Weekly Roundup
DeepSeek breaks IP law, innovation through competition, and the "American news swamp"
At least in the tech world, DeepSeek continues to control the headlines. One of the reasons is that the model employed “distillation” at scale with models they were not given access to. Distillation is when a larger, more complex model trains a smaller one. This process can be done in a variety of ways. In DeepSeek’s case, they basically asked other models thousands of questions and developed their own model based on those answers.
One of the immediate takeaways is the prospect that AI doesn’t need a billion-dollar price tag, something especially attractive to engineers in nearly every country besides America.1 As frustrating as it is for security hawks, sanctions and export controls may have spurred more cost-effective innovation, much like in Huawei’s case, which was forced to diversify its portfolio because of U.S. actions. Capitalism, it seems, found a way. Or perhaps it could just be that the “first mover” in AI was always going to bear the brunt of financial and technical costs.
IP problems
Billions of dollars in market cap are traded on the notion that intellectual property rights are protected and can subsequently be sold, either in products or services or as rights to others. It is unclear what the ease of stealing (or innovating) from large, complex, pre-existing models means for the business plans of U.S. companies. Is the new model patentable? Did it violate existing patents and copyrights? Does it matter that the models it “stole” from those that did the same thing with the original data? DeepSeek’s problems in this area are global.
Chinese AI company DeepSeek has been accused of IP theft, faced privacy inquiries in Europe, and has been the target of an enormous cyberattack. Now, it appears the company has a new headache on its hands: a U.S. trademark conflict.
On Tuesday, DeepSeek filed an application with the U.S. Patent and Trademark Office (USPTO) seeking to trademark its AI chatbot apps, products, and tools. But it was a hair too late. Thirty-six hours earlier, another firm had filed for the trademark “DeepSeek”: a Delaware-based company going by the name “Delson Group Inc.”
An interesting overlay on this problem is a recent announcement from the U.S. Copyright Office that AI outputs are copyrightable, provided “where a human author has determined sufficient expressive elements.” The law may not matter a whole lot in these early days of model training, but instead, it may just be who can amass market share based on good old-fashioned capitalist competition.
Innovation through competition
Traditionally, antitrust legislation has been the guarantee of a competitive market.
Based on historical patterns of cases gone moot (Ref: antitrust litigation against Microsoft in the 1990s), it’s clear the US government has been reaching for the wrong knobs and levers to spur innovation. Instead of lawsuits and regulation, it could increase funding for basic scientific research and support startup ecosystems with more financial incentives and tax benefits. The list could go on.
The Semafor article brings up a good point if read through the lens of prioritizing resources, but the Biden Administration gave a slew of financial incentives as well as started or followed through on antitrust lawsuits. The new Administration also began its first antitrust suit this week, suing to block HP from acquiring Juniper Networks.
Startups always need money, but the regulatory hurdles are high to export outside the U.S. or sell their products and services to the government. Access to innovating-spurring IP without the threat of a lawsuit could distill vital technology throughout the U.S. market and force companies to innovate better products. Releasing patents would have to be done strategically, or else there would be no incentive to compete in the first place. Still, certain technologies may certainly warrant it, especially if we are in an AI arms race. If so, we want as many people in the fight as possible.
In other regulation news:
Meta in Talks to Reincorporate in Texas or Another State, Exit Delaware
Musk’s X Adds Nestlé, Colgate, Shell, Other Brands to Ad-Boycott Suit
AI systems with ‘unacceptable risk’ are now banned in the EU. Something to watch is how the EU defines, in practice, “AI that manipulates a person’s decisions subliminally or deceptively.”
News on the news
The “American News Swamp”
Social media platforms operating in the UK should be forced or incentivised to promote content from trusted sources to avoid an “American news swamp”, according to the boss of broadcaster Channel 4.
Alex Mahon said content from respected broadcasters was being “shadow banned” by computer technology that selected what was promoted online, and regulated providers should be given “algorithmic prominence” to prevent this from happening.
Meta settles on Trump. “Meta said on Wednesday that it had agreed to pay President Trump $25 million to settle a 2021 lawsuit he filed over the suspension of his Facebook and Instagram accounts after the Jan. 6 riot at the Capitol.”
Paramount settles on Trump. Paramount is in talks to settle a lawsuit “accusing the company of deceptively editing a “60 Minutes” interview with Vice President Kamala Harris, many legal experts dismissed the litigation as a far-fetched attempt to punish an out-of-favor news outlet.”
The new sleeper cells? New information is available documenting Russian intervention in Romanian elections. “Moreover, according to the documents, ‘almost 800 TikTok accounts created by a “foreign state” in 2016 were suddenly activated last month to full capacity’ backing one of the candidates. Another 25,000 TikTok accounts had become active only two weeks before the first round.’”
The new oligarchy
The expert class, dutifully civil servants, indigo blob, deep-state, or however you want to call them, is out. Tech money is in. Some of the old oligarchs have been fired; some may receive buyouts. Tech money is allegedly sleeping in the White House and has access to high-level secret Treasury files. Regulation will now be conducted based on the judgment of the accelerationist crowd. If you want to know more about them, here are some reading recommendations:
Nate Silver’s On the Edge provides a window into the world from which the “tech bro” comes, what drives their decision-making, and their frustrations with the old oligarchy.
Original sources, such as Thiel’s Oped last month or Marc Andreessen’s Techno-Optimist Manifesto are always valuable,
as are their counter-points.
It might also be helpful to view this change as the by-product of the previous oligarchy’s calcification. After all, Treasury files of the new oligarchy were leaked, and some of the EOs President Trump revoked have been on file since the Johnson Administration. Whether or not you think the previous oligarchy was the right one or not, or perhaps not an oligarchy at all, you are still going to need to understand where new policies will come from. They aren’t as whimsical as you think.
I say “prospect” because it is still unclear how much “further” AI can advance without massive amounts of capital. Consumer demand, for example, will be massive, which means more compute and power, even if that compute is more efficient.