We know that AI illustrators are bad at depicting hands (and handshakes) – but why, exactly? CC-licensed photo by JourneyPure Rehab 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.
It’s Friday, so there’s another post due at the Social Warming Substack at about 0845 UK time.
A selection of 10 links for you. Maybe seafood on the menu? I’m @charlesarthur on Twitter. Observations and links welcome.
Floating pirate-themed restaurant sinks off Thailand coast • BBC News
»
A pirate-themed floating restaurant off the coast of Thailand has sunk in rough water.
Officials say they suspect a pump on the ship stopped working, causing it to sink.
No one was reported injured.
«
Floating ❌
Pirate 🤷♂️
Restaurant ❌
Sic transit gloria mundi.
unique link to this extract
Twitter to remove free API access in latest money making quest • The Verge
Jess Weatherbed:
»
The decision to remove free access to Twitter’s API follows the platform updating its developer rules to ban third-party clients, causing popular third-party Twitter apps like Twitterrific and Tweetbot to abandon the platform.
Many small developers have used Twitter’s free API access to create fun tools and useful bots like novelty weather trackers and black-and-white image colorizers which are not intended to earn income or turn a profit. As a result, it’s likely that many bots and tools utilizing Twitter’s free API access will need to charge a fee or be shut down. It would also impact third parties like students and scientists who use the platform to study online behavior and gather information for research papers.
And as those developers also say, you’re giving us a whole.. seven days for this momentous change, and no idea what the costs are going to be. Musk hinted in a tweet late on Thursday that it might be $100/month. (I wrote about this further on the Social Warming Substack today: see top of this post for the link.)
unique link to this extract
AI image generators keep messing up hands. Here’s why • Buzzfeed News
Pranav Dixit:
»
“It’s generally understood that within AI datasets, human images display hands less visibly than they do faces,” a Stability AI spokesperson told BuzzFeed News. “Hands also tend to be much smaller in the source images, as they are relatively rarely visible in large form.”
To understand more, I got in touch with Amelia Winger-Bearskin, an artist and an associate professor of AI and the arts at the University of Florida, who has been analyzing the aesthetics of AI art on her blog. “I am obsessed with this question!” Winger-Bearskin exclaimed on our video call.
Generative artificial intelligence that’s trained on billions of images scraped from the internet, Winger-Bearskin explained, does not really understand what a “hand” is, at least not in the way it connects anatomically to a human body.
“It’s just looking at how hands are represented” in the images that it has been trained on, she said. “Hands, in images, are quite nuanced,” she adds. “They’re usually holding on to something. Or sometimes, they’re holding on to another person.”
In the photographs, paintings, and screenshots that AI learns from, hands may be holding onto drapery or clutching a microphone. They may be waving or facing the camera in a way where just a few fingers are visible. Or they may be balled up into fists where no fingers are visible.
«
So it’s our fault? Now I’m wondering what happens if you ask these systems to picture feet.
unique link to this extract
Back at Google, cofounder Sergey Brin just filed his first code request in years • Forbes
Richard Nieva:
»
As the battle in artificial intelligence technology heats up between Silicon Valley companies, Google cofounder Sergey Brin is getting hands-on again with software code, after years of day-to-day absence.
On Jan. 24, Brin appeared to file his first request in years for access to code, according to screenshots viewed by Forbes. Two sources said the request was related to LaMDA, Google’s natural language chatbot—a project initially announced in 2021, but which has recently garnered increased attention as Google tries to fend off rival OpenAI, which released the popular ChatGPT bot in November.
Brin filed a “CL,” short for “changelist,” to gain access to the data that trains LaMDA, one person who saw the request said. It was a two line change to a configuration file to add his username to the code, that person said. Several dozen engineers gave the request LGTM approval, short for “looks good to me.” Some of the approvals came from workers outside of that team, seemingly just eager to be able to say they gave code review approval to the company cofounder, that person added.
«
OK, so now it’s serious.
unique link to this extract
ChatGPT sets record for fastest-growing user base in history, report says • Ars Technica
Benj Edwards:
»
On Wednesday, Reuters reported that AI bot ChatGPT reached an estimated 100 million active monthly users last month, a mere two months from launch, making it the “fastest-growing consumer application in history,” according to a UBS investment bank research note. By comparison, TikTok took nine months to reach 100 million monthly users, and Instagram about 2.5 years, according to UBS researcher Lloyd Walmsley.
“In 20 years following the Internet space, we cannot recall a faster ramp in a consumer internet app,” Reuters quotes Walmsley as writing in the UBS note.
Reuters says the UBS data comes from analytics firm Similar Web, which states that around 13 million unique visitors used ChatGPT every day in January, doubling the number of users in December.
…Also on Wednesday, OpenAI announced ChatGPT Plus, a $20 per month subscription service that will offer users faster response times, preferential access to ChatGPT during peak times, and priority access to new features. It’s an attempt to keep up with the intense demand for ChatGPT that has often seen the site deny users due to overwhelming activity.
«
Why VR/AR gets farther away as it comes into focus • MatthewBall.vc
Matthew Ball:
»
In 2023, it’s difficult to say that a critical mass of consumers or businesses believe there’s a “killer” AR/VR/MR experience in market today; just familiar promises of the killer use cases that might be a few years away. These devices are even farther from substituting for the devices we currently use (and it doesn’t seem like they’re on precipice of mainstream adoption, either). There are some games with strong sales—a few titles have done over $100MM—but none where one might argue that, if only graphics were to improve by X%, large swaths of the population would use VR devices or those titles on a regular basis.
I strongly prefer doing VR-based presentations to those on Zoom—where I spend 30-60 minutes staring at a camera as though no one else is there. But the experience remains fraught; functionality is limited; and onboarding other individuals is rarely worth the benefit because its participants seem to find these benefits both few and small. When the iPhone launched, Steve Jobs touted it did three distinct things—MP3 player, phone, internet communicator—better at launch than the single-use devices then on the market. The following year, the iPhone launched its App Store and “There’s an App for That” proliferated, with tens of millions doing everything they could on the device. The “killer app” was that it already had dozens of them.
«
This is the germ of the piece. You’re welcome to read the whole thing, but it’s really long, and Ball could probably afford to hire an editor next time.
unique link to this extract
Our real-names policy is taming the online trolls • The Times
Rose Wild is the Feedback editor at The Times (of London, if that helps):
»
In early December we notified subscribers who use our digital platforms that we would soon be requiring everyone who posts comments to use their real names. At that time only 36% of commenters were doing so. Enforcing the change is an ongoing process but we are now up to 83% and, with apologies to “Maggie Thatcher” and “Jacques Oeuf”, their pseudonyms and joke handles will soon be a thing of the past. The object of the change, as we explained, was to weed out trolls and stamp out the obnoxious posts that were turning some comment threads into playground punch-ups.
We’ve already seen a substantial drop in “toxic” comments, ie, those that are removed because they breach our rules. In the month before the announcement we had an average of 1,348 daily comments flagged by our systems as toxic. In the past month the average was 815, a 40% decrease.
More than a thousand comments have so far been posted on the interview with the actress and comedian Emily Atack in which she discusses the obscene images and threats of violence she receives from men on social media.
Surprisingly, not all comments were sympathetic. As one reader, Derrick Murray, remarked, many seemed to be “either apologists for the weird men who behave like this, or blaming her for being on social media”.
Actually, the response was far from all bad and prompted an interesting conversation about anonymity online. As Brendan Linnane put it: “If all social media had a ‘real name’ comments policy and were treated as a publisher the criminal behaviour Atack suffers would disappear overnight.” Leanora Munn added: “To those who have moaned about having to use their real name, this should be an indication of why such practice should be employed by every online media outlet.”
«
Remarkable that this change – which shouldn’t work! – seems to have worked. Is it something about traceability across platforms? (And how do they verify someone’s name?)
unique link to this extract
Netflix reveals how it will block account sharing • Pocket Lint
Rik Henderson:
»
At the end of 2022, it estimated that more than 100 million households participate in account sharing globally.
One way of countering this will be a paid account sharing plan, whereby it is possible to pay a little extra on the existing subscription fee to share access with family members in other locations. This is currently being trialled in a few South and Central American countries, such as Argentina and Honduras.
However, it has also revealed how it plans to curb unpaid account sharing, with some clarification appearing in a new FAQ section on its website (thanks to GHacks and 9to5Mac).
Using multiple sources of information, including IP addresses, device IDs and account activity, it will determine whether a user is watching Netflix content from home, away or if a totally different household is accessing the service.
To ensure uninterrupted access, Netflix content must be watched via the primary Wi-Fi source at the account holder’s home at least once every 31 days. This creates a “trusted device”, claims the streaming service. If a non trusted or verified device outside that network also tries to access the service for an “extended period of time”, it may be blocked.
For those worried about access when travelling or while on holiday, the primary account holder will be able to do so without any issues. Other profiles on the same account can also access the service for up to seven consecutive days through a temporary code that can be requested by the account holder.
«
Going to create plenty of fun and games for Netflix keeping tabs on all this. But it was inevitable.
unique link to this extract
The shocking state of enthusiast apps on Android • Birchtree
Matt Birchler:
»
I recently commented on Mastodon that I thought when it comes to third party apps, iOS is remarkably far ahead of Android. My feeling is that you can take the best app in a category on Android, and that would be the 3rd to 5th best app in that category on iOS.
It’s harsh, I know, but I really think it’s true for basically every category of app I care about.
Someone responded to me saying that there are a bunch on Android apps that are better than their iOS equivalents. I wanted to be open-minded, so I asked what apps they would recommend I look at to see how Android is ahead of iOS. They recommended a text editor with a UI that looked more like Notepad++ than a modern writing tool.
But they also suggested this RSS reader called Read You. Keep in mind this is supposed to be the best RSS reader, and the app that is going to show me how wrong I am to think iOS apps are ahead of Android ones.
First, this app is in beta, and is far from feature complete, so this isn’t criticism of a production app, it’s more criticism of the suggesting this is as good as it gets on Android or iOS.
«
He makes his points clearly and strongly. John Gruber builds on this and makes a key point: there are some things that matter which don’t submit simply to tickboxes, such as artistic quality or the pleasure of a smooth scroll.
unique link to this extract
How a tiny radioactive capsule was found in Australia’s vast outback • BBC News
Antoinette Radford on the hunt for a gamma- and beta-emitting pea-sized capsule containing caesium-137 which literally fell off a lorry:
»
Radiation portal monitors detect gamma radiation and are typically used at airports to scan individuals to ensure they do not have radioactive substances on them. Gamma spectrometers measure the intensity of the radiation.
Mr Ray said the new detection equipment could be attached to vehicles so searches could be done from moving vehicles at about 50km/h. “It will take approximately five days to travel the original route, an estimated 1400km, with crews travelling north and south along Great Northern Highway,” he said.
But by the end of 31 January, the capsule continued to evade search crews. “More than 660km has been searched so far – thank you to all agencies for their support,” the Department of Fire and Emergency Services said.
So the next morning, when the government revealed the capsule had been found just two metres off the side of the highway at 11:13 local time Wednesday, it seemed the all-but-impossible had been achieved. Authorities said search crews had “quite literally found the needle in the haystack”.
“You can only imagine it’s a pretty lonely stretch of road from Newman down to Perth,” Fire and Emergency Services Commissioner Darren Klemm said at a press conference on Wednesday. “You can’t help but imagine there was an element of surprise from the people in the car when the equipment did spike up.”
While hesitant to give the exact location the radioactive capsule was found, Mr Klemm described it as “the best possible outcome”.
Local media reports suggest it was found some 74km from Newman – so around 200km from the mine site. No one appeared to have been injured by the capsule, according to authorities, and it did not seem to have moved from where it fell. Mr Klemm said the additional resources from the federal government proved key to finding the capsule.
He said the survey equipment used from the start to detect radioactivity – paired with the specialised equipment that physically located the capsule – is how a car driving past at 70km/h found it. A check of the serial code on the capsule confirmed it was the right one.
«
They checked the serial number? How many did they think they were looking for?
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