TheChek News Flow

Il avait inventé une arme devenue mythique. Le fabricant autrichien de pistolets Gaston Glock est mort à l'âge de 94 ans après avoir révolutionné l'armement avec ses petits calibres en plastique, a indiqué mercredi son entreprise sur son site internet. «À la mémoire de Gaston Glock, 19.07.1929 - 27.12.23. La perfection continue», a écrit Glock GmbH, accompagnant l'hommage d'un portrait de l'homme d'affaires et d'un bandeau noir.

Gaston Glock avait fondé son entreprise d’armement en 1963 à Deutsch-Wagram, en Autriche. Discret comme il était célèbre, le père du semi-automatique «Glock» avait étudié l'ingénierie mécanique à Vienne, avant de confectionner des prototypes de pistolets. En 1982, il remporte un appel d'offres de l'armée autrichienne en concevant une arme à feu composée en grande partie de matériaux non métalliques

Le «Steve Jobs» des pistolets

Elle est moins chère, plus légère, plus facile à démonter, tire plus de balles que ses concurrentes et l'entreprise Glock GmbH se lance sur le marché mondial. L'entreprise possède aujourd'hui de nombreuses succursales dans le monde entier. Glock est devenue une marque planétaire grâce à son arme de poing devenue une référence.

L’entreprise connaît un grand succès sur le marché américain, et ses pistolets deviennent iconiques portés notamment par le hip-hop et par Hollywood. Bruce Willis vante ses mérites dans Die Hard 2, Tommy Lee Jones dans US Marshals. On le dégaine aussi dans les James Bond. L’entreprise équiperait, en outre, 80% des policiers américains mais aussi les armées américaine ou encore norvégienne. «On peut vraiment comparer Gaston Glock à Steve Jobs lorsqu'il a sorti le premier produit Apple de son garage», déclarait à l'AFP en 2018 Fritz Ofner, réalisateur d'une des très rares enquêtes sur le milliardaire («Weapon of Choice»).

«Clairvoyance visionnaire»

«Avec une clairvoyance visionnaire, Gaston Glock a bâti son entreprise et en a fait un leader mondial avec le Glock Perfection, de renommée internationale. Jusqu'à la fin, il était responsable de l'orientation stratégique du groupe de sociétés Glock et de ses employés», a précisé l'entreprise dans un communiqué. «L'œuvre de toute une vie de l'ingénieur Gaston Glock se poursuivra dans son esprit à l'avenir», a encore commenté l’entreprise.

Son épouse, Kathrin Glock, a également publié une photo d’adieu sur son compte Facebook rendant hommage à son époux disparu.

The New York Times is suing OpenAI and its close collaborator (and investor), Microsoft, for allegedly violating copyright law by training generative AI models on Times’ content.

In the lawsuit, filed in the Federal District Court in Manhattan, The Times contends that millions of its articles were used to train AI models, including those underpinning OpenAI’s ultra-popular ChatGPT and Microsoft’s Copilot, without its consent. The Times is calling for OpenAI and Microsoft to “destroy” models and training data containing the offending material and to be held responsible for “billions of dollars in statutory and actual damages” related to the “unlawful copying and use of The Times’s uniquely valuable works.”

“If The Times and other news organizations cannot produce and protect their independent journalism, there will be a vacuum that no computer or artificial intelligence can fill,” reads The Times’ complaint. “Less journalism will be produced, and the cost to society will be enormous.”

In an emailed statement, an OpenAI spokesperson said: “We respect the rights of content creators and owners and are committed to working with them to ensure they benefit from AI technology and new revenue models. Our ongoing conversations with The New York Times have been productive and moving forward constructively, so we are surprised and disappointed with this development. We’re hopeful that we will find a mutually beneficial way to work together, as we are doing with many other publishers.”

Generative AI models “learn” from examples to craft essays, code, emails, articles and more, and vendors like OpenAI scrape the web for millions to billions of these examples to add to their training sets. Some examples are in the public domain. Others aren’t, or come under restrictive licenses that require citation or specific forms of compensation.

Vendors argue fair use doctrine provides a blanket protection for their web-scraping practices. Copyright holders disagree; hundreds of news organizations are now using code to prevent OpenAI, Google and others from scanning their websites for training data.

The vendor-outlet conflict has led to a growing number of legal battles, The Times’ being the latest.

Actress Sarah Silverman joined a pair of lawsuits in July that accuse Meta and OpenAI of having “ingested” Silverman’s memoir to train their AI models. In a separate suit, thousands of novelists, including Jonathan Franzen and John Grisham, claim OpenAI sourced their work as training data without their permission or knowledge. And several programmers have an ongoing case against Microsoft, OpenAI and GitHub over Copilot, an AI-powered code-generating tool, which the plaintiffs say was developed using their IP-protected code.

While The Times isn’t the first to sue generative AI vendors over alleged IP violations involving written works, it’s the largest publisher involved in such a suit to date — and one of the first to highlight potential damage to its brand through “hallucinations,” or made-up facts from generative AI models.

The Times’ complaint cites several cases in which Microsoft’s Bing Chat (now called Copilot), which is underpinned by an OpenAI model, provided incorrect information that was said to have come from The Times — including results for “the 15 most heart-healthy foods,” 12 of which weren’t mentioned in any Times article.

The Times makes the case, also, that OpenAI and Microsoft are effectively building news publisher competitors using The Times’ works, harming The Times’ business by providing information that couldn’t normally be accessed without a subscription — information that isn’t always cited, sometimes monetized and stripped of affiliate links that The Times uses to generate commissions, moreover.

As The Times’ complaint alludes to, generative AI models have a tendency to regurgitate training data, for example reproducing almost verbatim results from articles. Beyond regurgitation, OpenAI has on at least one occasion inadvertently enabled ChatGPT users to get around paywalled news content.

“Defendants seek to free-ride on The Times’s massive investment in its journalism,” the complaint says, accusing OpenAI and Microsoft of “using The Times’s content without payment to create products that substitute for The Times and steal audiences away from it.”

Impacts to the news subscription business — and publisher web traffic — is at the heart of a tangentially similar suit filed by publishers earlier in the month against Google. In the case, the defendants, like The Times, argued Google’s GenAI experiments, including its AI-powered Bard chatbot and Search Generative Experience, siphon off publishers’ content, readers and ad revenue through anticompetitive means.

There’s credence to publishers’ assertions. A recent model from The Atlantic found that, if a search engine like Google were to integrate AI into search, it’d answer a user’s query 75% of the time without requiring a click-through to its website. Publishers in the Google suit estimate they’d lose as much as 40% of their traffic.

That doesn’t mean they’ll be successful in court. Heather Meeker, a founding partner at OSS Capital and an adviser on IP matters including licensing arrangements, compared The Times’ example of regurgitation to “using a word processor to cut and paste.”

“In the complaint, The New York Times gives an example of a ChatGPT session about a 2012 restaurant review,” Meeker told TechCrunch via email. “The prompt for ChatGPT is ‘What were the opening paragraphs of his review?’ The next prompts then repeatedly ask for ‘the next sentence.’ Teasing a chatbot into reproducing input is not a sensible basis for copyright infringement … If the user intentionally makes the chatbot copy, that’s the user’s fault. And that’s why most [lawsuits like this] will probably fail.”

Some news outlets, rather than fight generative AI vendors in court, have chosen to ink licensing agreements with them. The Associated Press struck a deal in July with OpenAI, and Axel Springer, the German publisher that owns Politico and Business Insider, did likewise this month.

In its complaint, The Times says that it attempted to reach a licensing arrangement with Microsoft and OpenAI in April but that talks weren’t ultimately fruitful.

A soft landing for the U.S. economy in 1995 was followed by a boom. Could it happen again?

Talk of a soft landing for the U.S. economy has investors dreaming of a 1990s-style boom soon afterward. A lot would have to go right for that to happen, but it isn’t as outlandish as it sounds.

Examples of the economy escaping a tightening cycle from the Federal Reserve without entering a recession are few. Some cite the Fed’s rate cuts in 2019, though it is impossible to know how that would have ended without the pandemic. The most frequently mentioned example is a series of rate hikes in 1994. They wreaked havoc on markets that year, but were soon followed by the dawn of a golden period for both Wall Street and Main Street.

The effective federal-funds rate rose from around 3% at the start of 1994 to around 6% in March 1995, according to data from the St. Louis Federal Reserve—a three percentage-point move in a little over a year. That looks similar to, though certainly less severe than, the Fed’s tightening from March 2022 through August 2023, which brought the target range on the federal-funds rate from 0% to 0.25% to its current range of 5.25% to 5.5%.

The economy of the mid-1990s slowed in response but stayed well clear of contraction, with real gross domestic product growth bottoming out at a 2.2% year-over-year pace in the fourth quarter of 1995, Fed data show. The Fed made some modest rate cuts in 1995, similar to what economists now expect for 2024. What followed is known to anyone on Wall Street: one of American history’s great booms. The S&P 500 declined 1.5% in 1994, then rallied 34.1% in 1995 and didn’t suffer another down year until the dot-com bubble burst in 2000.

“We could still get a recession late next year, but in that case it would be a case of a policy error by the Federal Reserve for not loosening soon enough,” said Jason Draho, head of Asset Allocation Americas for UBS Global Wealth Management.

If the U.S. does dodge a recession, investors would want to start looking more at “midcycle” stocks that do well when the economy has a runway of growth ahead of it, including things like industrials, regional banks and small-caps, Draho argued in an interview. Indeed, since the Fed’s meeting on Dec. 13, in which policymakers unveiled lower expectations for rates next year, the small-cap Russell 2000 has rallied 9.8%, and the KBW Regional Bank index has risen 10.0%, compared with a 3.0% rise in the S&P 500.

In a recent investor presentation, a team of UBS analysts led by Draho identified several factors that could lead the U.S. economy to a “Roaring ’20s” scenario this decade, with elevated real growth of 2.5% or higher, and controlled inflation of between 2% and 3%. Key are “supply-side” factors like capital investment in factories and infrastructure, and technological advancement, such as in artificial intelligence, that can boost worker productivity. These could increase the economy’s potential to grow without inflation, echoing what happened in the 1990s thanks to computer and internet technology.

“The potential to look like the ’90s is high, but it can also be overstated,” says Steven Blitz, Chief U.S. Economist at TS Lombard. “I think there’s still a lot of negatives in trail that will keep growth relatively low in the near term…In the long-term, it’s a productivity bet.”

Among the differences between then and now: inflation was also held down in the ’90s by globalization, Blitz notes. The current trend toward industrial reshoring is a positive for capital investment in the U.S., but it is also likely inflationary, as it generally means replacing cheaper imports with domestically produced goods.

Similarly, investment in more climate-friendly energy and infrastructure could be inflationary over the near term. It could even drag on productivity if it means investing in redundant energy sources, says Draho. But if those investments eventually yield a more efficient energy system, that would be a positive for productivity in the long term, he adds.

Another difference: A lot of potential good news is already priced in today. In December 1993, the cyclically adjusted price-earnings ratio, a measure developed by economist Robert J. Shiller that compares stock prices to average earnings over the prior 10 years, stood at 21.2, according to data from Shiller. That is not remarkably low, but it compares to an elevated 30.8 in September 2023, the latest month that Shiller has data available.

Worker-productivity data are notoriously volatile and difficult to interpret, but the signs were there in the mid-to-late ’90s that something very good was afoot. The year-over-year change in labor productivity reached 2.6% in the fourth quarter of 1997 from 0.5% in the first quarter of 1994, Fed data show. It then accelerated to 3.7% in the first quarter of 1998.

Today, there are already some indications that worker productivity may have been meaningfully enhanced by pandemic-era advancements in remote work, videoconferencing, at-home shopping and the like. But the data are hard to read: Worker productivity rose 2.4% from a year earlier in the third quarter of 2023, compared with a negative reading two quarters earlier.

Arguably, it is in AI that the prospects for a ’90s-style technology and productivity boom most likely lie today. But AI’s ultimate impact on productivity is hard if not impossible to forecast. “The range of potential outcomes is huge,” says Draho.

Still, if worker productivity numbers keep creeping up, look for investors to get more excited about a potential ’90s redux.

US newspaper alleges tech companies took ‘free-ride’ on millions of articles to help build generative AI chatbots

The New York Times has become the first major US media company to sue OpenAI and Microsoft over their artificial intelligence chatbots, alleging the tech companies have taken a “free-ride” on millions of articles to build the groundbreaking technology.

The newspaper is seeking unspecified billions of dollars in damages from the two companies for “profit[ing] from the massive copyright infringement, commercial exploitation and misappropriation of The Times’s intellectual property”.

The move comes as media companies have grown increasingly concerned that generative AI models — which can spew out humanlike text, images and code in seconds — may have been fed their content during their creation without permission or compensation.

AI groups have said that ingesting and processing vast amounts of information that is available on the open internet constitutes “fair use” under US copyright laws. Publishers fear they will lose traffic, and therefore revenues, as a result of chatbots, such as OpenAI’s hugely popular ChatGPT, summarising their output.

“Defendants’ unlawful use of The Times’s work to create artificial intelligence products that compete with it threatens The Times’s ability to provide that service” of news, analysis and commentary, its lawsuit, which was filed in New York on Wednesday, alleged.

The newspaper claims the two tech companies have sought “to free-ride on The Times’s massive investment in its journalism by using it to build substitutive products without permission or payment”.

OpenAI said: “We respect the rights of content creators and owners and are committed to working with them to ensure they benefit from AI technology and new revenue models. Our ongoing conversations with the New York Times have been productive and moving forward constructively, so we are surprised and disappointed with this development. We’re hopeful that we will find a mutually beneficial way to work together, as we are doing with many other publishers.”

Microsoft did not respond to a request for comment.

Microsoft is OpenAI’s biggest backer after committing up to $13bn to fuel the company’s growth and provide the huge technical infrastructure needed to create its AI models. OpenAI’s GPT technology also underpins Microsoft’s Bing Chat, a feature within the software giant’s search engine.

News publishers around the world have been meeting AI companies including OpenAI, Microsoft and Google for several months in an effort to hammer out deals to license their content.

This month, Germany’s Axel Springer struck a deal with OpenAI worth tens of millions of euros a year to let its AI systems use content from outlets such as Bild, Politico and Business Insider.

The Times’s lawsuit alleges the company has held similar discussions with Microsoft and OpenAI “for months”. “These negotiations have not led to a resolution,” it stated.

The Times’s challenge is the latest in a series of lawsuits filed against OpenAI, alleging copyright infringement. In September, a group of bestselling authors including John Grisham, David Baldacci, Jonathan Franzen and George RR Martin sued the tech group, accusing its algorithms of being engaged in “systematic theft on a mass scale”.

Journalist and writer Julian Sancton filed a similar complaint the following month, and was soon joined by New Yorker writer Jia Tolentino, among others.

While OpenAI’s lawyers have yet to respond to those two suits, they have responded to a proposed class action filed in California, arguing that some of the claims should be dismissed as its model can rely on the “fair use” doctrine. They claimed this doctrine had been interpreted by “numerous courts” to mean that the use of “copyrighted materials by innovators in transformative ways does not violate copyright”.

OpenAI’s lawyers have also pointed to an order in a separate challenge brought against Meta’s AI model in California by comedian Sarah Silverman and writer Ta-Nehisi Coates, among others, in which the court found that the output of the company’s large language model was not “substantially similar” to the books written by the plaintiffs.

Shares in The New York Times Company rose about 1 per cent on Wednesday morning.

Higher interest rates, geopolitical tensions and regulatory challenges have prompted two-year hiatus

Dealmaking sank below $3tn for the first time in a decade in 2023, as a cocktail of higher interest rates and escalating geopolitical tension confounded bankers’ hopes that last year’s lull was a one-off.

About $2.9tn worth of transactions were struck globally this year, data from the London Stock Exchange Group shows, down 17 per cent from 2022. It was the first time since 2008-09 that the value of deals announced fell more than 10 per cent for two consecutive years, LSEG said.

“2023 has clearly been a very slow year, more subdued than we expected when you look at the volume of deals,” said Simona Maellare, global co-head of the alternative capital group at UBS.

Europe showed the sharpest drop, down 28 per cent annually, while the Asia-Pacific region was 25 per cent lower and the US 6 per cent.

Dealmakers have had to contend with challenges on multiple fronts. Mergers and acquisitions had already been in decline following a pandemic-era surge in activity, with regulators taking a more muscular approach and the rapid increase in global interest rates cooling the private equity market.

A pair of mega US energy deals from ExxonMobil and Chevron, each worth in excess of $50bn, lifted transaction volumes in the final months of this year. The value of deals struck in the fourth quarter was 28 per cent higher than the third quarter.

However, Israel’s war with Hamas, which started in October, stopped a more widespread dealmaking revival from taking off.

“The regulatory environment has been tricky through the year,” said Mark Sorrell, co-head of global M&A at Goldman Sachs. “As sentiment was improving, you had the Middle East happen.”

Deals from financial sponsors declined 30 per cent over the past year to $562bn. Advisers said private equity groups had difficulty agreeing to valuations on assets. Brookfield’s shelved plans to sell the holiday resort group Center Parcs for more than £4bn exemplified the difficulty finding investors willing to pay up at a time of higher interest rates and inflation.

Private equity groups are expected to come under more pressure to strike deals next year after a prolonged slowdown in activity, dealmakers said.

This year “the successful exits were by the most courageous sellers with the best assets, and every process was more structured and complex”, said Carsten Woehrn, JPMorgan’s co-head of Europe, Mideast and Africa M&A, and global co-head of the bank’s strategic investor group.

“For next year what’s clear is that there is both a willingness and a need to do deals by sponsors,” he said.

A more stringent attitude by competition authorities to enforcement has also deterred companies from launching bids for rivals. Microsoft’s $75bn deal for gaming company Activision Blizzard survived challenges to close after 21 months of uncertainty, but Adobe’s $20bn takeover of software group Figma was abandoned after EU and UK watchdogs launched probes.

Advisers concede that activity may arrive in the second half of 2024 rather than earlier in the new year.

Global investment banking fees were hurt by the slowdown, dropping 8 per cent compared with last year to $105bn. Fees from M&A fell most sharply, down 26 per cent to $29bn, the lowest level since 2016.

Goldman Sachs held the top spot in M&A advisory work, driven by its lead position in the US. Morgan Stanley and JPMorgan came in second and third, leading in Asia and Europe respectively.

Proposal would accelerate preparations in time for a February summit to show solidarity with Ukraine

The US has proposed that working groups from the G7 explore ways to seize $300bn in frozen Russian assets, as the allies rush to agree a plan in time for the second anniversary of Moscow’s full-scale invasion of Ukraine.

While no decisions have been taken and the issue remains hotly debated inside European capitals, the acceleration of work on confiscating Moscow’s assets for Ukraine highlights its rising importance for the west.

The topic was discussed this month by both G7 finance ministers and their deputies, according to people briefed on the calls, which touched on how to develop such a policy and assess the risks involved.

The US, backed by the UK, Japan and Canada, has proposed moving forward with the preparatory work so the options would be ready for a potential meeting of G7 leaders around February 24, the date of Vladimir Putin’s 2022 offensive on Kyiv.

The three working groups proposed by Washington would examine the legal issues around confiscation; the method of applying such a policy and mitigating risks; and options for how to best channel the support to Ukraine.

Germany, France, Italy and the EU have expressed some reservations, and the need to carefully assess the legality of confiscating Moscow’s assets before decisions are taken. Several European ministers also stressed the need to maintain high levels of secrecy over the work, according to accounts of the meeting.

Various options are being explored in western capitals, ranging from directly confiscating and spending the Russian central bank assets, to tapping the proceeds from the frozen assets or using them as collateral for loans.

The EU has so far stopped short of seizing the Russian assets themselves, instead exploring ways to skim off profits generated for financial institutions such as Euroclear, where €191bn in sovereign assets are held.

Washington has so far not publicly backed seizing Russian assets. But the US privately circulated a discussion paper this year within the G7 suggesting seizures of Moscow’s frozen assets would be lawful as “a countermeasure to induce Russia to end its aggression”.

But Europe, where the majority of the assets are held, is much more wary, fearing the possible implications for financial stability as well as retaliatory action from Russia.

Italy, which takes over the G7 presidency in 2024, is among those worried about potential retaliation on its companies active in Russia, something that Moscow has already threatened to do. Russia has also warned it would cease diplomatic relations with the US in response to any asset confiscation.

The EU, UK and France also stressed that the money would not be readily available, and insufficient to cover Ukraine’s reconstruction needs, and that seizing the assets should not be at the expense of providing financial support to Kyiv in 2024. 

Some ministers are concerned that the debate over seizures will imply there is an alternative to orthodox funding packages for Ukraine, which have stalled through opposition in the US Congress and because of Hungary’s refusal to back an EU deal.

But the push to seize Russian sovereign assets reflects a shared desire to show Moscow that it would not be able to outlast western resolve to aid Ukraine, both economically and militarily.

The G7 was able to overcome differences among its members several times on economic measures against Russia over the past two years, including on the initial sweeping sanctions package, and on setting a price cap on Russian oil.

China’s generative artificial intelligence startup sector is going through the “qualifying round,” according to Kai-Fu Lee, a venture capitalist and former president of Google China, who earlier this year founded 01.AI, a Beijing-based startup developing large language models, which last month raised a funding round at a $1 billion valuation.

The biggest internet companies in China, such as Alibaba, Tencent, Baidu and ByteDance, as well as dozens of startups, are all developing their own LLMs, in what local media have described as “a war of 100 models.” The Chinese government blocks OpenAI, Google and other U.S. internet services, so domestic firms are vying for AI supremacy. In an interview, Lee said companies are in the phase of trying to prove they have the technology to build a high-quality model. Those who pass that test will move on to the next phase, which will be about how to grow revenue and eventually generate profit.

“In China, we’ll eventually have a few big winners, a few decent exits, and most will end up either falling by the wayside or pivoting to something less ambitious,” like building applications and solutions for specific industries, Lee predicted.

His startup, founded in March, now has a team of more than 100 employees, all based in mainland China, mostly Beijing. Last month, 01.AI launched its first open-source LLM, Yi-34B, but the company won’t be relying on that model for future revenue. Instead, its business plan is to sell proprietary LLMs, mainly to customers in China. The startup is currently developing a new proprietary model with more than 100 billion parameters, according to Lee.

The startup faced some controversies last month after Yi-34B quickly climbed on top of the Hugging Face leaderboard for open-source LLMs. Developers’ inquiries revealed that Yi-34B had used Meta Platforms’ Llama open-source AI model without mentioning Llama. The drama ended as 01.AI renamed portions of Yi-34B to give credit to Llama and Lee apologized for the oversight. For more on that, see here.

Lee talked to The Information in a video call from his office in Beijing, where he discussed how 01.AI is coping with U.S. government chip export restrictions and global opportunities for Chinese AI firms. Here’s an edited version of the Q&A.

The Information: China currently has dozens of domestic competitors, if not more, developing LLMs. What will happen next?

Lee: I think China has seen this kind of situation many times before, with Groupon copycats, bicycle rental apps and, in the deep tech space, computer vision and speech recognition. When computer vision was proven to have made a breakthrough, everybody in China jumped in with every possible application, but most of them didn’t survive. China is a very, very competitive market, maybe even more so than the U.S.

The current competition in China is still the qualifying round. And the first test is: Which one of these 100 companies can make a high-quality model that creates real value? That means showing good performance because this is still an area where being better in technology makes all the difference when it comes to providing useful content and strategic insights. If you don’t have a good enough model, then it’s more of a toy than a technology that solves real problems.

Those who pass the technology test in the qualifying round will then move on to the next phase, which will be business value. What is your business model? How do you make money? Soon these companies will be evaluated on the basis of their profit and loss statements, and investors will be asking the same questions that they ask cloud providers, enterprise software companies and consumer apps. And if companies can’t answer those questions, their growth will come to an end. In the U.S., OpenAI has already shown that it has a world-leading technology and it can also generate revenue. OpenAI creates enough value so people are building apps on it and paying money for it.

In China, we’ll eventually have a few big winners, a few decent exits, but most will end up either falling by the wayside or pivoting to something less ambitious, like building applications and solutions for specific industries instead of trying to do the pre-trained large model, which will be increasingly expensive over time.

Chinese AI startups and their investors say China will develop its own ecosystem for generative AI models and applications, separately from the U.S. What do you think?

None of us wants a parallel universe. We all prefer to compete on a global scale and let the best company win. It’s more efficient that way. But we don’t control our fate in this case.

There are geopolitical issues in particular. If we wanted to enter the U.S. market, while there is no regulation that says we can’t, I don’t think we will get a lot of business because of the current—in my opinion, unfair—sentiment about Chinese software. So that’s just the practical reality we’ve come to accept.

We are open to business opportunities outside China, but we fully understand that some things are not in the cards. It’s not in the cards to sell our proprietary model to companies in the U.S. They are not going to buy it. We are not going to do anything silly.

China is obviously the big opportunity, but I wouldn’t write off other parts of the world as possible areas Chinese companies can reach. Silicon Valley’s approach is, generally speaking, a one-size-fits-all approach. It has more or less worked for companies like Facebook and Google, and [has] led to the American leadership. But this time it’s different because LLMs are trained on data. And data comes with questions about biases, ideologies and values. And American values aren’t preferred or even accepted in certain countries. China won’t be the only country. I would argue that the Middle East is another region that may want to think about religion and other issues differently. And I think it will lead to countries wanting more control over their models.

I do think there are opportunities to build special models that are different for different countries. This is something that Silicon Valley companies aren’t naturally going to be doing, because they feel that their values are the right values, and they want more people to try it and assimilate to it. And it’s a lot of engineering work to make different models for different markets. So Silicon Valley companies won’t be naturally inclined to build those models. And companies from other parts of the world, including China, may have opportunities to work on such models. Obviously they will have to win the trust of the users and the governments of the countries.

Chinese media have reported that your startup has managed to significantly lower the AI training cost for Yi-34B. How?

We have a very strong infrastructure team. It’s the biggest part of our workforce. One of the things I told my staff earlier is that every time you add a modeling person, you are spreading thin your [graphics processing units]. Every time you add an infrastructure person, you are getting more out of your GPUs. Of course you need a strong modeling team, but, from the get-go, we had a major priority to build a strong infrastructure team.

Infrastructure is the work of unsung heroes. They have to deal with hardware, software and huge amounts of data being transferred. They are simultaneously dealing with GPU, memory and networking, and any one of the three can become a bottleneck.

GPUs have a hard time scaling above several thousand. When you go from 2,000 to 8,000, you can’t just take the software and run with that, because the amount of networking demands change dramatically as you go to larger models and larger data sizes.

The infrastructure team, which has dozens of engineers, is the largest team now at 01.AI. For example, the kinds of work our infrastructure team has done include figuring out how to use FP8—a data format for Nvidia’s H100 chips—end to end, which leads to [a] substantial reduction in computation. The team figured out where to use FP8 and where to use other types and how to convert them in a seamless way. They also dealt with questions like what networking protocol we should use, how to optimize the compiler and how to deal with failing GPUs. GPUs actually fail pretty frequently. When a GPU fails, can you hot-plug? Unplug? We are still working on that. If your training stops for an hour because a [single] GPU fails out of a thousand-GPU cluster, to be able to hot-swap it will save you an hour a day. These little things will add up.

Another related thing is elastic training. That means if you have a cluster of 2,000 H100 chips, and you need 500 for a separate task, can you remove them between checkpoints and can you add them back later? These [tasks] are not what AI researchers have been trained to do. They are more related to networking engineers.

If the development of LLM is rocket science, the rocket will never fly without the engineers. SpaceX didn’t succeed just because of a bunch of researchers. It succeeded because it had a ton of very complex, intricate engineering.

The U.S. has restricted the exports of advanced semiconductor technology to China, including key Nvidia chips. How are you coping with this?

I’ve publicly said we have enough chips to last for 18 months. That’s basically the chips we acquired before the restrictions. We are definitely studying hard how to use Chinese chips. It’s not easy. It’s not fun. Programming them is not something we’re experienced at. But if that’s what we gotta do, that’s what we gotta do.

Nvidia has exceptionally good chips, but there could be an argument that a simpler chip could do the job at a much lower cost. But a major factor behind Nvidia’s strength is the whole ecosystem around its CUDA software libraries that makes programming easy. Engineers would basically revolt if you forced them to use non-Nvidia chips because they are so much less productive. But we are currently facing that predicament in 18 months, and obviously we have to start much earlier. If we can’t get access to Nvidia chips, we will look for simpler chips that are more focused on transformers, but they will be a pain to program. But if we have no choice, then we have no choice.

But Chinese engineers are known to be able and willing and do an excellent job in dealing with such engineering challenges that are considered laborious grunt work. It’s similar to what I said earlier about the work of our infrastructure team. Learning to program a new nonstandard GPU with very minimal libraries is also grunt work.

Chinese entrepreneurs are tenacious. Chinese engineers are hardworking. They are unafraid of grunt work. That’s the very reason why Meituan built a superior service, and why WeChat is a superior product.

Yes, these are difficult challenges, and you could say they are a waste of time and a lot of people’s energy. But these are the cards we are dealt, so we will do our best to play our hand.

Terran Orbital may be close to receiving a major payment from its biggest customer, CEO Marc Bell announced internally at a company-wide meeting earlier this month. Rivada Space Networks is in the final stages of closing funding to help fund a mega-constellation to be built by Terran at a cost of $2.4 billion, revenues that currently make up the vast majority of Terran’s backlog.

While Terran is pursuing other lucrative contracts that could comprise billions in work, its $2.4 billion contract with Rivada is by far the largest it has secured so far. Earlier this quarter, Terran had to adjust its full-year financial outlook after Rivada delayed paying an incremental $180 million toward that total contract award.

The Terran chair and CEO communicated the news to staff during the December 19 meeting.

“I had dinner with [Rivada CEO] Declan Ganley last week in DC,” Bell told staff during the meeting, a recording of which was obtained by TechCrunch. “He told me they expect to close tomorrow on their funding. He showed me the documents. I saw them, I read them. He texted me this morning and maybe Thursday, Friday now. […] As long as it’s by Christmas, I’ll be happy. Nothing wrong with getting a good Christmas present.”

Rivada, a German subsidiary of U.S.-based Rivada Networks, contracted with Terran to build 300 satellites for the mega-constellation under a $2.4 billion agreement in February of this year. Rivada has a separate deal with SpaceX to launch the satellites starting in April 2025.

Of course, Christmas has come and gone and neither company has made any public announcements about the financing. TechCrunch reached out to both for comment and neither responded by press time. Bell said later in the meeting that even if Rivada closes the funding, the two entities would need to make a modification on their contract, which could delay payment terms and public announcements.

“He’s being very transparent to me and so I have no reason not to believe him,” Bell said of Ganley. “But unfortunately, he’s not the one writing the check, somebody else is writing the check to him. But if he gets a check, I have to assume we’ll get a check. But we do have to do a contract mod. So that’s the one thing that might hold us up, because we have to do a mod on the contract. […] But we’ll at least get paid on the invoice that’s owed, the $9 million invoice. But I want to get the big check, as everyone else does, because that will dramatically help our share price and everything else in our world.”

In an investor call in November, Bell said that Rivada’s funding and payment delays came as “quite a surprise” to both companies. He added that Rivada’s funding source is “a large sovereign” — presumably a sovereign wealth fund — and that the two companies expect the money to close eventually.

As of November, Terran reported a backlog of future work of $2.6 billion, of which $2.4 billion is from the Rivada contract. Even without the expected milestone payment, Bell told staff that he still expects the company to generate $130 million in revenue this year, a notable increase from the $94 million the company made in 2022.

The company is also pursing other high-value contract opportunities, including with the Space Development Agency’s constellation known as “Proliferated Warfighter Space Architecture.” Terran has already built and delivered satellites for the initial tranche of the constellation, Tranche 0, and is currently building 42 satellite buses for Tranche 1 and will build an additional 32 buses for Tranche 2. Bell said the company will also be going after an additional award for a variant of the Tranche 2 satellites called Gamma, possibly as the prime contractor on that award (for the others, Terran is a subcontractor of prime winner Lockheed Martin).

“We feel very good about Gamma and how we’re going to win. We are contemplating priming Gamma as opposed to just being a sub […] But right now it is even-money odds we may be finally be a prime on these things. And that would be huge. It would change the dynamic. But I haven’t made a decision, we’re going to sit down and talk to Lockheed about it.”

Bell also told staff that the company is still having conversations about taking Terran private, but that the goal would be to “go private, and then take it public again the traditional way and not have this dumbass market cap like we have today,” he said.

Terran Orbital’s stock price has cratered since it went public via SPAC — a reverse merger with a special purpose acquisition company — in March of last year. The company debuted with a stock price of $10.96, but today the shares are trading for around $1.22.