Could the technology that created ChatGPT help us understand sperm whales’ language? And would we know what they meant if it could? CC-licensed photo by Gregory “Slobirdr” Smith 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 was another post at the Social Warming Substack last Friday: it’s about how the flywheel of Twitter is slowing down.
A selection of 9 links for you. Try the squid. I’m @charlesarthur on Twitter. On Mastodon: https://newsie.social/@charlesarthur. Observations and links welcome.
It’s not just the Discord leak. Group chats are the internet’s new chaos machine • The Atlantic
In every group chat, no matter the size, participants fall into informal roles. There is usually a leader—a person whose posting frequency drives the group or sets the agenda. Often, there are lurkers who rarely chime in. Different chats, depending on the size, develop their own sets of social rules and hierarchies. “The key to every group chat is mutually assured destruction,” the New York Times reporter Astead Herndon tweeted in 2021. “If you’re the only one dropping tea, you’re at risk. [If] one person is a little too silent, they gotta go.” Larger group chats are not immune to the more toxic dynamics of social media, where competition for attention and herd behavior cause infighting, splintering, and back-channeling.
According to the [Washington] Post’s reporting, [21-year-old Jack] Teixeira was fixated on capturing the attention of—and winning approval from—his Discord community. “He got upset” when people in the chat ignored his long, detailed summaries of classified documents, and he threatened to stop posting altogether, one server member told the newspaper. Eventually, Teixiera started sharing photos of the classified documents with the chat because they were more engaging. As the national-security reporter Spencer Ackerman wrote this week, Teixeira “didn’t leak for patriotism, principle, or even money.” His motivation was far less aspirational but, as Ackerman notes, it was “uncomfortably familiar”: He was showing off for the group chat.
Group chats aren’t just good for triggering geopolitical crises—they’re also an effective means to start a bank run, as the world learned last month. The investor panic that led to the swift collapse of Silicon Valley Bank in March was effectively caused by runaway group-chat dynamics. “It wasn’t phone calls; it wasn’t social media,” a start-up founder told Bloomberg in March. “It was private chat rooms and message groups.”
…This presents a major issue: Unlike traditional social media or even forums and message boards, group chats are nearly impossible to monitor. As law enforcement, journalists, and researchers have learned, trying to track extremist groups such as QAnon or right-wing militias is much harder when they retreat to smaller, private chat apps.
Elon Musk just shut down automation for important public safety accounts • Mashable
Since acquiring Twitter, Elon Musk maintained that one of his major objectives was to eliminate the bots.
Last night, Twitter did just that. One problem, though: The bots blocked are the good ones.
Numerous public service Twitter accounts have lost their ability to automatically post breaking news and events. Twitter has been removing API access, which allows many of these accounts to post in an authorised way by the platform, as it switches to Musk’s new high-priced paid API system.
Many of these affected Twitter accounts have automated updates, but aren’t the type of hands-off bot accounts that some may think of when they hear the term “bot.”
For example, numerous National Weather Service accounts that provide consistent updates, both automated and manually posted by humans, shared that they could no longer provide their up-to-the-minute, potentially life-saving updates.
Would it be too contrarian to say that if you’re getting your weather alerts from Twitter, you’re doing it wrong? It seems to me that it was only ever making up space, and that nobody’s really going to miss those ones. But there are lots of rodomontade bots posting far less important (or not at all important) information that people are going to miss.
unique link to this extract
Twitter clashes with Brazil over school violence posts • Bloomberg via Yahoo
Andrew Rosati, Rachel Gamarski and Daniel Carvalho:
A sudden spike in killings in education centers across Brazil — where school shootings are uncommon — prompted authorities this week to clamp down on social media companies that were hosting messages that lauded the attacks.
Twitter, which has espoused a free-speech ethos since being taken over by billionaire Elon Musk last year, initially resisted over 500 requests from Brazil’s Justice Ministry to take down posts and profiles. The stance sparked widespread backlash and the issuing of an executive decree late Wednesday that threatens Twitter and other platforms with fines — or even a potential ban for failure to comply.
The Justice Ministry’s offensive comes as the nation is still reeling from the murder of four children caused by a hatchet-wielding man at a daycare center in southern state of Santa Catarina this month.
Investigators often point to the existence of active groups in the far-reaches of the Internet that celebrate such attacks as one of the reason violence in schools keeps multiplying.
Twitter didn’t respond to a request for comment. Internet activists and enraged Twitter users struck back at the website with posts containing #TwitterApoiaMassacres, or Twitter supports massacres, which later appeared to be blocked on the platform.
Fun how Bloomberg has to explain that “school shootings are uncommon” for its American audience. Of course Brazil is now run by Lula, not Bolsonaro; the latter delighted in stirring up trouble online. For Twitter, though, the question is: what did it really do wrong? There’s no clear indication that the attacks were inspired by Twitter users. The reference the authorities are using is Columbine – another US school massacre.
unique link to this extract
Artist refuses prize after his AI image wins at top photo contest • PetaPixel
A photographer has stirred up fresh controversy and debate after his artificial intelligence (AI) image won first prize at one of the world’s most prestigious photography competitions. He has since declined to accept the prize while the contest has remained silent on the matter.
Berlin-based “photomedia artist” Boris Eldagsen participated this year in the World Photography Organization’s Sony World Photography Awards, a leading photo contest that offers prizes that include $5,000 cash, Sony camera equipment, a trip to London for the awards ceremony, and/or worldwide publicity through a book and exhibition.
Eldagsen submitted an image titled THE ELECTRICIAN to the Creative category of the 2023 Open competition. It picture appears to be a portrait of two women captured with a photographic process from the early days of photography.
The photo is part of a series titled PSEUDOMNESIA: Fake Memories that Eldagsen has been working on since 2022.
“PSEUDOMNESIA is the Latin term for pseudo memory, a fake memory, such as a spurious recollection of events that never took place, as opposed to a memory that is merely inaccurate,” the artist writes on the project page. “The following images have been co-produced by the means of AI (artificial intelligence) image generators.
“Using the visual language of the 1940s, Boris Eldagsen produces his images as fake memories of a past, that never existed, that no one photographed. These images were imagined by language and re-edited more between 20 to 40 times through AI image generators, combining ‘inpainting’, ‘outpainting’, and ‘prompt whispering’ techniques.
Eldagsen’s picture is very impressive: you won’t be saying “AI can’t do hands/fingers” any more. But he argues that it’s not a photo, so shouldn’t win a photo contest.
Low heat pump uptake ‘an embarrassment’ • The Times
A flagship green energy scheme to encourage homeowners to ditch their gas boilers has been branded an “embarrassment” after uptake in the first year was only a third of the planned level.
The government’s boiler upgrade scheme had a budget of £150m to subsidise 30,000 ground or air source heat pump and biomass boiler installations between its launch in May last year and the end of March this year. Yet figures published by the energy regulator Ofgem show that fewer than 10,000 installations were completed under the scheme in this period.
Ministers have set a target of 600,000 heat pump installations a year within the next five years and the sale of gas boilers will be banned by 2035.
Heat pumps work by drawing heat from the air or ground to provide hot water and central heating.
An air source pump can cost between £7,000 and £14,000 to buy and install, while a ground source pump costs between £15,000 and £35,000. The government scheme provides a subsidy of between £5,000 and £6,000 depending on the type of system used. In addition to the grant, there is no VAT on installation, offering a further saving of thousands of pounds.
…[Mike Foster, chief executive of the trade body the Energy and Utilities Alliance which represents the heating and hot water industry] said: “It takes a certain type of genius to fail to give away £150 million of taxpayers’ money and this wretched scheme looks like it has done just that. When will the government actually listen to the people, the majority of whom simply cannot afford a heat pump, subsidised or not.
“It does little for carbon saving compared to investment on insulation. It does not help people keep bills low. It takes from the poor to give to the wealthy and it is an embarrassment of a policy.”
He added: “More taxpayer-subsidised heat pumps have probably been fitted in Cornish holiday homes than the whole of Britain’s second city, Birmingham. That is shameful. People are still hurting with high energy bills, insulating the homes of those most in need should be the priority, not giving hard-earned taxpayers’ cash to those who were going to buy a heat pump anyway. It’s utterly wasteful.”
Probably they would work better in (generally warm) Cornwall than (rather less generally warm) Birmingham, but the “hot water industry” rep has a point. Insulation would be a better idea.
unique link to this extract
Montana lawmakers approve statewide ban on TikTok • WSJ
Meghan Bobrowsky and Stu Woo:
Montana lawmakers on Friday approved a first-of-its-kind bill to ban TikTok across the state, setting the stage for future court battles that could determine the fate of the popular, Chinese-owned social-media app in the US.
The Montana House voted 54-43 to send the bill to Gov. Greg Gianforte’s desk. The governor’s office declined to say whether he would sign the bill but noted Mr. Gianforte had previously banned TikTok on government-issued devices and urged the state university system to do the same.
The bill said the ban would go into effect on Jan. 1, 2024. It would prohibit TikTok, owned by Beijing-based ByteDance Ltd., from operating within the state, and would also bar app stores from offering TikTok within the state. It would fine any entity violating this law $10,000 per violation. It is unclear how some elements of the legislation would be enforced.
…The bill’s authors ahead of the vote said they expect legal challenges that could ultimately reach the US Supreme Court should Mr. Gianforte sign the legislation.
Critics including the American Civil Liberties Union said the bill amounts to censorship and violates free-speech rights protected under the First Amendment.
Nobody is quite sure how this would work: how can Apple or Google know that someone’s in Montana (pop. 1 million, area 147,040 square miles)? What if someone downloads it while out of state and comes back into the state? Would ISPs have to block it? What about people using VPNs? As ever, the internet is mighty uncooperative with laws that try to limit it geographically.
unique link to this extract
Toward understanding the communication in sperm whales • ScienceDirect
Jacob Andreas et al from various academic institutions think it might be possible to use a Large Language Model approach to translate sperm whale speech:
the recent state-of-the-art Transformer models such as GPT-3 (Brown et al., 2020) was pre-trained on a large language corpus comprising over 10^11 data points. While unsupervised structure discovery is also possible without self-supervised representation learning, recent studies have also shown that unsupervised structure discovery can provide benefits.
It is difficult to make an exact analogy between tokens in human languages and whale vocalizations. And, for comparison, the Dominica Sperm Whale Project (DSWP) dataset contains less than 10^4 coda clicks collected over a longitudinal study since 2005. It is thus apparent that one of the key challenges toward the analysis of sperm whale (and more broadly, animal) communications using modern deep learning techniques is the need for sizable datasets capturing a wide range of attributes.
Secondly, human linguistic corpora are easier to deal with because they are typically pre-analyzed (i.e., already presented in the form of words or letters) and verification against ground truth is available, whereas in bioacoustic communication data, the relevant units must be inferred bottom-up with no ground truth available. Given this highly complex learning objective, we expect larger datasets will facilitate the discovery of meaning-carrying units.
It’s ambitious, but I’m reminded of the Wittgenstein saying: if a lion could speak, we could not understand it. A sperm whale’s world and motivations are so different from ours that it would be almost impossible to find the interface between them.
unique link to this extract
How nerds (almost) took over the world • Financial Times
Self-confessed nerd Stephen Bush:
The new Dungeons & Dragons film sums up the cultural moment. If you’ve seen a Marvel superhero film you’ll be familiar with the approach it takes: a lot of quips and self-consciously ironic dialogue. It is hardly a classic of modern cinema but it is not actively bad either. But while it shares a setting and captures something of the anarchic feeling of a session playing Dungeons & Dragons, the interest in what makes a game of Dungeons & Dragons “fun” feels wholly absent from the film. Instead what we are treated to is a dutiful run-through of some of the franchise’s most famous locations and spells.
That might be the most harmful consequence of nerds’ greater purchasing power: that instead of seeking new things to do with old stories, much of our common culture is dominated by low-quality remakes, made to secure a fast buck rather than to tell a good story.
We nerds are drivers of the problem, too. Sometimes it can appear that the thing we dislike most of all is someone making the thing we love more accessible or widely known. The TikToker Francis Bourgeois — real name Luke Nicolson — is castigated for being an insufficiently “real” trainspotter and for making money out of his hobby by appearing in adverts for Gucci, despite engaging millions of new fans in the trainspotting world. Star Wars fans appear to be divided between those who complain when the franchise simply replays the hits of the past as in The Rise of Skywalker, and those who become bitterly angry when it doesn’t, as in The Last Jedi: sometimes the biggest thing the nerd dollar does is pay to keep things exactly as they are.
The dreadful inertia of the Nerd Franchises – Marvel, Star Wars, Lord of the sodding Rings – really has squashed culture. The escapees from beneath it, notably Succession, intentionally restrict themselves: you won’t see spinoffs or prequels because they’re sufficient unto themselves.
unique link to this extract
The end of faking it in Silicon Valley • The New York Times
Not only has funding dried up for cash-burning startups over the last year, but now, fraud is also in the air, as investors scrutinize startup claims more closely and a tech downturn reveals who has been taking the industry’s “fake it till you make it” ethos too far.
Take what happened in the past two weeks: Charlie Javice, the founder of the financial aid startup Frank, was arrested, accused of falsifying customer data. A jury found Rishi Shah, a co-founder of the advertising software startup Outcome Health, guilty of defrauding customers and investors. And a judge ordered Elizabeth Holmes, the founder who defrauded investors at her blood testing startup Theranos, to begin an 11-year prison sentence on April 27.
Those developments follow the February arrests of Carlos Watson, the founder of Ozy Media, and Christopher Kirchner, the founder of software company Slync, both accused of defrauding investors. Still to come is the fraud trial of Manish Lachwani, a co-founder of the software startup HeadSpin, set to begin in May, and that of Sam Bankman-Fried, the founder of the cryptocurrency exchange FTX, who faces 13 fraud charges later this year.
Taken together, the chorus of charges, convictions and sentences have created a feeling that the startup world’s fast and loose fakery actually has consequences. Despite this generation’s many high-profile scandals (Uber, WeWork) and downfalls (Juicero), few startup founders, aside from Ms. Holmes, ever faced criminal charges for pushing the boundaries of business puffery as they disrupted us into the future.
Actions have consequences? This is incredible.
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