A DeepMind system has invented – or discovered? – an entirely new, and faster, way to multiply certain matrices. That could help GPU algorithms in future.
You can sign up to receive each day’s Start Up post by email. You’ll need to click a confirmation link, so no spam.
On Friday, there’ll be another post due at the Social Warming Substack at about 0845 UK time.
A selection of 9 links for you. No motion capture, please. I’m @charlesarthur on Twitter. Observations and links welcome.
Visions of a technologically ascendent China keep American strategists up at night. They see the contours of a surveillance state implementing the will of President Xi Jinping by algorithmic edict at home and projecting computing power abroad. To erase those contours for good, on October 7th President Joe Biden’s administration announced the most sweeping set of export controls in decades. The new rules cut off people and firms in China from many advanced technologies of American origin, and from products made using these. The list includes chips used for artificial intelligence (ai), software to design advanced chips and the machine tools to manufacture them. Selling such things to China is now barred without explicit permission from America’s government. Rulebreakers risk being cut off from American tech themselves.
The share prices of affected Chinese firms have sunk. China’s biggest producer of memory chips, the state-owned ymtc, has 60 days to allow American officials to inspect its operations for compliance. American companies that sell advanced semiconductor technology to China have also been hit, even as they reel from a deep cyclical slump in demand for their wares. This week it emerged that Intel, America’s chipmaking champion with Chinese sales of $21bn last year, is about to axe thousands of jobs.
Another piece (because this has huge ramifications) from the Center For Strategic and International Studies:
The most important chokepoints in the context of this discussion are AI chip designs, electronic design automation software, semiconductor manufacturing equipment, and equipment components. The Biden administration’s latest actions simultaneously exploit US dominance across all four of these chokepoints. In doing so, these actions demonstrate an unprecedented degree of US government intervention to not only preserve chokepoint control but also begin a new US policy of actively strangling large segments of the Chinese technology industry—strangling with an intent to kill.
And there’s a Twitter thread (on a single page) on the effects this is having in China, which kicks off with this:
To put it simply, Biden has forced all Americans working in China to pick between quitting their jobs and losing American citizenship. Every American executive and engineer working in China’s semiconductor manufacturing industry resigned yesterday, paralyzing Chinese manufacturing overnight.
One round of sanctions from Biden did more damage than all four years of performative sanctioning under Trump.
the mathematician Volker Strassen proved in 1969 that multiplying a matrix of two rows of two numbers with another of the same size doesn’t necessarily involve eight multiplications and that, with a clever trick, it can be reduced to seven. This approach, called the Strassen algorithm, requires some extra addition, but this is acceptable because additions in a computer take far less time than multiplications.
The algorithm has stood as the most efficient approach on most matrix sizes for more than 50 years, although some slight improvements that aren’t easily adapted to computer code have been found. But DeepMind’s AI has now discovered a faster technique that works perfectly on current hardware. The company’s new AI, AlphaTensor, started with no knowledge of any solutions and was presented with the problem of creating a working algorithm that completed the task with the minimum number of steps.
It found an algorithm for multiplying two matrices of four rows of four numbers using just 47 multiplications, which outperforms Strassen’s 49 multiplications. It also developed improved techniques for multiplying matrices of other sizes, 70 in total.
AlphaTensor discovered thousands of functional algorithms for each size of matrix, including 14,000 for 4×4 matrices alone. But only a small minority were better than the state of the art. The research builds on AlphaZero, DeepMind’s game-playing model, and has been two years in the making.
Hussein Fawzi at DeepMind says the results are mathematically sound, but are far from intuitive for humans. “We don’t really know why the system came up with this, essentially,” he says. “Why is it the best way of multiplying matrices? It’s unclear.”
The Strassen algorithm is a bit mindbending in its own right, since it requires creating new matrices, so seems like it’s entailing much more work. But more intriguing is that the humans don’t quite know why this works; it needs to be reverse engineered to understand.
unique link to this extract
In the autumn of 1954 a young and rather daring psychologist called Leon Festinger infiltrated a UFO cult in Minnesota. The cult insiders believed that they would be picked up in a spaceship at midnight on December 21 and transported to a new planetary utopia on the edge of the galaxy. They had sold their possessions and told the local newspaper what was about to unfold.
Festinger was not interested in the prophecy per se. Many such groups have risen up from time to time in different parts of America. Rather, he was fascinated in what would happen after the prophecy failed. Would the cult members admit their folly? Would they go back to their lives? Would they become reacquainted with reality?
In fact the 35-year-old Festinger, who had spent his career examining dogmatic thinking in all its forms, had a different expectation. He thought their convictions would become even more entrenched. It would be too psychologically threatening to admit they were wrong, too mortifying to confront the stares of those who had warned them. Sure enough, as the clock ticked past midnight, the cult members rapidly found an alternative explanation. The planetary timetable had shifted: the spaceship would now come two years later. Within a week they were back out on a recruitment drive.
I mention this research because it offers the only lens through which to make sense of what has unfolded over the past six years of British politics.
Very astute. Let’s see how The Markets like the appointment of Jeremy Hunt – someone very much not in the cult (voted to Remain, backed Rishi Sunak, doesn’t agree with supply-side economics) – this morning.
unique link to this extract
Exclusive: Musk’s SpaceX says it can no longer pay for critical satellite services in Ukraine, asks Pentagon to pick up the tab • CNN Politics
Since they first started arriving in Ukraine last spring, the Starlink satellite internet terminals made by Elon Musk’s SpaceX have been a vital source of communication for Ukraine’s military, allowing it to fight and stay connected even as cellular phone and internet networks have been destroyed in its war with Russia.
So far roughly 20,000 Starlink satellite units have been donated to Ukraine, with Musk tweeting on Friday the “operation has cost SpaceX $80m and will exceed $100m by the end of the year.”
But those charitable contributions could be coming to an end, as SpaceX has warned the Pentagon that it may stop funding the service in Ukraine unless the US military kicks in tens of millions of dollars per month.
Documents obtained by CNN show that last month Musk’s SpaceX sent a letter to the Pentagon saying it can no longer continue to fund the Starlink service as it has. The letter also requested that the Pentagon take over funding for Ukraine’s government and military use of Starlink, which SpaceX claims would cost more than $120m for the rest of the year and could cost close to $400m for the next 12 months.
“We are not in a position to further donate terminals to Ukraine, or fund the existing terminals for an indefinite period of time,” SpaceX’s director of government sales wrote to the Pentagon in the September letter.
Among the SpaceX documents sent to the Pentagon and seen by CNN is a previously unreported direct request made to Musk in July by the Ukrainian military’s commanding general, General Valerii Zaluzhniy, for almost 8,000 more Starlink terminals.
In a separate cover letter to the Pentagon, an outside consultant working for SpaceX wrote, “SpaceX faces terribly difficult decisions here. I do not think they have the financial ability to provide any additional terminals or service as requested by General Zaluzhniy.”
First: if only SpaceX knew someone of immense wealth – perhaps worth billions personally – to whom it could turn for funding. (Though the world’s richest man seems to think Ukraine should just roll over.)
Second: how is this costing SpaceX any money? It puts up the satellites: that’s a capital cost, but one it has to bear to run its business. It has to keep the satellites in position: that’s an operating cost, but one it has to bear to run its business. It has ground stations: that’s an operating cost, etc etc. The bandwidth, though, is essentially free. It only “costs” in that use of bandwidth may be exclusionary to paying users. Though there probably aren’t that many in Ukraine.
unique link to this extract
The Kettle Companion is an assisted living product, that helps those who live apart to stay connected, by illuminating when a loved one activates their kettle at home.
This is signaled through a monitoring plug and communicated via Wi-Fi to a paired Kettle Companion in another user’s home. Additionally, if there is a change in pattern of use, for instance, an elderly parent has not had their habitual morning cup of tea by the usual time, the paired Kettle Companion will illuminate red. A text message alert can also be sent to the owner of this appliance, prompting them to check on their loved one.
Very neat little idea. Nothing about price, but you’d hope it would be.. tolerable?
unique link to this extract
Jeff Horwitz, Salvador Rodriguez and Meghan Bobrowsky:
While Mr. Zuckerberg has said the transition to a more immersive online experience will take years, the company’s flagship metaverse offering for consumers, Horizon Worlds, is falling short of internal performance expectations.
Meta initially set a goal of reaching 500,000 monthly active users for Horizon Worlds by the end of this year, but in recent weeks revised that figure to 280,000. The current tally is less than 200,000, the documents show.
Most visitors to Horizon generally don’t return to the app after the first month, and the user base has steadily declined since the spring, according to the documents, which include internal memos from employees.
By comparison, Meta’s social-media products, including Facebook, Instagram and WhatsApp, together attract more than 3.5 billion average monthly users—a figure equivalent to almost half the world’s population. Horizon is currently reaching less than the population of Sioux Falls, S.D.
Horizon is designed to be a sprawling collection of interactive virtual spaces, or worlds, in which users appearing as avatars can shop, party and work. Yet there are rarely any girls in the Hot Girl Summer Rooftop Pool Party, and in Murder Village there is often no one to kill. Even the company’s showcase worlds, such as Questy’s, a virtual arcade featured in a Super Bowl commercial earlier this year, are mostly barren of users.
According to internal statistics, only 9% of worlds built by creators are ever visited by at least 50 people. Most are never visited at all.
I’m trying to think of ways by which this could really, really turn round for Meta, but honestly can’t. It requires a huge shift in how people work akin to the smartphone, but VR is far, far away from that. We could have holographic TV first.
unique link to this extract
While the updates bringing full-body avatars aren’t expected until 2023, Zuckerberg was clearly seen jumping around in the video, giving everyone an early look at the tech. Or was he?
Anyone who has ever been around—*checks the culture*—any piece of marketing ever made should know by now that not everything is as it seems when a company is trying to sell you something. And in this case, the video Meta showed off was made with some help.
As UploadVR’s Ian Hamilton has since reported, Meta has issued a follow-up statement, which says, “To enable this preview of what’s to come, the segment featured animations created from motion capture.”
Deep down, of course, you all knew this. From vertical slices at E3 to photo tricks shown at Apple events, there are always grains of salt we need to chew on every time a company trying to sell us something that isn’t out yet.
But there’s something especially funny about this in particular, that a project that has spent billions of dollars to look like a Kinect demo—a piece of hardware first shown off in 2009—has ended up with its own dumb feet-related moment.
unique link to this extract
This is clearly an area Zuckerberg is passionate about to a truly painful degree. So far, though, the best use case — the best use case — for even the more credulous believers is meetings. I cannot imagine buying dedicated expensive hardware for meetings, but I am probably not in the right market; two-and-a-half years into working from home and I still have not bought a ring light. Regardless, that sounds pretty dull. Are businesses champing at the bit to have staff sit in a virtual board room instead of just on a call? Is this solving a meaningful problem for them?
Zuckerberg preemptively responded to criticisms like these by reminding everyone that this category is just getting started. But that is a bit of misdirection. Oculus, the virtual reality hardware company Meta bought, was founded in 2012; Meta bought it in 2014. On a technical level, Meta can point to plenty of improvements. But it is much more difficult for anyone to point to clarifications in the concept and purpose of virtual reality. Again, I would be an idiot to argue there are none at all, but this week’s keynote would have been a great time for Meta to illustrate something new and enrich the story. So far, it does not have legs.
VR has been around for ages: I first tried it out in the early 1990s, on games, though it was also being tried for some less trivial applications. The fact that it keeps not getting traction suggests to me that it’s like 3D TV or films: a technological breakthrough that completely fails to enthral users.
unique link to this extract
Alaska snow crab season cancelled as officials investigate disappearance of an estimated 1 billion crabs • CBS News
In a major blow to America’s seafood industry, the Alaska Department of Fish and Game has, for the first time in state history, canceled the winter snow crab season in the Bering Sea due to their falling numbers. While restaurant menus will suffer, scientists worry what the sudden population plunge means for the health of the Arctic ecosystem.
An estimated one billion crabs have mysteriously disappeared in two years, state officials said. It marks a 90% drop in their population.
“Did they run up north to get that colder water?” asked Gabriel Prout, whose Kodiak Island fishing business relies heavily on the snow crab population. “Did they completely cross the border? Did they walk off the continental shelf on the edge there, over the Bering Sea?”
Ben Daly, a researcher with ADF&G, is investigating where the crabs have gone. He monitors the health of the state’s fisheries, which produce 60% of the nation’s seafood.
“Disease is one possibility,” Daly told CBS News.
He also points to climate change. According to the National Oceanic and Atmospheric Administration, Alaska is the fastest warming state in the country, and is losing billions of tons of ice each year — critical for crabs that need cold water to survive.
“Environmental conditions are changing rapidly,” Daly said. “We’ve seen warm conditions in the Bering Sea the last couple of years, and we’re seeing a response in a cold adapted species, so it’s pretty obvious this is connected. It is a canary in a coal mine for other species that need cold water.”
|• Why do social networks drive us a little mad?
• Why does angry content seem to dominate what we see?
• How much of a role do algorithms play in affecting what we see and do online?
• What can we do about it?
• Did Facebook have any inkling of what was coming in Myanmar in 2016?
Read Social Warming, my latest book, and find answers – and more.
Errata, corrigenda and ai no corrida: none notified