TheChek News Flow

The Indian Ocean nation of the Maldives will ban Israelis from the luxury tourist hot spot, the office of the president said Sunday, announcing a national rally in “solidarity with Palestine”.

The Maldives, a tiny Islamic republic of more than 1,000 strategically located coral islets, is known for its secluded sandy white beaches, shallow turquoise lagoons and Robinson Crusoe-style getaways.

President Mohamed Muizzu has “resolved to impose a ban on Israeli passports,” a spokesman for his office said in a statement, without giving details of when the new law would take effect.

Muizzu also announced a national fundraising campaign called “Maldivians in Solidarity with Palestine”.

The Maldives had lifted a previous ban on Israeli tourists in the early 1990s and moved to restore relations in 2010.

However, normalisation attempts were scuttled following the toppling of president Mohamed Nasheed in February 2012.

Opposition parties and government allies in the Maldives have been putting pressure on Muizzu to ban Israelis, as a sign of protest against the Gaza war.

Official data showed the number of Israelis visiting the Maldives dropped to 528 in the first four months of this year, down 88 percent compared to the corresponding period last year.

The war was sparked by Hamas’s unprecedented October 7 attack on Israel, which resulted in the deaths of 1,189 people, mostly civilians, according to an AFP tally based on Israeli official figures.

Militants also took 252 hostages, 121 of whom remain in Gaza, including 37 the army says are dead.

Israel’s retaliatory offensive has killed at least 36,379 people in Gaza, mostly civilians, according to the Hamas-run territory’s health ministry.

Silicon Valley chipmaker seeks to accelerate pace of development with new ‘Rubin’ processors

Nvidia announced the next generation of its artificial intelligence processors on Sunday in a surprise move less than three months after its most recent launch.

At the Computex conference in Taipei, the chipmaker’s chief executive Jensen Huang unveiled “Rubin”, the successor to its “Blackwell” chips for data centres, which are currently in production after being announced in March.

The unexpected move to reveal its next wave of products before Blackwell has even started shipping to customers shows how the world’s most valuable chipmaker is racing to entrench its dominance of AI processors, which has propelled it into the ranks of the world’s most valuable companies.

“A new computing age is starting,” Huang said, as Nvidia also unveiled new AI chip deals with PC makers.

Rubin is set to start shipping in 2026 and promises improved power efficiency, as the Silicon Valley-based company attempts to address concerns that Big Tech’s expansion of AI data centres is putting strain on the energy grid in some regions.

The announcement was light on detail but Huang said Nvidia was working on a “one-year rhythm” of building new AI platforms.

Nvidia’s pace of innovation has taken on outsized importance to the wider stock market, as traders bet on whether the huge AI-driven rally in a handful of Big Tech companies can continue.

The chipmaker added around $350bn to its market capitalisation after it reported surging revenue growth, and the company is closing in on Apple to become the second most valuable US company after Microsoft.

While Nvidia today sells the majority of the AI chips needed to train large language models, such as OpenAI’s GPT, the company faces growing competition from AMD and Intel, as well as bespoke chips developed by cloud computing providers including Amazon, Google and Microsoft.

Nvidia’s Blackwell chip is being rolled out barely a year after its current generation “Hopper” chips were unveiled.

The company on Sunday also announced a new Vera Arm-based central processing unit, as Nvidia seeks to make more of the chips that go into AI data centres.

CPUs, a market dominated by Intel and AMD, are traditionally the workhorse of any computer but Huang is attempting to reshape the server market around its AI chips, as artificial intelligence takes a growing share of data centre workloads.

Nvidia started out more than 30 years ago making graphics processing units, which acted as a sidekick to Intel CPUs in video gaming PCs. But more than 15 years ago, Huang realised that the technology inside its GPUs was also suited to other data intensive computing tasks, such as AI.

The company is now trying to boost its PC chip business by capitalising on its dominance in AI chips for data centres.

Huang on Sunday also announced deals with two PC makers, Asus and MSI, that will launch machines using Nvidia’s GeForce RTX graphics processing units to support a range of AI tasks, from running digital assistants to video editing and coding.

“Your future laptop will be constantly helping you in the background,” Huang said. “The PC will run apps that are enhanced by AI, from writing, photo editing, to digital humans that are AIs,” said Huang.

Nvidia did not specify when the Asus and MSI laptops will go on sale.

A range of PC makers and component providers are expected to use the Computex event to make announcements to position themselves as beneficiaries of an expected “AI PC” wave.

Microsoft recently unveiled a string of AI-enhanced PCs and tablets fitted out with its Copilot assistant tool, powered by Qualcomm’s chips, which will begin to launch later this month. Microsoft has said it expects to include the Nvidia chips and AMD’s Radeon graphics chips in its PCs in future. 

PC sales have slumped since the pandemic but analysts expect that when demand revives companies will increasingly opt for AI PCs embedded with powerful chips to run AI applications, rather than relying solely on the cloud.

“AI PCs will bring the most exciting innovation to the PC industry in the last two to three decades, since the creation of the World Wide Web in the late 1980s,” Morgan Stanley analysts wrote in a note last month.

They said that running AI applications on devices would be cheaper and more flexible than on the cloud and would also have benefits for data privacy. AI PCs will make up about 65 per cent of PC shipments by 2028, up from 2 per cent this year, Morgan Stanley predicted.

Media tycoon weds molecular biologist Elena Zhukova at his Bel Air vineyard

Rupert Murdoch married retired molecular biologist Elena Zhukova on Saturday, a family spokesperson said, the fifth wedding for the media mogul.

Murdoch, 93, handed over the reins of his two companies Fox and News Corp to his elder son Lachlan in November, resolving a closely watched succession saga that has fuelled tabloid gossip and television dramas for decades.

His wedding to Zhukova, 67, is not expected to alter the family trust which controls about 40 per cent of the voting shares for both companies.

Murdoch and Zhukova married at the businessman’s California vineyard. Zhukova is the mother of philanthropist Dasha Zhukova Niarchos, whose first marriage to Russian oligarch Roman Abramovich ended in divorce.

Australian-born Murdoch has six children from three of his previous marriages who share as beneficiaries of the trust, albeit with differing voting privileges. Last year, Murdoch was briefly engaged to former police chaplain Ann Lesley Smith, though the wedding was called off. He divorced the former model Jerry Hall in 2022 after six years of marriage.

Despite stepping back from official duties across his media empire, Murdoch is still seen as its figurehead.

Last month, Prince Harry failed in his attempt to bring legal action against Murdoch directly as part of his lawsuit, with other politicians and celebrities, over alleged phone-hacking claims against News Group Newspapers, part of the News Corp empire which publishes The Sun and used to publish the now-defunct News of the World.

Lachlan Murdoch, meanwhile, has been contending with flagging viewership and advertisers at Fox News through the US presidential primary season.

He told analysts last month that he was “disappointed” by trends in the spring but remained confident the 2024 election would ultimately boost the group’s local television networks through the remainder of the year.

Positive results from the ADRIATIC Phase III trial showed AstraZeneca’s IMFINZI® (durvalumab) ?demonstrated statistically significant and clinically meaningful improvements in the dual primary endpoints of overall survival (OS) and progression-free survival (PFS) compared to placebo for patients with limited-stage small cell lung cancer (LS-SCLC) who had not progressed following standard-of-care concurrent chemoradiotherapy (cCRT).

These results will be presented today during the Plenary Session at the 2024 American Society of Clinical Oncology (ASCO) Annual Meeting (abstract #LBA5).

Results from the planned interim analysis showed IMFINZI reduced the risk of death by 27% versus placebo (based on an OS hazard ratio [HR] of 0.73; 95% confidence interval [CI] 0.57-0.93; p=0.0104). Estimated median OS was 55.9 months for IMFINZI versus 33.4 months for placebo. An estimated 57% of patients treated with IMFINZI were alive at three years compared to 48% on placebo. IMFINZI also reduced the risk of disease progression or death by 24% (based on a PFS HR of 0.76; 95% CI 0.61-0.95; p=0.0161) versus placebo. Median PFS was 16.6 months for IMFINZI versus 9.2 months for placebo. An estimated 46% of patients treated with IMFINZI had not experienced disease progression at two years compared to 34% on placebo.

The OS and PFS benefits observed were generally consistent across key prespecified patient subgroups including age, sex, race, disease stage1 at diagnosis, prior radiation and whether patients received prophylactic cranial irradiation. - Source TradeTheNews.com

- GILD and RCUS announced new data from Cohort B of ARC-9, a Phase 1b/2 study evaluating the safety and efficacy of etrumadenant, a dual A2a/b adenosine receptor antagonist, plus anti-PD-1 monoclonal antibody zimberelimab, FOLFOX chemotherapy and bevacizumab (EZFB) in third-line metastatic colorectal cancer (mCRC). These results will be presented today during an oral session at the 2024 American Society of Clinical Oncology (ASCO) Annual Meeting by Zev A. Wainberg, M.D., MSc, Co-Director of the GI Oncology Program at University of California Los Angeles and a principal investigator of the ARC-9 trial (Abstract 3508).

Cohort B of ARC-9 randomized 112 patients with comparable baseline characteristics between two arms: EZFB or regorafenib. At the time of data cut-off (November 13, 2023) median follow-up was 20.4 months. Patient baseline characteristics were similar to those of third-line patients who have progressed on oxaliplatin- and irinotecan-based regimens in mCRC1. OS and PFS were consistently longer in the EZFB arm versus regorafenib, in all sub-groups analyzed, including in patients with liver metastases.

The EZFB regimen had a safety profile consistent with the known safety profiles of each individual molecule to date, without unexpected toxicities. A higher percentage of patients treated with regorafenib (17%) had a treatment emergent adverse event (TEAE) leading to discontinuation of all study drugs than those treated with EZFB (5%). A lower percentage of patients experienced Grade =3 TEAEs attributed to etrumadenant or zimberelimab versus regorafenib (23.0% vs 25.7%).

Etrumadenant and zimberelimab are investigational molecules. Neither Gilead nor Arcus has received approval from any regulatory authority for any use of these molecules, and their safety and efficacy for the treatment of colorectal cancer have not been established.

Last week, China launched a $48 billion fund to build a world-beating semiconductor industry, and surrounded Taiwan in a military war game.

“If you want to be scared in a geopolitical way,” says semiconductor investor Paul “Chip” Schorr, “if China were to take over Taiwan, you could still build a toaster oven, because the chips you need are available from other locations. But advanced servers based on Nvidia chips? Not so much.”

One of the longest active private-equity specialists in technology, Schorr lined up $120 million from the U.S. Commerce Department in May to help expand Polar Semiconductor’s factory in Minnesota, which makes power management chips for auto and aerospace customers.

The federal contribution is part of the $53 billion that’s aimed at returning chip production to American shores, under the Biden Administration’s Chips and Science Act.

“What the Chips Act is doing is leveling the playing field,” Schorr says. Chinese factories may be less burdened by things like environmental rules and building codes. The Chips Act money helps U.S. factories match China’s cost profile.

Schorr knows the semiconductor business. At Citi’s venture fund in 1997, he arranged one of the chip industry’s first management buyouts. It carved out three units of National Semiconductor and put them into Fairchild Semiconductor. After tripling its profit, Fairchild came public in 1999. Schorr went on to run technology investing at Blackstone and then the J.P. Morgan-affiliated One Equity Partners.

When Schorr was starting in the 1990s, America manufactured all the world’s advanced semiconductors. Now, more than 90% of leading-edge processor chips are made in Taiwan by firms like Taiwan Semiconductor Manufacturing. Nearly all the memory and logic chips that accelerate artificial intelligence are fabricated there, in Korea, or Japan.

Globalization cut costs, but it created unanticipated dependencies. “With Covid, people really suddenly found out where their supply chains were,” Schorr says.

Chip shortages stalled U.S. auto makers. Car prices spiked nearly 30% in 2021 and jump-started today’s inflation.

The next year, Biden signed the Chips Act, with bipartisan support. In February 2023, Commerce Secretary Gina Raimondo began taking applications for the Act’s $53 billion in investment and other incentives.

“Failure is not an option here,” she said at the launch.

The Chips Act awards began this year. Micron Technology, Samsung Technology, and Taiwan Semi each will get $6 billion to build U.S. chip factories. Intel will get more than $8 billion.

Private capital is investing alongside the government. The Polar foundry’s upgrade will be backed by midsize private-equity firms like Prysm Capital and Schorr’s Niobrara Capital. Apollo Global Management has backed other chip fabs.

America has a mixed record with industrial policy subsidies in synthetic fuels and solar power. Schorr is optimistic that the Chips Act is getting it right.

“If you want to feel better about our government, go meet the Chips Act office,” he says. “It is incredibly well staffed, from Gina Raimondo on down.”

China can pour money into chip fabs, but it lacks the drivers of semiconductor success. The leading production process engineers in the world are in the Western ecosystem, says Schorr, and U.S. firms like Nvidia and Advanced Micro Devices lead chip design.

“If you can get the same cost here as you can get there, but you know that the supply chain is safer here, then you’re absolutely going to want to use a fab here,” he says.

AI training data has a price tag that only Big Tech can afford

Data is at the heart of today’s advanced AI systems, but it’s costing more and more — making it out of reach for all but the wealthiest tech companies.

Last year, James Betker, a researcher at OpenAI, penned a post on his personal blog about the nature of generative AI models and the datasets on which they’re trained. In it, Betker claimed that training data — not a model’s design, architecture or any other characteristic — was the key to increasingly sophisticated, capable AI systems.

“Trained on the same data set for long enough, pretty much every model converges to the same point,” Betker wrote.

Is Betker right? Is training data the biggest determiner of what a model can do, whether it’s answer a question, draw human hands, or generate a realistic cityscape?

It’s certainly plausible.

Generative AI systems are basically probabilistic models — a huge pile of statistics. They guess based on vast amounts of examples which data makes the most “sense” to place where (e.g., the word “go” before “to the market” in the sentence “I go to the market”). It seems intuitive, then, that the more examples a model has to go on, the better the performance of models trained on those examples.

“It does seem like the performance gains are coming from data,” Kyle Lo, a senior applied research scientist at the Allen Institute for AI (AI2), a AI research nonprofit, told TechCrunch, “at least once you have a stable training setup.”

Lo gave the example of Meta’s Llama 3, a text-generating model released earlier this year, which outperforms AI2’s own OLMo model despite being architecturally very similar. Llama 3 was trained on significantly more data than OLMo, which Lo believes explains its superiority on many popular AI benchmarks.

(I’ll point out here that the benchmarks in wide use in the AI industry today aren’t necessarily the best gauge of a model’s performance, but outside of qualitative tests like our own, they’re one of the few measures we have to go on.)

That’s not to suggest that training on exponentially larger datasets is a sure-fire path to exponentially better models. Models operate on a “garbage in, garbage out” paradigm, Lo notes, and so data curation and quality matter a great deal, perhaps more than sheer quantity.

“It is possible that a small model with carefully designed data outperforms a large model,” he added. “For example, Falcon 180B, a large model, is ranked 63rd on the LMSYS benchmark, while Llama 2 13B, a much smaller model, is ranked 56th.”

In an interview with TechCrunch last October, OpenAI researcher Gabriel Goh said that higher-quality annotations contributed enormously to the enhanced image quality in DALL-E 3, OpenAI’s text-to-image model, over its predecessor DALL-E 2. “I think this is the main source of the improvements,” he said. “The text annotations are a lot better than they were [with DALL-E 2] — it’s not even comparable.”

Many AI models, including DALL-E 3 and DALL-E 2, are trained by having human annotators label data so that a model can learn to associate those labels with other, observed characteristics of that data. For example, a model that’s fed lots of cat pictures with annotations for each breed will eventually “learn” to associate terms like bobtail and shorthair with their distinctive visual traits.

Experts like Lo worry that the growing emphasis on large, high-quality training datasets will centralize AI development into the few players with billion-dollar budgets that can afford to acquire these sets. Major innovation in synthetic data or fundamental architecture could disrupt the status quo, but neither appear to be on the near horizon.

“Overall, entities governing content that’s potentially useful for AI development are incentivized to lock up their materials,” Lo said. “And as access to data closes up, we’re basically blessing a few early movers on data acquisition and pulling up the ladder so nobody else can get access to data to catch up.”

Indeed, where the race to scoop up more training data hasn’t led to unethical (and perhaps even illegal) behavior like secretly aggregating copyrighted content, it has rewarded tech giants with deep pockets to spend on data licensing.

Generative AI models such as OpenAI’s are trained mostly on images, text, audio, videos and other data — some copyrighted — sourced from public web pages (including, problematically, AI-generated ones). The OpenAIs of the world assert that fair use shields them from legal reprisal. Many rights holders disagree — but, at least for now, they can’t do much to prevent this practice.

There are many, many examples of generative AI vendors acquiring massive datasets through questionable means in order to train their models. OpenAI reportedly transcribed more than a million hours of YouTube videos without YouTube’s blessing — or the blessing of creators — to feed to its flagship model GPT-4. Google recently broadened its terms of service in part to be able to tap public Google Docs, restaurant reviews on Google Maps and other online material for its AI products. And Meta is said to have considered risking lawsuits to train its models on IP-protected content.

Meanwhile, companies large and small are relying on workers in third-world countries paid only a few dollars per hour to create annotations for training sets. Some of these annotators — employed by mammoth startups like Scale AI — work literal days on end to complete tasks that expose them to graphic depictions of violence and bloodshed without any benefits or guarantees of future gigs.

In other words, even the more aboveboard data deals aren’t exactly fostering an open and equitable generative AI ecosystem.

OpenAI has spent hundreds of millions of dollars licensing content from news publishers, stock media libraries and more to train its AI models — a budget far beyond that of most academic research groups, nonprofits and startups. Meta has gone so far as to weigh acquiring the publisher Simon & Schuster for the rights to e-book excerpts (ultimately, Simon & Schuster sold to private equity firm KKR for $1.62 billion in 2023).

With the market for AI training data expected to grow from roughly $2.5 billion now to close to $30 billion within a decade, data brokers and platforms are rushing to charge top dollar — in some cases over the objections of their user bases.

Stock media library Shutterstock has inked deals with AI vendors ranging from $25 million to $50 million, while Reddit claims to have made hundreds of millions from licensing data to orgs such as Google and OpenAI. Few platforms with abundant data accumulated organically over the years haven’t signed agreements with generative AI developers, it seems — from Photobucket to Tumblr to Q&A site Stack Overflow.

It’s the platforms’ data to sell — at least depending on which legal arguments you believe. But in most cases, users aren’t seeing a dime of the profits. And it’s harming the wider AI research community.

“Smaller players won’t be able to afford these data licenses, and therefore won’t be able to develop or study AI models,” Lo said. “I worry this could lead to a lack of independent scrutiny of AI development practices.”

If there’s a ray of sunshine through the gloom, it’s the few independent, not-for-profit efforts to create massive datasets anyone can use to train a generative AI model.

EleutherAI, a grassroots nonprofit research group that began as a loose-knit Discord collective in 2020, is working with the University of Toronto, AI2 and independent researchers to create The Pile v2, a set of billions of text passages primarily sourced from the public domain.

In April, AI startup Hugging Face released FineWeb, a filtered version of the Common Crawl — the eponymous dataset maintained by the nonprofit Common Crawl, composed of billions upon billions of web pages — that Hugging Face claims improves model performance on many benchmarks.

A few efforts to release open training datasets, like the group LAION’s image sets, have run up against copyright, data privacy and other, equally serious ethical and legal challenges. But some of the more dedicated data curators have pledged to do better. The Pile v2, for example, removes problematic copyrighted material found in its progenitor dataset, The Pile.

The question is whether any of these open efforts can hope to maintain pace with Big Tech. As long as data collection and curation remains a matter of resources, the answer is likely no — at least not until some research breakthrough levels the playing field.