Start Up No.1898: what AI tells us about human grammar, pricing the ‘moron risk premium’, get rich hacking legally!, and more

Clocksgoingbackfordaylight 57053721
An academic has calculated that sticking with Daylight Savings Time would save people serious money on energy bills. So why don’t we do it? (Picture by Diffusion Bee on the prompt: “clocks going back for daylight savings time”.)

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 about 0845 UK time. Free signup.

A selection of 9 links for you. Not involved in stand-up rows. I’m @charlesarthur on Twitter. Observations and links welcome.

AI is changing scientists’ understanding of language learning – and raising questions about an innate grammar • The Conversation

Morten Christiansen (prof of psychology, Cornell U) and Pablo Contreras Kallens (PhD student in psych, Cornell U):


many language scientists – including Noam Chomsky, a founder of modern linguistics – believe that language learners require a kind of glue to rein in the unruly nature of everyday language. And that glue is grammar: a system of rules for generating grammatical sentences.

Children must have a grammar template wired into their brains to help them overcome the limitations of their language experience – or so the thinking goes.

This template, for example, might contain a “super-rule” that dictates how new pieces are added to existing phrases. Children then only need to learn whether their native language is one, like English, where the verb goes before the object (as in “I eat sushi”), or one like Japanese, where the verb goes after the object (in Japanese, the same sentence is structured as “I sushi eat”).

But new insights into language learning are coming from an unlikely source: artificial intelligence. A new breed of large AI language models can write newspaper articles, poetry and computer code and answer questions truthfully after being exposed to vast amounts of language input. And even more astonishingly, they all do it without the help of grammar.

Even if their choice of words is sometimes strange, nonsensical or contains racist, sexist and other harmful biases, one thing is very clear: the overwhelming majority of the output of these AI language models is grammatically correct. And yet, there are no grammar templates or rules hardwired into them – they rely on linguistic experience alone, messy as it may be.

GPT-3, arguably the most well-known of these models, is a gigantic deep-learning neural network with 175 billion parameters. It was trained to predict the next word in a sentence given what came before across hundreds of billions of words from the internet, books and Wikipedia. When it made a wrong prediction, its parameters were adjusted using an automatic learning algorithm.


If Chomsky lives to see his fondest theory overturned, then it will be a moment to savour, given how fabulously wrong he has been about so many things outside his specialist field.
unique link to this extract

Quantifying Britain’s moron risk premium • Financial Times

Louis Ashworth:


Britain has finally got its longed-for exceptionalism in recent weeks, experiencing a yadda yadda yadda (you know the story).

Not a whole lot of value has emerged from the past month’s omnishambles, but one valuable development is the coining of “moron risk premium” — in short, the extra money the UK is paying to borrow because its leaders are a few sandwiches short of a tea party.

TS Lombard’s inimitable Dario Perkins appears to have coined the term, which is now pretty widespread.

Section I: The signs you might have MRP
If you’re just getting to grips with MRP, here’s what we’re talking about. The yield on 30-year gilts has shot up far more quickly than other countries’ equivalent bonds since Liz Truss and the wild bunch took power in early September:

In case you’re worried this focuses unfairly on the time since the author joined Alphaville Truss entered Number 10 (on September 5), here’s a year-to-date view:

As that graph shows more clearly, the UK has a potential partner in Italy, whose own MRP peaked in mid-June as former prime minister Mario Draghi’s leadership fell into crisis.


A funny-yet-serious piece, as the entire edifice of the government falls to pieces around Truss’s ears. (Free to read.)
unique link to this extract

Save energy by not turning clocks back in October, says expert • The Guardian

Rachel Hall:


Households could save more than £400 a year on energy bills if clocks are not put back at the end of October, according to an expert, who said it would help people with the cost of living crisis and reduce pressure on the National Grid this winter.

Evening energy demand peaks between 5pm and 7pm during winter, when the sun has already set after daylight savings time (DST). If clocks didn’t go back, it would remain light for at least part of this time, reducing carbon emissions and energy demand.

Prof Aoife Foley, a clean energy expert at Queen’s University Belfast, said: “By simply forgoing the winter DST in October, we save energy because it is brighter in the evening during winter, so we reduce commercial and residential electrical demand as people leave work earlier, and go home earlier, meaning less lighting and heating is needed.”

This would help the government tackle the “energy war” in Europe resulting from the Ukraine invasion, she said. “Dependent on weather conditions this winter it is very likely we may need to start rationing energy very seriously to avoid bigger energy issues in December and January when gas reserves start to run low,” she said.

Foley’s calculations suggest that households could save £1.20 a day and more than £400 a year on electricity bills if clocks are not put back at the end of October, although exact amounts depend on tariffs.

There has long been debate over whether to scrap DST, which was introduced in 1916 to reduce energy demand during the war by prolonging evening daylight in summer. It still benefits some farmers, but is less popular among people who would prefer more light later in the day in winter, and is thought to cause sleep disturbance.


Well I’m going to implement this in my house, and damn the rest of you.
unique link to this extract

Ask an AI art generator for any image. The results are amazing—and terrifying • WSJ

Joanna Stern looks at AI art generators, and it’s much as you’d expect, but this is interesting:


What about the bias? 

My first query for “a technology columnist writing a column” in Dall-E 2 returned four images of white men. Another I conducted of “a man commuting to work” returned four images of white men. In DreamStudio a prompt for a basketball player on the moon returned an image of a Black man.

The source material for training the AI is found across the internet. “We are aware that the data is heavily biased toward western culture and white male culture,” said Jean Oh, an associate research professor at the Robotics Institute at Carnegie Mellon University. “These models can amplify these biases, generating more stereotypical images.”

An OpenAI spokeswoman said the company continues to do research on mitigating bias and improve results. It recently modified Dall-E to diversify its results when a query doesn’t include race or gender—I did see a few examples of this. Both OpenAI and Stability AI suggest you can add specific prompts to increase the diversity of image results.


Something we’ll need to keep tabs on.
unique link to this extract

75% of the time we spend with our kids in our lifetime will be spent by age 12 • 1000 Hours Outside

Ginny Yurich:


Every year at the start of the summer vacation I see all sorts of posts about the 18 summers we get with our kids. The intention of these posts is exactly the same as mine. They are a reminder that we need to fight to slow down and to simplify. We need to pray for perspective on the days that drag on. We need to put down the screens and connect. 

While our intentions may be the same I fully disagree with the number. Maybe those who write articles about the 18 summers with our kids still have only little ones at home. Maybe they have forgotten the summers when they turned 15 and then 16 and were able to drive and have jobs. There is a significant developmental shift that happens during childhood around age 12 (occasionally earlier) and with that often comes a change in family dynamics. Summers begin to have a different look and then eventually parents and siblings become more of a background object, a less integral piece of the puzzle. In fact, 75% of the time we spend with our kids in our lifetime will be spent by age 12.

Don’t get me wrong. Growing up is a good thing, something to be celebrated. Ultimately we want our kids to take on the challenges of the real world and they begin to do this in stages. Drivers training. First jobs. Dating. What all of this means is that  it’s highly unlikely that any of us will get 18 endless summers with our kids. Maybe we will get 13 or 14 if we are fortunate. Our time truly is limited.


It’s quite a sobering thought. OK, the first three years are totally exhausting, which means the next two are mostly spent recovering from them (unless another child happens along, in which case rinse, repeat). But the years after 12 are very high quality, in general. So it’s not all bad.
unique link to this extract

Mango Markets exploiter comes clean, claims all actions were legal • The Block

Osato Avan-Nomayo:


Avraham Eisenberg, the man behind the $114m exploit on Mango Markets, has confirmed that he orchestrated the attack on the DeFi platform in a statement issued today.

“I was involved with a team that operated a highly profitable trading strategy last week,” Eisenberg confirmed, adding, “I believe all of our actions were legal open market actions, using the protocol as designed, even if the development team did not fully anticipate all the consequences of setting parameters the way they are.” Eisenberg declined to comment on the size of his team when asked by The Block.

This legal trading strategy required $10m on Eisenberg’s part to drain $114m from Mango Markets. The “trade” worked by manipulating the price oracle to inflate the mango token price three-fold from $0.30 to $0.91. This boosted the value of Eisenberg’s collateral, allowing him and his team to borrow more funds from the protocol.

Eisenberg’s name was linked to the attack barely a day later. Independent reporter Chris Burnet published an article providing some evidence connecting Eisenberg to the attack. The evidence included leaked screenshots of Discord chats describing the planned attack as well as suspicious on-chain activities following the incident. This is not the first time Eisenberg has been linked to a DeFi exploit. Earlier this year he was accused of defrauding FortressDAO investors to the tune of $14m. 

With regards to FortressDAO, Eisenberg said, “In February, Fortress DAO voted for a full redemption of the Treasury and I helped implement that. By the end of March, this redemption was complete and any fort token holders were able to exit for a proportional share of the Treasury.”


Eisenberg’s form of words – “legal open market actions” – is the same as the defence lawyer for a hacker facing a charge saying “but, your honour, my client simply asked the computer for access, and the computer granted it. That cannot be illegal, because it was allowed.” I’m pretty sure I heard that exact argument, literally in Southwark Crown Court in the 1990s.

And that is the problem for Mango. Their screwy protocol, their problem, Eisenberg’s advantage. Oh, and the hacker got off, via a jury decision.
unique link to this extract

Broadband customers face up to 14% hike in bills, warns Which? • The Guardian

Mark Sweney:


Broadband bills could surge by as much as £113 [£9.40 per month] next year if a number of the UK’s biggest telecoms firms push ahead with inflation-busting price increases next spring, says consumer watchdog Which?

Many of the country’s main internet providers – including the largest player BT, along with TalkTalk, EE, Plusnet and Vodafone – use a mechanism to increase the cost of bills annually by the rate of inflation as measured by the consumer prices index (CPI) in January, plus 3.9%.

The Bank of England forecasts inflation at just below 10% for January, meaning millions of broadband customers will face a 14% mid-contract increase in their bills.

Which?’s latest broadband survey found that a typical BT customer is facing the largest potential increase of £113 compared with what they were paying in January this year.

Customers of Plusnet, also owned by BT, will face the smallest hike of £87.15 [£7.25 per month] among the five telecoms companies that use the mechanism surveyed by Which?

Given the telecoms companies pushed through inflation-busting rises of around 10% in April, next spring their customers will have seen their bills increase by between £120 and £156 in just two years.

“It is unacceptable that many broadband customers are facing price hikes during an unrelenting cost of living crisis,” said Rocio Concha, director of policy and advocacy at Which? “Customers should be allowed to leave their contract without penalty if prices are hiked mid-contract, regardless of whether or not these increases can be said to be ‘transparent’.”


It’s also wild that broadband companies, which truly aren’t reliant on workers toiling in the data factories to turn out stuff (sure, they have workers who want pay rises, but they aren’t their total opex) should raise prices like this. My advice: call them and say you’re looking to change. Weirdly, they’ll want to keep you and when they look down the back of the sofa, they’ll find a special, cheaper offer.
unique link to this extract

Global smartphone market fell 9% as consumers trim spending • Canalys


In Q3 2022, the global smartphone market recorded its third consecutive decline this year, dropping 9% year-on-year, marking the worst Q3 since 2014. The gloomy economic outlook has led consumers to delay purchasing electronic hardware and prioritize other essential spending. This will likely continue to dampen the smartphone market for the next six to nine months.

Samsung retained its leading position with a 22% market share driven by heavy promotions to reduce channel inventory. Apple was the only vendor in the top five to record positive growth, improving its market position further with an 18% share during the market downturn thanks to relatively resilient demand for iPhones. Xiaomi, OPPO and vivo continued to take a cautious approach to overseas expansion given domestic market uncertainty, retaining 14%, 10% and 9% global market shares, respectively. 


By Canalys’s numbers, in the past 15 quarters (since Q1 2019) the smartphone market has only had growth above 5% twice – both in the first half of 2021.

The precise size of the market isn’t important any more, but has become an interesting proxy for worldwide consumer sentiment about discretionary big-ticket (comparatively) purchases – whether bought on monthly repayments or by a single payment.

This data suggests that economically, things haven’t been too good for quite a while, and now are getting worse.
unique link to this extract

Experts grade Facebook, TikTok, Twitter, YouTube on readiness to handle midterm election misinformation • The Conversation

Anjana Susarla:


The 2016 US election was a wake-up call about the dangers of political misinformation on social media. With two more election cycles rife with misinformation under their belts, social media companies have experience identifying and countering misinformation. However, the nature of the threat misinformation poses to society continues to shift in form and targets. The big lie about the 2020 presidential election has become a major theme, and immigrant communities are increasingly in the crosshairs of disinformation campaigns – deliberate efforts to spread misinformation.

Social media companies have announced plans to deal with misinformation in the 2022 midterm elections, but the companies vary in their approaches and effectiveness. We asked experts on social media to grade how ready Facebook, TikTok, Twitter and YouTube are to handle the task.


The experts consulted are Dam Hee Kim, assistant professor of Communication at the University of Arizona; Anjana Susarla, professor of Information Systems, Michigan State University; and Scott Shackelford, professor of Business Law and Ethics, Indiana University.

So consider: they looked at Facebook, TikTok, Twitter and YouTube. How would you grade each, on an A (great) to F (appalling) scale? Now see how the experts think they do. Bear in mind this is only really for US elections; for other countries, things tend to be worse.
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

1 thought on “Start Up No.1898: what AI tells us about human grammar, pricing the ‘moron risk premium’, get rich hacking legally!, and more

  1. I’d claim Noam Chomsky’s language theory is in fact being validated by advances in AI. The article strikes me as very badly reasoned and confused about how different methods can obtain an identical result. A simple example: Airplanes use the exact same underlying physics as birds. But airplanes don’t flap their wings. Someone who argued that airplanes disproved the air theory of bird flight because the wings are fixed, would have misunderstood the underlying theory. Chomsky’s theory does not say there is no data “training” involved in human language learning. He says basically it’s not a huge lookup table, and there are deep structures. This isn’t contradicted by being able to do a linguistic task with a huge lookup table! However, the enormous advances in AI are not just bigger and bigger lookup tables (though they are a key part), but crucially an understanding of how to put all this data into deeper structures. That it’s not the same structures as human use, is like arguing that airplane wings need to flap like birds.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

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