury Department.

11:41And the administration tried to make it clear that this wasn't meant as a way to tamp down the economic relationships between the US and China, but to focus on the national security threat. Said an administration official, I want to be clear, this is a national security action, not an economic one. This executive order is aimed at narrowly protecting our national security interests. And yet still, there has been a ton of bluster that even more export restrictions were coming. Part of the delay has been that there has been real tension with the U.S. technology industry and these companies that make a ton of revenue from selling to China.

12:11Once again, the framing is the same. Commerce Secretary Gina Raimondo told reporters that the goal is to limit China's quote, access to advanced semiconductors that could fuel breakthroughs in artificial intelligence and sophisticated computers. She pointed out again that these chips are critical to Chinese military applications. And so what are the actual changes? Well, one, in the wake of last year's restrictions, some companies, including NVIDIA, had developed specifically less powered chips designed for that Chinese market. Those chips are now banned without a license. What's more, the government has now designated what they call a gray zone of chips that are just below banned power thresholds and which will now require notification to the government who has the power to deny their sales.

12:48U.S. chip industry group the Semiconductor Industry Association was not a fan of the new rules, saying in a statement, Overly broad, unilateral controls risk harming the U.S. semiconductor ecosystem without advancing national security as they encourage overseas customers to look elsewhere. Chinese officials have criticized the controls as violating international trade rules, as well as accusing them of being an attempt to constrain China's economic growth. When discussing the potential of these rules earlier this year, NVIDIA's CEO, Colette Kress, said that while the restrictions of sales on the less powerful chips to China wouldn't necessarily have an immediate financial impact because the demand would be filled by other places, she said that long-term, restrictions on China will, quote, result in a permanent loss of opportunities for the U.S.

13:27industry to compete. Now, in addition to just placing more restrictions on how powerful chips that can be exported are, they've also tried to close the loophole around overseas subsidiaries. Basically, these export restrictions now include overseas subsidiaries of Chinese companies, as well as 21 additional countries. Going even farther, the Commerce Department said it's launching a rulemaking process to also restrict China's ability to access AI compute via the cloud. Yet some people are calling for even more stringent actions. What's clear is that this is going to be a continued battle, and it's one that will shape the way the industry evolves.

13:59Now, reinforcing the urgency depending on your point of view, on the same day that this was announced, Chinese tech giant Baidu also announced its new Ernie 4.0 platform with claims that it can perform at the same level as GPT-4. At the company's annual flagship event, CEO Robin Li said that the Ernie bot is quote, not inferior to any aspect of GPT-4. Li showed off a number of different use cases, generating a commercial for a car, solving complicated math problems, creating a plot for a martial arts novel, etc. Now, this tool is not available to the public, so right now we're just taking the company's word for it.

14:31But some analysts are still taking note. Charlie Dye, the Vice President and Research Director of Technology at Forrester, called Baidu, quote, the first vendor in China to claim it can perform at GPT-4 levels. He said, We need more benchmarking evidence to prove it, but I'm cautiously optimistic that this is China's GPT-4 moment, given its long-term investment in AI and machine learning. Now when it comes to market reactions, Chinese investors did not appear particularly impressed. Baidu shares were actually down 1.4 % following the presentation. Now one more interesting dimension of the China AI story brings us back over to the UK.

15:03As you well know, the UK government under Rishi Sunak has set out to become a leader in AI policy and AI safety. As part of that, they are hosting a big AI safety summit on November 1st and 2nd, and last month faced a lot of backlash when they invited China to that event. Now, this wasn't just some generic anti-China sentiment. The reaction was specifically due to the fact that UK police had just recently arrested a parliamentary aide on suspicion of spying for China. This Chinese spying scandal was all over the UK news, the exact same time as they were being extended an invite to this global summit.

15:35In response, influential MP Ian Duncan Smith had called on the government to ban China from the summit, but Sunak's government said that it was not going to rescind that invite. A few days later, it did come out that the government was considering banning Chinese officials from half of the event, suggesting that one day would be more general conversations while another day would be more sensitive conversations, but ultimately nothing much came of that. This week, however, we did get some updates about the event. Specifically, we got more information about what the schedule was actually going to look like.

16:01On the first day, there's an opening plenary, followed by a set of roundtable discussions on topics like risks to global safety from frontier AI misuse, risks from unpredictable advances in frontier AI capability, risks of loss of control over Frontier AI, risks of the integration of Frontier AI into society. Basically, you can tell that these guys are not just dealing with questions of bias or hallucination, but are actually thinking about X-risk type issues. Next up is another set of roundtable discussions on improving Frontier AI safety. What should Frontier AI developers do to scale responsibly?

16:32What should national policymakers do in relation to the risk and opportunities of AI? What should the international community do and what should the scientific community do? The day concludes with a panel discussion on AI for good. Now, if that is the content-y day, day two is the big talking day. The only information that is given is this. The Prime Minister will convene a small group of governments, companies, and experts to further the discussion on what steps can be taken to address the risks in emerging AI technology and ensure it is used as a force for good. In parallel, UK Technology Secretary Michelle Donilon will reconvene international counterparties to agree to next steps.

17:05Adding some stress to that whole process, the BBC reported this week that German Chancellor Olaf Schultz could decide to specifically snub the event. BBC writes, German Chancellor Olaf Schultz may turn down his invitation to a major UK summit on artificial intelligence. While no guest list has been published of an expected 100 participants, some within the sector say it's unclear if the event will attract top leaders. Though no final decision has been made, it is now seen as unlikely that the German Chancellor will attend. That could spark concerns of a domino effect with other world leaders, such as the French President Emmanuel Macron, also unconfirmed.

17:36The BBC piece points out that it is also not clear if China has decided to accept that invitation or not. So my friends, we will wrap there. So much geopolitical intrigue around this little technology industry we have here. And given that this summit is just two weeks away, I think you can count on a lot more in the coming days. Appreciate you guys listening or watching as always. And until next time, peace.

18:03Thank you.