Start Up No.1928: Substack and the journalists, solar silicon costs plummet, the AI publisher, Twitter’s algorithmic push, and more


The promise of quantum computers never seems to be fulfilled, no matter how much money and effort – and how many milestones – are passed. CC-licensed photo by IBM Research 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. Not part of George Santos’s CV. I’m @charlesarthur on Twitter. Observations and links welcome.


Is Substack the future of media? • New Statesman

Harry Lambert:

»

Newsletter-first media companies are being built by former print editors, most notably Graydon Carter, the former editor of Vanity Fair. Carter recently raised a second round of $17m for his newsletter title Air Mail; it launched in 2019 and already has over 220,000 paying and trial subscribers.

Carter’s departure from Vanity Fair after 25 years as editor may have signalled the end of the magazine era and the birth of the newsletter age. Air Mail exhibits many of the mannerisms of a print magazine and, as Carter has said, his job as editor has not changed; it is just the means of delivery that are different.

Meanwhile, new publications are being built on Substack. Editor Joshi Herrmann’s bid to deliver local news in newsletter form is bringing in around £250,000 a year for his nascent company The Mill in Manchester and other outlets based in Liverpool and Sheffield. Byline Times’s political editor Adam Bienkov has inspired his employer to launch on the platform.

“I think we are where the US was 18 months ago, and we are going to get some big names coming over to Substack,” says Sam Freedman. While the bigger social media platforms have tried to nullify its threat, Twitter and Facebook have both wound down their rival efforts to host newsletters. Instagram has introduced a payment model, in which users can charge each other for access to exclusive content – something Musk could adopt to prop up collapsing revenues at Twitter. In December, he expressed interest in buying Substack and integrating it into Twitter.

The big platforms will need to act. High-profile users on Twitter, Instagram and YouTube are increasingly asking themselves a question provoked by Substack’s precedent: why have I spent years building an audience over which I have no ownership, and for which I am not paid? A stampede to Substack may be about to begin.

«

Especially, as Lambert notes, given the mess Musk is making of Twitter. Although the tricky thing is that earnings there will follow a power law, just like so many things on the internet: huge for some, small for most.
unique link to this extract


Hype around quantum computing recedes over lack of practical uses • Financial Times

Richard Waters:

»

Are today’s rudimentary quantum computers already on the verge of significant feats beyond the reach of traditional computers? Or have their capabilities been exaggerated, as practical uses for the technology recede into the future?

These questions have been thrown into sharp relief in recent days by a claim from a group of Chinese researchers to have come up with a way to break the RSA encryption that underpins much of today’s online communications.

The likelihood that quantum computers would be able to crack online encryption was widely believed a danger that could lie a decade or more in the future. But the 24 researchers, from a number of China’s top universities and government-backed laboratories, said their research showed it could be possible using quantum technology that is already available.

…[CEO Steve] Brierley at [quantum software company] Riverlane said it “can’t possibly work” because the Chinese researchers had assumed that a quantum computer would be able to simply run a vast number of computations simultaneously, rather than trying to gain an advantage through applying the system’s quantum properties.

Peter Shor, the American mathematician who first proposed a way for quantum computers to crack encryption, predicted that the inability to run all the computations at once meant it would take “millions of years” for a quantum computer to run the calculation proposed in the [Chinese] paper.

…Four years ago, John Preskill, a professor of theoretical physics at the California Institute of Technology, predicted that quantum systems would start to outperform and might have commercial uses once they reached 50-100 qubits in size. But that moment has come and gone without quantum systems showing any clear superiority. IBM unveiled a 127-qubit computer more than a year ago, and last month announced that a new 433-qubit processor would be available in the first quarter of 2023.

«

In my decades in journalism, two technologies have promised to upend everything once they came on stream: fusion and quantum computing. Decades later, both are still decades away.
unique link to this extract


Silicon cost per watt down 96% over last two decades • pv magazine USA

John Fitzgerald Weaver:

»

Research by Fraunhofer ISE shows that since 2004 the material usage of polysilicon per watt of solar cell has dropped by approximately 87%. The data suggests that in 2004, 16 grams of silicon were needed to produce a single watt of solar cell. By 2021, that number had shrunk to just over 2 grams.

For example, when the world’s largest solar farm – at just over 5 MW – turned on in Germany in 2004, it was using 150-watt solar panels. At the time, constructing one of these modules would have consumed 2,400 grams of the processed material.

In 2021, Maxeon signed a deal that 1.8 million of its Performance 5 UPP solar modules would be the powerhouse of the world’s eighth largest solar facility – the Primergy Solar farm in Nevada. If we assume that this 545-watt panel uses 2.2 grams of silicon per watt, we get 1,199 grams per module.

That’s approximately 360% higher output per solar panel — using only half of the silicon!

Of course, we’re going to use massively more silicon in 2023 than we did in 2004. In 2004, we deployed 1,044 MW of solar power, using just over 16,000 t of silicon globally. According to Bloomberg, 268 GW of solar was deployed in 2022, which is over 250 times more capacity than what was deployed in 2004. At 2.2 grams per watt, the 268 GW used approximately 590,000 kg of silicon, or 35 times more silicon than was used in 2004.

«

unique link to this extract


CNET is quietly publishing entire articles generated by AI • Futurism

Frank Landymore:

»

The articles are published under the unassuming appellation of “CNET Money Staff,” and encompass topics like “Should You Break an Early CD for a Better Rate?” or “What is Zelle and How Does It Work?”

That byline obviously does not paint the full picture, and so your average reader visiting the site likely would have no idea that what they’re reading is AI-generated. It’s only when you click on “CNET Money Staff,” that the actual “authorship” is revealed.

“This article was generated using automation technology,” reads a dropdown description, “and thoroughly edited and fact-checked by an editor on our editorial staff.”

Since the program began, CNET has put out around 73 AI-generated articles. That’s not a whole lot for a site that big, and absent an official announcement of the program, it appears leadership is trying to keep the experiment as lowkey as possible. CNET did not respond to questions about the AI-generated articles.

Based on Breton’s observations, though, some of the articles appear to be pulling in large amounts of traffic, in spite of Google having vowed to penalize AI-generated content last year. Futurism has reached out to Google for comment.

AI-generated articles are not new — there’s tons littering the internet already, some as low tech as copying a human written article and swapping certain words out with synonyms to obfuscate the plagiarism.

«

As Tom Goodwin observed on Twitter, this could be seen as disrespectful to the reader: if you can’t be bothered to get a human to write the article, why expect them to read it? Though some stuff maybe doesn’t need it.
unique link to this extract


Twitter defaults to a For You page now, just like TikTok • The Verge

Mitchell Clark:

»

Twitter is changing how you move between the algorithmically-driven timeline and the reverse chronological one and making the algorithmic feed the default.

In a change rolling out to iOS users first, the company has taken away the star button at the top right that let you switch between two feeds. In its place are two tabs — one labeled “For You” and the other “Following” — and when you open the app, you’ll see the For You tab first.

For You, which shares a name with TikTok’s algorithmically-driven feed, is similar to the old “Home” option, which shows you tweets from the people you follow out of order, interspersed with tweets it thinks you may like. (This isn’t the first time Twitter has copied a TikTok feature.) Following is what used to be called “Latest Tweets.”

While the change makes it easier to switch between them, taking a single swipe instead of a few taps, there is something that’s been lost — if you set your app to show you the “Latest Tweets,” that would typically stick.

Now, however, even when you close the app and reopen it, you’re shown the For You feed. That is a bit of a bummer; there are a lot of people who far prefer the reverse-chronological feed, and previous attempts to get rid of it or make it harder to access have typically sparked ire. While this change does technically make it easier to get to from the algorithmic timeline, it definitely makes the latter option harder to ignore.

«

People are (predictably! It’s social media!) being performatively upset at this. At a glance, though, it seems like a much simpler way of giving you the options of algorithmic or reverse-chronological timelines via just a swipe (or single tap), and makes them much clearer. The Verge has a GIF showing the interaction. The problem is that one’s choice doesn’t stick: it always reverts back to the algorithmic timeline.
unique link to this extract


Update about an alleged incident regarding Twitter user data being sold online • Twitter Blog

»

In December 2022, additional press reports published that someone claimed that they have access to over 400 million Twitter-associated user emails and phone numbers, and that the data had been exposed through the same vulnerability discovered in January 2022. Recently, in January 2023, a similar attempt to sell data from 200 million Twitter-associated accounts was reported in the media.

After a comprehensive investigation, our Incident Response and Privacy and Data Protection teams concluded that: 

• 5.4 million user accounts reported in November were found to be the same as those exposed in August 2022
• 400 million instances of user data in the second alleged breach could not be correlated with the previously reported incident, nor with any new incident
• 200 million dataset could not be correlated with the previously reported incident or any data originating from an exploitation of Twitter systems
• Both datasets were the same, though the second one had the duplicated entries removed
• None of the datasets analyzed contained passwords or information that could lead to passwords being compromised.

Therefore, based on information and intel analyzed to investigate the issue, there is no evidence that the data being sold online was obtained by exploiting a vulnerability of Twitter systems. [Emphasis in original – Ed.] The data is likely a collection of data already publicly available online through different sources. 

«

This is responding to a bug bounty claim from January 2022. Not sure this is all going to be over quite so quickly.
unique link to this extract


Sam Bankman-Fried’s supersized bet: $1bn for a bitcoin miner on the Kazakh Steppe • WSJ

Eliot Brown and Yuliya Chernova:

»

Just before crypto markets plunged last year, Sam Bankman-Fried‘s hedge fund made a $1bn bet on Genesis Digital Assets, a Cyprus-registered bitcoin miner rigged to consume a small city’s worth of electricity in Kazakhstan.

The cash injection from Mr. Bankman-Fried’s Alameda Research LLC was supersized even for the red-hot crypto startup world, and it dwarfed the FTX founder’s other investments in private companies.

Genesis Digital is now one of the largest assets in the FTX bankruptcy, the product of a globe-spanning spending spree in which Mr. Bankman-Fried’s companies poured money into crypto tokens, arena branding deals and Bahamas real estate.

Extracting value from Genesis Digital could prove difficult. Bitcoin-mining company valuations have plummeted. Only around half of the Alameda funds in Genesis Digital went to the company’s operations, according to Cyprus corporate-registration documents and people familiar with Genesis Digital. More than $500m bought existing shares from two Genesis Digital co-founders, a detail that hasn’t been previously reported.

«

When the full story is finally assembled on the past few crypto years – the bitcoin mining, the NFTs, the shadow banking, the “exchanges” front-running – people will reel at how absurdly profligate people were with other people’s money.
unique link to this extract


Facebook’s partner in Africa, Sama, quits content moderation • Time

Billy Perrigo:

»

Facebook’s largest content moderation provider in Africa announced Tuesday it would be “discontinuing” its work for the social media giant, nearly a year after a TIME investigation found low pay, trauma and alleged union-busting at its Nairobi office.

The company, Sama, is currently the co-defendant, along with Meta, in a Kenyan lawsuit brought by former content moderator Daniel Motaung, who alleges both companies are guilty of multiple violations of the Kenyan constitution.

Sama blamed the decision on the “current economic climate,” and said it would entail letting go of approximately 3% of its staff, mostly from Nairobi.

A Meta spokesperson confirmed the end of the contract in a statement. “We respect Sama’s decision to exit the content review services it provides to social media platforms. We’ll work with our partners during this transition to ensure there’s no impact on our ability to review content.”

Sama’s contract to review harmful content for Meta, Facebook’s parent company, was worth $3.9m in 2022, according to internal Sama documents reviewed by TIME

«

Doesn’t sound like a lot of money, honestly.
unique link to this extract


Global PC shipments decline by 16% to 285m in 2022 • Canalys

»

The global PC market ended 2022 on a low note, with total shipments of desktops and notebooks down by 29% to 65.4 million units in Q4. This represents a fourth consecutive quarter of decline as holiday season spending was muted amid a worsening economic environment.

This pegged total shipments for full-year 2022 at 285.1m units, a 16% drop from the highs of 2021 when all end-user segments saw peak demand. Nevertheless, shipment volumes remain favorable compared to pre-pandemic, with total 2022 shipments 7% higher than in 2019.

Notebooks underwent a larger decline, down by 30% to 51.4m units in Q4 2022 and 19% to 223.8m units for the full year. Desktops fared slightly better, undergoing a decline of 24% to 14.1m units in Q4 and a drop of 7% to 61.3m units across 2022.

«

To save you doing the maths: the laptop/desktop ratio for the year was 4:1 (ie 80% laptops), and the same in Q4.

For all those dreaming of a Mac Pro, imagine that Apple sells ~30m Macs in the year, and that it’s in the same ratio: that’s 24m laptops, 6m desktops in a year, split between the iMac, Mac Studio and Mac Pro.

Apply the same ratio: iMac sells 80% of the 6m (=4.8m), Mac Studio sells 80% of the remainder (=0.96m) and the Mac Pro gets the rest: that’s 6-5.76 = 240,000 Mac Pros per year. That isn’t nothing, but it’s also two orders of magnitude less than the laptops. What sort of price do you put on it?
unique link to this extract


Pegasus: the spyware hiding in plain sight – Book review • The Guardian

I reviewed this new book by Laurent Richard and Sandrine Rigaud, two French journalists who tracked down the misuse of NSO’s smartphone-spying software:

»

The fundamental problem with Pegasus is that of any superpower: it’s too easy, and tempting, to misuse. NSO, and especially its chief executive, have publicly insisted that sales are conditional on the software being used only to target criminals. (And never American phone numbers; NSO knows not to anger the biggest beast.) But plenty of authoritarian states, and those wobbling on the edge, see telling the truth as a criminal act – and thus target journalists and lawyers too.

NSO implies that it can’t know which individuals have been targeted. The opening of Pegasus appears to contradict that: two journalists, Laurent Richard and Sandrine Rigaud of the French investigative journalism outlet Forbidden Stories, receive a list of 50,000 phone numbers from all over the world with a mysterious series of dates and times attached. As they discover, the numbers, dates and times accord with mobile phones in multiple countries, and the time of attempted or successful infection. (The leak’s timing overlaps intriguingly with a case heard in London in 2021, during which it emerged that Pegasus was used to spy on a British lawyer, Baroness Shackleton, and her client, Princess Haya, who was seeking a divorce from Sheikh Mohammed bin Rashid al-Maktoum, the ruler of Dubai.)

«

Pegasus very much proves the point Apple made when the FBI demanded it create a back door to break into an iPhone in 2016: if the power exists, it will be misused.
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: Thanks to the many people who pointed out that yesterday’s article about the early days of Wired UK was by none other than Mr NTK himself, Danny O’Brien.

Leave a comment

This site uses Akismet to reduce spam. Learn how your comment data is processed.