Bad news, folks: an AI played a game of Tetris so good that the gameplay itself couldn’t keep up. CC-licensed photo by Conor Lawless 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.
A selection of 10 links for you. Friday, yeah? I’m @charlesarthur on Twitter. Observations and links welcome.
AI plays the best game of Tetris you’ve ever seen • Gizmodo
Andrew Liszewski:
»
Like human players, [Greg] Cannon’s impressive StackRabbit AI gets better at playing Tetris through repeatedly playing and analyzing the game to develop improved strategies. But unlike human players, StackRabbit has nerves of steel and doesn’t start to panic as the ever-growing stack of tetrominoes approaches the top of the play board, which it pairs with lightning-quick reflexes to play one of the most mesmerizing and impressive rounds of Tetris you’ve probably ever seen.
Clearing four lines at once is not only satisfying, it’s also the best way to quickly rack up points when playing Tetris. But while, theoretically, a talented player could stack tetrominoes indefinitely, the 8-bit NES version of the game (which is used for The Classic Tetris World Championships) starts to melt down as gameplay approaches level 29 where the game’s speed doubles. The developers assumed this was the point where human players wouldn’t be able to keep up, and while some have managed to make it just past level 29, the game starts to quickly exhibit graphical glitches as the load on the NES’s processor increases.
Human players have managed to hit NES Tetris high scores of over 1.6 million points, but with artificial human limits removed, Cannon’s StackRabbit AI managed to reach level ••• of the game with a score of [—-] points after around an hour and five minutes of gameplay. Watching the AI’s unbelievable run is often as confusing as it is mesmerizing as in later levels the game starts using the wrong graphical elements to build the tetromino pieces.
«
Try to guess how many levels it gets to, and how many points. Hint: levels into three figures, points in eight figures.
Aligning language models to follow instructions • OpenAI
»
The OpenAI API is powered by GPT-3 language models which can be coaxed to perform natural language tasks using carefully engineered text prompts. But these models can also generate outputs that are untruthful, toxic, or reflect harmful sentiments. This is in part because GPT-3 is trained to predict the next word on a large dataset of Internet text, rather than to safely perform the language task that the user wants. In other words, these models aren’t aligned with their users.
To make our models safer, more helpful, and more aligned, we use an existing technique called reinforcement learning from human feedback (RLHF). On prompts submitted by our customers to the API*, our labellers provide demonstrations of the desired model behavior, and rank several outputs from our models. We then use this data to fine-tune GPT-3.
The resulting InstructGPT models are much better at following instructions than GPT-3. They also make up facts less often, and show small decreases in toxic output generation.
* (We only use prompts submitted through the Playground to an earlier version of the InstructGPT models that was deployed in January 2021. Our human annotators remove personal identifiable information from all prompts before adding it to the training set.)
«
The difference is pretty dramatic – when GPT-3 is challenged to “Explain the moon landing to a 6 year old in a few sentences” it does it in one. The previous untrained version got nowhere near it.
unique link to this extract
Putin backs bitcoin and crypto mining despite Bank of Russia’s ban plans • Bloomberg
Evgenia Pismennaya:
»
President Vladimir Putin backs a Russian government proposal to tax and regulate mining of cryptocurrencies, rejecting the central bank’s proposal to ban it completely, according to three people familiar with the matter.
Putin supports the proposal, which would allow mining to continue, as Russia has many regions with a surplus of electricity, including Irkutsk, Krasnoyarsk and Karelia, the people said, asking not to be identified because the information is not public.
Kremlin spokesman Dmitry Peskov declined to say what Putin’s stance was and said that the president ordered the government and central bank to work out their differences.
The central bank continues to oppose mining on environmental grounds and because it creates incentives to bypass regulations, its press service said in a response to questions.
The government’s press service did not respond to a request to comment on the status of the talks.
Putin’s position is good news for an industry that has suffered numerous setbacks recently, including China’s complete ban last year and Kazakhstan temporarily unplugging miners this week as the country faced blackouts. Russia became the world’s third biggest crypto miner in 2021, after the US and Kazakhstan, according to Cambridge University data released in October.
«
Note though that it’s “tax and regulate”. Not just “allow”.
unique link to this extract
Gender-based online violence spikes after prominent media attacks • Brooking Institute
Megan Brown, Zeve Sanderson, and Maria Alejandra Silva Ortega:
»
On March 9, 2021, Fox News host Tucker Carlson took aim at a favorite target: a New York Times journalist. In the crosshairs was tech reporter Taylor Lorenz, one of the paper’s rising stars who had recently described the toll of online harassment. The torrent of online hate she receives had “destroyed her life” she said, in a tweet supporting the launch of the Online Violence Response Hub.
“Destroyed her life, really? By most people’s standards, Taylor Lorenz would seem to have a pretty good life, one of the best lives in the country, in fact,” Carlson intoned. “Lots of people are suffering right now, but no one is suffering quite as much as Taylor Lorenz is suffering.”
After Carlson mocked Lorenz in his segment, her social media mentions and inbox were again filled with violent threats and harassment—a dynamic likely familiar to many women with a public presence online. The vitriol Lorenz endured was an example of gender-based online violence, which UNESCO recently characterized as online rhetoric against women designed to “induce fear, silence, and retreat; and … chill their active participation” in public debate. Yet as problematic as the phenomenon is, it remains relatively understudied. To better understand how gender-based online violence takes place, we therefore examined three instances in which female journalists were attacked by prominent male media figures on either social media or broadcast media, and then tracked the ways in which online violence against them spiked.
«
Sometimes social warming is speeded up by external factors. Fox News is basically a blowtorch.
unique link to this extract
Apple creates personal safety guide as AirTag concerns mount • The Verge
Victoria Song:
»
On Tuesday, Apple quietly launched a Personal Safety User Guide to help “anyone who is concerned about or experience technology-enabled abuse, stalking or harassment.” The guide is a resource hub to help people figure out what their options are if they wish to remove someone’s access to shared information, as well as personal safety features available across the Apple ecosystem. Most notably, it includes a “Stay safe with AirTag and other Find My accessories” page at a time when an increasing number of people have come forward about being stalked with the devices.
As pointed out by 9to5Mac, the hub is mostly repackaging a data privacy guide that was first published about a year ago. Overall, it’s a good thing to create an easily accessible resource to help people keep their information safe or find out what to do in the event their safety is threatened.
…While the guide is helpful, the timing is unsurprising. Several outlets, including CNBC, BBC News, The Guardian, and The New York Times, have run stories in the past few weeks detailing multiple instances of users receiving alerts they’d been tracked by an unknown AirTag. Others have shared their personal experiences directly to social media like TikTok, and in early January, Sports Illustrated model Brooks Nader shared her own experience in her Instagram Stories.
When the AirTags launched in April 2021, Apple emphasized the devices had anti-stalking measures built-in. That included notifications sent to iPhones if an AirTag was detected moving with them over time and sound alerts. However, some reviewers heavily criticized the measures as being insufficient, especially since it initially took three days for the AirTags to play a sound alert. Apple then changed that to a random period between eight and 24 hours after being separated from the owner’s iPhone.
«
No good (or bad) technology goes unpunished.
unique link to this extract
Woman puts Apple AirTag in a box and catches mover lying about his location • The Washington Post
Marisa Iati:
»
By the time her cross-country move came around in December, Valerie McNulty had heard too many horror stories about delays and theft during military relocations to take any chances.
So she took the bracelet with an Apple AirTag that her 4-year-old son had been wearing to school and dropped it into a box of his toys. Off it went, from the family’s home in Fort Carson, Colo., toward their new base in Fort Drum, N.Y. — until the expected delivery date came and passed with no sign of their belongings.
In the days that followed, McNulty used the AirTag’s Bluetooth-enabled tracking capability to catch her moving-truck driver in a lie and pinpoint the location of her family’s items.
…In McNulty’s case, her AirTag might have saved the day. After she reported her family’s missing items in early January, she said, the company that the Army had contracted with to coordinate their move promised that the boxes would arrive the next day — a Saturday.
Then the driver called. He said he had just picked up their items in Colorado and would drop them off that Monday, McNulty recalled.
That’s when McNulty, 33, swiped to her iPhone’s Find My app and checked the location of her AirTag. It was in Pennsylvania, she said, about four hours from her family’s new home.
McNulty pointed this out to the driver, who, she said, promptly hung up the phone. He called back a few minutes later and said he could get her the delivery that Sunday or Monday, she said. Then he called one more time, McNulty recalled, and said he was going to see his romantic partner in New Jersey but would still bring her the truck the next day.
McNulty took screenshots of the AirTag’s location through the night, she said, to hold the driver accountable to his word. The next day, Jan. 8, the truck finally arrived.
«
Soooo… AirTags good?
unique link to this extract
Why Spotify kept Joe Rogan over Neil Young • The Verge
Ashley Carman:
»
Spotify’s thinking is obvious. What does losing Young mean? The company’s not financially dependent on his streams or subscribers — Drake or Taylor Swift might be a different story — and barring a mass exodus of subscribers over his missing catalog, things remain business as usual. In fact, the company loses money every time someone streams Young’s songs, which is why Spotify wanted to get into podcasting in the first place. It makes money every time someone listens to Rogan.
On the flip side, I’m not sure what Young’s label, Warner Records, gets out of this. Maybe it wants leverage in a negotiation or to change the conversation around streaming? I’m not sure, but I do assume some sort of politicking is happening behind the scenes that could somehow net a win for Warner. Maybe people listen to more Young elsewhere? Buy some CDs? Unclear.
Still, the takeaway from the skirmish is clear: Spotify can’t afford to ostracize Rogan or his audience. The company specifically licensed his show with the goal of both converting listeners to the platform and making money through ad sales. JRE has become the lynchpin to its entire podcasting apparatus.
A source previously told me that if marketers buy ads on Rogan, they have to buy ads on the rest of Spotify’s catalog, too, meaning Rogan’s success brings more advertisers to the rest of Spotify’s investments. Without him, Spotify has Call Her Daddy and Armchair Expert, but neither reaches Rogan’s scale. It’s easy to see why Spotify didn’t cave so easily.
However, the thing that interests me more is what this says about Spotify’s approach to moderation. When we think about moderation issues on social media platforms, it’s typically one in which algorithms promote and monetize sensational, inflammatory, and problematic content. Up until now, podcasting has mostly remained out of the conversation. The industry relies on word of mouth and curated lists, and the hope is that software recommendations will do more some day in the future.
…Even more oddly, Spotify told The Wall Street Journal yesterday it has taken down 20,000 podcast episodes in violation of “detailed content policies” related to COVID-19. It’s unclear if what I quoted above is the detailed policy or if it goes beyond a sentence. Regardless, the open question is what Spotify considers crossing its line. Has Rogan not crossed it? What did these other podcasters say to be taken down? Also, who is this in-house team? Who’s flagging these violations — software or humans?.
«
That point about forcing advertisers to buy space elsewhere on the catalog(ue) is noteworthy. Rogan’s definitely earning his keep for Spotify if so. But, also: questions about moderation.
unique link to this extract
Everything new in iOS 15.4 and iPadOS 15.4: Face ID with a mask, emojis, Apple Card widget, Universal Control and more • MacRumors
Juli Clover:
»
Apple today seeded the first betas of iOS 15.4, iPadOS 15.4 to developers for testing purposes, adding a slew of new features to the latest iOS operating systems. iOS 15.4 is the biggest update that we’ve had to iOS 15 to date, and it brings Universal Control, Face ID with a mask, new emojis, and tons more.
• Face ID with a mask: with iOS 15.4, there is now an option to unlock your iPhone while wearing a mask. Apple warns that full face ID is the more secure option, but mask Face ID is now available.
Face ID with a mask can be enabled after updating to iOS 15.4, and it is designed to use the area around your eyes for authentication purposes. It works with glasses, but it is not compatible with sunglasses, and you must be looking at the iPhone to unlock it with a mask on. Face ID with a mask looks to be limited to the iPhone 12 and newer.
• Universal Control: iPadOS 15.4 and macOS Monterey 12.3 enable the long-awaited Universal Control feature, which is designed to allow you use a single cursor and keyboard to control the iPads and Macs that are signed into your iCloud account.
«
Face ID with a mask has taken two years (though Samsung’s Galaxy range offered it from the start of offering face recognition – hard to feel that’s good?). Just as, at least in the UK, most mask mandates are ending.
Universal Control feels like one of those things which is wonderful in a demo but of very limited utility most of the rest of the time. How close does the other device need to be? (Close enough you can make sense of what you’re manipulating, I guess.) How many times do you want to manipulate things on other devices, if they’re Apple devices signed into your account? Those should sync via iCloud. Unless, of course, that’s broken.
unique link to this extract
Has Godwin’s law, the rule of Nazi comparisons, been disproved?
Stephen Harrison:
»
The moment in the debate when someone glibly calls their ideological opponent a “Nazi” is what the French call “the Godwin point.”
But recent academic research makes a bold claim: Godwin’s law does not work in practice. The study’s authors reviewed a sampling of nearly 200 million Reddit posts and found that references to “Hitler” and “Nazis” did not occur with a high degree of frequency. In fact, after a certain point, the probability of observing these words actually decreased.
That’s a counterintuitive outcome, to say the least. Based on the abundance of literature comparing Trump’s presidency to fascist regimes, or the way Fox News commentators link Anthony Fauci to Nazi doctors, not to mention troubling coverage of a neo-Nazi resurgence in Germany, one would expect that online Hitler comparisons would unfortunately be on the rise.
That’s not the case, according to the research. As the study’s authors put it, the Reddit results “suggest that it is not inevitable that conversations eventually disintegrate into reductio ad Hitlerum.” Godwin’s law is nearly as old as the internet itself, but does this study show that it is dead?
Dariusz Jemielniak, one of the study’s authors, said in an email that killing Godwin’s law was not his goal. What he wanted instead was to test it, something that at first seemed impracticable. The project began with Jemielniak and colleagues relaxing in a garden, chatting about vocabulary in online communities—typical behavior for their group of data scientists, apparently—when he mentioned that Godwin’s law could theoretically be testable if only they could find a large enough data set. According to Jemielniak, Gabriele Fariello, a statistician who teaches at Harvard, said, “Hold my beer,” and they were off.
«
So they’ve disproved Godwin’s Law and Betteridge’s Law with one story. (Thanks G for the link.)
unique link to this extract
Anti-vaxxers making ‘at least $2.5m’ a year from publishing on Substack • The Guardian
Dan Milmo:
»
Prominent figures in the anti-vaccine movement including Dr Joseph Mercola and Alex Berenson have large followings on Substack, which has more than 1 million paying subscribers who sign up for individual newsletters from an array of authors who include novelist Salman Rushdie, the writer musician Patti Smith and former Downing Street adviser Dominic Cummings.
Mercola, a US alternative medicine doctor and prolific producer of anti-vaccine content, and Alex Berenson, a journalist banned from Twitter last year after questioning the efficacy of Covid-19 vaccines, are among five vaccine sceptics on the platform who earn themselves and Substack a minimum of $2.5m a year from their newsletters. Under Substack’s business model, writers keep about 90% of the subscription income, with the platform taking 10% and payment company Stripe charging the writers 3% of their take.
Research by the Center for Countering Digital Hate, a campaign group, showed that Mercola’s newsletters made a minimum of $1m a year from charging subscribers an annual fee of $50, with Berenson making at least $1.2m from charging people $60. Three other vaccine sceptic newsletters, from tech entrepreneur Steven Kirsch, virologist Robert Malone and anonymous writer Eugyppius, generate about $300,000 between them.
Imran Ahmed, chief executive of CCDH, said companies like Substack were under “no obligation” to amplify vaccine scepticism and make money from it. “They could just say no. This isn’t about freedom; this is about profiting from lies … Substack should immediately stop profiting from medical misinformation that can seriously harm readers.”
«
It is a pity that people like Berenson are making such huge amounts of money for spreading nonsense. Set this situation 15 years (maybe even just 10) earlier, and he would have struggled to get the software that would let him do this.
There’s also a huge motivation for Berenson et al not to alter their position in any way, and once the pandemic dies down (as surely it will) they will have to pivot to some other divisive topic in order to keep the money rolling in.
unique link to this extract
Errata, corrigenda and ai no corrida: Andrew Brown offers this observation on Agatha Christie’s ability to hire a maid and a nurse, but not buy a car – a situation which, I commented, is reversed now despite there being more people on the planet:
The humans we hire cheaply — at least as cheaply relative to our salaries as maids were — are all out of sight and half way round the world. And they tend not to work for us exclusively or directly. But if you took out of our lives everything built or maintained by people paid £2,600 a year after food and housing, a lot would vanish that we now take for granted. The electrification of the home abolished such an *enormous* amount of physical labour.
| • 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? Social Warming, my latest book, and find answers – and more. |
The Overspill: when there's more that I want to say 