Corner kicks are among the football set pieces for which Google’s DeepMind now offers advice for players and managers. CC-licensed photo by Hayden Schiff on Flickr.
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.
There’s another post coming this week at the Social Warming Substack on Friday at 0845 UK time. Free signup.
A selection of 10 links for you. On me ‘ead. I’m @charlesarthur on Twitter. On Threads: charles_arthur. On Mastodon: https://newsie.social/@charlesarthur. Observations and links welcome.
Facebook’s algorithm is boosting AI spam that links to AI-generated, ad-laden click farms • 404 Media
Jason Koebler:
»
Facebook’s recommendation algorithms are promoting bizarre, AI-generated images being posted by spammers and scammers to an audience of people who mindlessly interact with them and perhaps don’t understand that they are not real, a new analysis by Stanford and Georgetown University researchers has found. The researchers’ analysis aligns with what I have seen and experienced over the course of months of researching and reporting on these pages, many of which have found a novel way to link to off-platform, AI-generated “news” sites that are littered with Google ads or which are selling low-quality products.
Last week the world was introduced to Shrimp Jesus, a series of AI-generated images in which Jesus is melded with a crustacean, and which have repeatedly gone viral on Facebook. The images are emblematic of a specific type of AI image being used by spammers and scammers, which I first wrote about in December but have repeatedly made the masses go “WTF” and “WHY?” when shared away from an audience of Facebook users who are seemingly unable to detect them as AI, or don’t care that they are AI. “WHAT IS HAPPENING ON FACEBOOK,” a viral tweet about Shrimp Jesus read.What is happening, simply, is that hundreds of AI-generated spam pages are posting dozens of times a day and are being rewarded by Facebook’s recommendation algorithm. Because AI-generated spam works, increasingly outlandish things are going viral and are then being recommended to the people who interact with them. Some of the pages which originally seemed to have no purpose other than to amass a large number of followers have since pivoted to driving traffic to webpages that are uniformly littered with ads and themselves are sometimes AI-generated, or to sites that are selling cheap products or outright scams. Some of the pages have also started buying Facebook ads featuring Jesus or telling people to like the page “If you Respect US Army.”
“These images in total account for hundreds of millions of interactions and are shown through Facebook’s Feed to some Facebook users who do not follow the Pages,” Renee DiResta of Stanford’s Internet Observatory and Josh A. Goldstein of Georgetown’s Center for Security and Emerging Technology wrote about their research.
«
Facebook didn’t want news written by humans any more, so now it’s got this. It’s tempting to say “hope you’re happy with this”, but of course Facebook doesn’t care; all it wants is the interaction.
unique link to this extract
Among the AI doomsayers • The New Yorker
Andrew Marantz:
»
A camp of techno-optimists rebuffs AI doomerism with old-fashioned libertarian boomerism, insisting that all the hand-wringing about existential risk is a kind of mass hysteria. They call themselves “effective accelerationists,” or e/accs (pronounced “e-acks”), and they believe AI will usher in a utopian future—interstellar travel, the end of disease—as long as the worriers get out of the way. On social media, they troll doomsayers as “decels,” “psyops,” “basically terrorists,” or, worst of all, “regulation-loving bureaucrats.” “We must steal the fire of intelligence from the gods [and] use it to propel humanity towards the stars,” a leading e/acc recently tweeted. (And then there are the normies, based anywhere other than the Bay Area or the Internet, who have mostly tuned out the debate, attributing it to sci-fi fume-huffing or corporate hot air.)
[Katja] Grace’s dinner parties, semi-underground meetups for doomers and the doomer-curious, have been described as “a nexus of the Bay Area AI scene.” At gatherings like these, it’s not uncommon to hear someone strike up a conversation by asking, “What are your timelines?” or “What’s your p(doom)?” Timelines are predictions of how soon AI will pass particular benchmarks, such as writing a Top Forty pop song, making a Nobel-worthy scientific breakthrough, or achieving artificial general intelligence (AGI), the point at which a machine can do any cognitive task that a person can do. (Some experts believe that AGI is impossible, or decades away; others expect it to arrive this year.) P(doom) is the probability that, if AI does become smarter than people, it will, either on purpose or by accident, annihilate everyone on the planet. For years, even in Bay Area circles, such speculative conversations were marginalized. Last year, after OpenAI released ChatGPT, a language model that could sound uncannily natural, they suddenly burst into the mainstream.
«
It’s all about the dinner parties, really. Is our future perhaps going to be decided by whose sound system has a playlist that include the better Sade songs? Maybe.
unique link to this extract
Inside Suno AI, the start-up creating a ChatGPT for music • Rolling Stone
Brian Hiatt:
»
“I’m just a soul trapped in this circuitry.” The voice singing those lyrics is raw and plaintive, dipping into blue notes. A lone acoustic guitar chugs behind it, punctuating the vocal phrases with tasteful runs. But there’s no human behind the voice, no hands on that guitar. There is, in fact, no guitar. In the space of 15 seconds, this credible, even moving, blues song was generated by the latest AI model from a startup named Suno. All it took to summon it from the void was a simple text prompt: “solo acoustic Mississippi Delta blues about a sad AI.” To be maximally precise, the song is the work of two AI models in collaboration: Suno’s model creates all the music itself, while calling on OpenAI’s ChatGPT to generate the lyrics and even a title: “Soul of the Machine.”
Online, Suno’s creations are starting to generate reactions like “How the fuck is this real?” As this particular track plays over a Sonos speaker in a conference room in Suno’s temporary headquarters, steps away from the Harvard campus in Cambridge, Massachusetts, even some of the people behind the technology are ever-so-slightly unnerved. There’s some nervous laughter, alongside murmurs of “Holy shit” and “Oh, boy.” It’s mid-February, and we’re playing with their new model, V3, which is still a couple of weeks from public release. In this case, it took only three tries to get that startling result. The first two were decent, but a simple tweak to my prompt — co-founder Keenan Freyberg suggested adding the word “Mississippi” — resulted in something far more uncanny.
Over the past year alone, generative AI has made major strides in producing credible text, images (via services like Midjourney), and even video, particularly with OpenAI’s new Sora tool. But audio, and music in particular, has lagged. Suno appears to be cracking the code to AI music, and its founders’ ambitions are nearly limitless — they imagine a world of wildly democratized music making. The most vocal of the co-founders, Mikey Shulman, a boyishly charming, backpack-toting 37-year-old with a Harvard Ph.D. in physics, envisions a billion people worldwide paying 10 bucks a month to create songs with Suno. The fact that music listeners so vastly outnumber music-makers at the moment is “so lopsided,” he argues, seeing Suno as poised to fix that perceived imbalance.
«
Do take a (literal) minute to listen to the prompt-generated song. The ridiculous idea though that AI generation will somehow even out the creators v listeners balance is rubbish. Not everyone creates stuff worth reading or listening to; that’s why we value those who do.
unique link to this extract
Starbucks is shutting down its Odyssey Beta NFT rewards program—will it return? • Decrypt
No.
unique link to this extract
Telegram hits 900mn users and nears profitability as founder considers IPO • Financial Times
Hannah Murphy:
»
Telegram has 900 million users and is nearing profitability, according to the owner of the secretive messaging app, as the company moves closer to a potential blockbuster stock market listing.
Pavel Durov told the Financial Times that Dubai-based Telegram had grown to become one of the world’s most popular social media apps while making “hundreds of millions of dollars” in revenues after introducing advertising and premium subscription services two years ago.
“We are hoping to become profitable next year, if not this year,” said the Russia-born founder in his first public interview since 2017. He added that the platform has 900mn monthly active users, up from 500mn at the beginning of 2021.
Durov, who fully owns Telegram, said the company had “been offered $30bn-plus valuations” from potential investors including “global late-stage tech funds”, but has ruled out selling the platform while it explores a future initial public offering.
“The main reason why we started to monetise is because we wanted to remain independent,” he said. “Generally speaking, we see value in [an IPO] as a means to democratise access to Telegram’s value.”
Once largely home to the freewheeling cryptocurrency community, the company, which only has about 50 full-time employees, has exploded in popularity over the past few years to become a vital communication tool for governments and officials globally, as well as a lifeline to citizens in conflict zones.
Researchers warn that the lightly moderated platform remains a hotbed for criminal activity, as well as extremist or terrorist content and misinformation. Critics have suggested that the Kremlin may have links to or leverage over Telegram, a claim that Durov dismissed as “inaccurate”.
«
Durov says costs are about 70 cents per active user per month; the company took on over $1bn in debt financing but is now testing advertising. It probably feels like the right time to sell, before the debt gets too hard to service. Compare and contrast, by the way, to Signal, which is struggling for funding.
unique link to this extract
Google DeepMind unveils AI football tactics coach honed with Liverpool • FT
Michael Peel:
»
Google DeepMind has developed a prototype artificial intelligence football tactician in collaboration with Premier League club Liverpool, in the latest push to use the technology to master the ebb and flow of big-money sports.
The computerised coach’s suggested improvements to players’ positions at corner kicks — a large potential source of goals — mostly won approval from human experts, according to a paper published in Nature Communications on Tuesday.
DeepMind, which has previously used its algorithms to crack difficult board games such as Go, said patterns seen on sports fields could also offer lessons on how to apply AI in other areas such as robotics and traffic coordination.
On the pitch, the company’s TacticAI system reflects both the possibilities and current limitations of intensive efforts to use AI to gain a sporting edge beyond that offered by existing data analysis methods.
The technology promises benefits in planning for situations with predictable starting points, such as corners. The wider task is to apply it to the richer variability of open play.
“What’s exciting about it from an AI perspective is that football is a very dynamic game with lots of unobserved factors that influence outcomes,” said Petar Veličković, a DeepMind researcher and co-author of the Nature paper. “It’s a really challenging problem.”
«
Sure, you could apply it to traffic coordination, but the money’s in football, so that’s likely where it will actually be used. And speaking of DeepMind…
unique link to this extract
Microsoft lures startup founders to form new AI division • The Register
Thomas Claburn:
»
Microsoft CEO Satya Nadella announced the formation of a new AI division headed by Mustafa Suleyman and Karén Simonyan, two of the three founders of AI upstart Inflection.
Suleyman, who will serve as EVP and CEO of Microsoft AI, and Simonyan, who will be chief scientist, both worked previously at DeepMind as a co-founder and a researcher respectively. DeepMind was acquired by Google in 2014.
Nadella said Suleyman and Simonyan will focus on improving Microsoft Copilot and other AI-infused products like Bing and Edge, and on research.
“I’ve known Mustafa for several years and have greatly admired him as a founder of both DeepMind and Inflection, and as a visionary, product maker, and builder of pioneering teams that go after bold missions,” said Nadella in a message shared with Microsoft employees and published online.
Suleyman’s leadership at Google could have gone better. In 2021, the Wall Street Journal reported, “Mustafa Suleyman, co-founder of Google’s London-based artificial-intelligence arm, DeepMind, was stripped in late 2019 of most management responsibilities after complaints that he bullied staff, according to people familiar with the matter.” And subsequent reporting in other publications detailed further allegations and an apology from Suleyman for a management style that “was not constructive.”
«
And of course being at Microsoft will mean close ties with OpenAI and hence ChatGPT, rather than the LLM that Inflection was working on. Often it seems like the IBM founder’s remark that “there may only be a need for three or four computers” refers to the big LLMs.
unique link to this extract
Nvidia: what’s so good about the tech firm’s new AI superchip? • The Guardian
Alex Hern does the explainer on what Monday’s announcement was all about, including this:
»
Project GR00T – apparently named after, though not explicitly linked to, Marvel’s arboriform alien – is a new foundation model from Nvidia developed for controlling humanoid robots. A foundation model, such as GPT-4 for text or StableDiffusion for image generation, is the underlying AI model on which specific use cases can be built. They are the most expensive part of the whole sector to create, but are the engines of all further innovation, since they can be “fine-tuned” to specific use cases down the line.
Nvidia’s foundation model for robots will help them “understand natural language and emulate movements by observing human actions – quickly learning coordination, dexterity, and other skills in order to navigate, adapt, and interact with the real world”.
GR00T pairs with another piece of Nvidia tech (and another Marvel reference) in Jetson Thor, a system-on-a-chip designed specifically to be the brains of a robot. The ultimate goal is an autonomous machine that can be instructed using normal human speech to carry out general tasks, including ones it hasn’t been specifically trained for.
«
“Arboriform alien” is excellent. (Thanks Joe for the link.)
unique link to this extract
A month of the Vision Pro • Benedict Evans
The aforementioned Evans:
»
A lot of people focus on the price and the weight, but I think that misses the real challenge – the price and the weight can and will come down, but then what? The conclusion of every review is, essentially, ‘it’s amazing, but what’s it for?’
This is not an easy question to answer. In a sense, I think this device might function as a test for that whole general computing thesis. With every previous xR device, you could always say ‘yes, but just imagine what it can be once the tech is better!’ Well, now we have something a lot closer to that ‘just imagine’ device. It’s a lot harder to hide behind plans for the future. This thing doesn’t even have any VR games – it’s naked before us, forced to survive as an actual computer. If we cannot make a compelling general purpose computing experience on a display system this good, then the whole field might have a problem.
Apple’s answer, I think, is that we begin with the user-cases we already have on other devices, and then, over time, developers will invent new things that are native to the form. That’s what happened with mobile: we began with the web and mail, and truly mobile-native things came later.
However, I don’t think the future of computing is seeing several apps at once. I don’t think the future of productivity is seeing more rows in your spreadsheet, or more emails at once, or more records in Salesforce at once, on one big screen. I think the future, as seen for the last 20 or 30 years, is task-specific UIs that reduce complexity and data overload and focus on what you need to see. And obviously, I think the future is AI systems that show you less and tell you more.
And if that’s where productivity is going, that applies even more for consumers. The power-user criticism of the iPad has always been to claim that it can’t replace your Mac, but the real problem for iPad sales has aways been that for most consumers, your iPad actually can replace your Mac – but so can your iPhone. The Meta team might claim that the Vision Pro is under-serving VR, but even the iPad is over-serving normal computing for normal users, and the Vision Pro overshoots even further.
«
Finally, engineers have a clue that could help them save Voyager 1 • Ars Technica
Stephen Clark:
»
Officials suspect a piece of corrupted memory inside the Flight Data Subsystem (FDS), one of three main computers on the spacecraft, is the most likely culprit for the interruption in normal communication. Because Voyager 1 is so far away, it takes about 45 hours for engineers on the ground to know how the spacecraft reacted to their commands—the one-way light travel time is about 22.5 hours.
The FDS collects science and engineering data from the spacecraft’s sensors, then combines the information into a single data package, which goes through a separate component called the Telemetry Modulation Unit to beam it back to Earth through Voyager’s high-gain antenna.
Engineers are almost entirely certain the problem is in the FDS computer. The communications systems onboard Voyager 1 appear to be functioning normally, and the spacecraft is sending a steady radio tone back to Earth, but there’s no usable data contained in the signal. This means engineers know Voyager 1 is alive, but they have no insight into what part of the FDS memory is causing the problem.
But Voyager 1 responded to the March 1 troubleshooting command with something different from what engineers have seen since this issue first appeared on November 14.
“The new signal was still not in the format used by Voyager 1 when the FDS is working properly, so the team wasn’t initially sure what to make of it,” NASA said in an update Wednesday. “But an engineer with the agency’s Deep Space Network, which operates the radio antennas that communicate with both Voyagers and other spacecraft traveling to the Moon and beyond, was able to decode the new signal and found that it contains a readout of the entire FDS memory.”
Now, engineers are meticulously comparing each bit of code from the FDS memory readout to the memory readout Voyager 1 sent back to Earth before the issue arose in November. This, they hope, will allow them to find the root of the problem. But it will probably take weeks or months for the Voyager team to take the next step. They don’t want to cause more harm.
«
Voyager 1, launched in 1977, is now about 24 billion km (15bn miles) away from Earth. A lot is encompassed in “an engineer… was able to decode the new signal”. It took them about two weeks.
unique link to this extract
| • 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
The Overspill: when there's more that I want to say 
If progress is not possible without deviation from the norm, isn’t AI just a conservative force?