FT : $10mn? $30mn? $100mn? The redefinition of the super-rich

$10mn? $30mn? $100mn? The redefinition of the super-rich
The rapid rise in billionaires has disrupted what it means to be part of the moneyed elite

Talk to 10 different wealth industry professionals about when you become super-rich (an ultra-high-net worth individual, or UHNW, in industry parlance) and you will get 10 different answers. For a law firm, it can mean having investable assets — spare cash not tied up in property — of $10mn; for a wealth manager, it can mean having at least $30mn; for an exclusive private members’ club, the hurdle can be as high as $100mn.

What they do agree on, however, is that the base figure is rising, and quickly. The monetary definition has shifted significantly, reflecting not just the growth in wealth globally, but also the changing expectations of what it takes to be considered part of this elite group.

David Gibson-Moore, president of consultancy Gulf Analytica, says the traditional $30mn level “allows for significant investments across multiple asset classes — stocks, bonds, real estate, private equity” — while also furnishing luxuries such as private-jet travel. But, over time, as the financial world has expanded and the accumulation of wealth has accelerated in certain sectors, particularly technology, “the bar for what it means to be ultra-wealthy has risen” he observes. “The $30mn threshold . . . doesn’t carry the same weight or exclusivity it once did. In today’s world, $30mn might secure you a luxurious lifestyle but, in the realms of the ultra-rich, it’s increasingly viewed as just the starting point,” Gibson-Moore adds.

“The ultra-rich today are being measured by new standards, with some financial commentators now suggesting $100mn is the new yardstick for anyone who wants to keep their head held high at private equity parties.”

Charlie Wells, managing director of high-end property buying agency Prime Purchase, agrees: “The dial keeps ticking upwards when it comes to defining ‘UHNW’. Forty years ago, a millionaire with a Rolls-Royce may have been the epitome of wealth. But, thanks to inflation, the numbers are constantly growing. Only recently, someone worth £20mn-plus would have been considered very wealthy but now you need £50mn-plus to be truly UHNW.”

This shift is driven by several factors. First, says Gibson-Moore, is the explosion of new wealth in technology and entrepreneurship. “Over the past two decades, we’ve seen the rise of tech billionaires, cryptocurrency pioneers and venture capitalists who have amassed fortunes at an unprecedented pace,” he says. “The ability to build companies worth billions seemingly overnight has compressed the time it takes to reach UHNW status and these new wealth holders often operate in a different financial universe than the more traditional wealthy class.”

Dominic Volek, group head of private clients at Henley & Partners, which advises wealthy individuals on citizenships and residencies, says: “There has been a jump in wealth creation — and one only needs to look at the tech sector, where billionaires are now common. The diversification into asset classes like cryptocurrencies and NFTs [non-fungible tokens] has also created UHNW individuals almost instantaneously.”

If you take $30mn as the accepted definition for what it takes to be a UHNW, data from consulting group Capgemini shows the number jumped from 157,000 in 2016 to 220,000 last year.

The scope of what super-rich individuals invest in has broadened, too. It is no longer just about having a diversified portfolio; today they might have stakes in disruptive tech start-ups, sustainable ventures or even space exploration. This new frontier of investment requires much larger sums of capital and comes with greater risks — but also offers the potential for exponential returns.

Inflation in luxury assets — such as property, fine art and collectibles — also means it takes far more to maintain a lifestyle traditionally associated with super-rich status. Volek says: “$30mn just doesn’t stretch as far as it did a decade ago.”

A painting by Jean-Michel Basquiat, for example, sold for $57.3mn at an auction in 2016 then again for $85mn six years later. Likewise, the price of entry into exclusive property markets such as Monaco, Mayfair or Aspen in Colorado has soared, with the average price of a house in London’s Grosvenor Square, for example, stretching to around £20mn. The costs of maintaining private aircraft, yachts and other luxury assets have similarly grown, making it far costlier to maintain the hallmarks of ultra-wealth. Experts suggest the annual running cost of a $10mn superyacht can now easily be as much as $1.5mn.

For the owners of R360, an invitation-only private members club, it is clear what the number should be to be considered ultra-rich and eligible for membership: $100mn. Barbara Goodstein, managing partner and chair of the New York chapter of R360, which offers members exclusive investment opportunities, confidential support groups and private getaways, says: “We focus on serving centimillionaires.”

She says, at that level, they get people who are “less focused on short-term investment opportunities and more interested in becoming stewards of wealth”. Goodstein notes that the $100mn threshold has been the criteria for membership at R360 since inception in 2021. “While we don’t anticipate an increase in the near future, we recognise that the average wealth of our members has steadily increased over the past few years and is now more than $400mn.”

The age of members ranges from 28 to 84 but Goodstein adds that R360 is seeing a notable rise in younger members, “with many recently successful entrepreneurs joining in their late twenties and early thirties”.

The question that follows — irrespective of the definition — is why it is so important to the very wealthy to be classified as such. What extra doors does it open?

UHNW individuals receive significantly different treatment because of the scale and complexity of their wealth, Volek adds. “Not only do they have access to better investment opportunities and more diversified portfolios, but they also have dedicated relationship managers who look after them and are available 24/7.”

They often get access to pre-initial public offering deals, private placements and other high-return opportunities that the “broader market doesn’t even know exist”, says Gibson-Moore. For example, Stripe, a fintech payment company, has conducted several rounds of private financing, most notably raising $600mn in 2021 at a valuation of $95bn. This funding round was only available to a select group of institutional investors and the super-rich.

Lifestyle perks can vary. One of the most lavish examples seen by Samuel Wu, chief investment officer of Hong-Kong-based Tridel Capital and co-founder of Chartwell Family Partners, was a European trip given to a super-rich family by a bank in return for their custom.

“More commonly, perks include invitations to concerts, meetings with celebrities [and] successful figures as well as economic leaders,’’ says Wu. ‘‘This is a form of marketing, and these costs are often reflected in the prices being charged. One particular Swiss bank is humorously known to run its entertainment programme better than its banking.” Wu says.

On the flipside, Volek at Henley & Partners says that he also knows of banks “off-boarding” private banking clients with less than $5mn in assets because they were deemed not profitable enough to have as customers. He says it shows that size now really matters in the world of the ultra-rich and that, while the definition might still be up for debate, the level at which you can be classified as UHNW is only going one way: up.

FT : Russia’s shadow fleet grows despite western crackdown

Russia’s shadow fleet grows despite western crackdown
Moscow’s capacity to transport oil on old underinsured ships has increased by 70% since last year, says report

Russia has expanded the capacity of its shadow fleet of oil tankers by almost 70 per cent year-on-year despite a recent crackdown on insurers and shipping companies enabling Moscow to circumvent western sanctions, new research has shown.

The volume of Russian oil transported by poorly maintained and underinsured tankers has increased from 2.4mn barrels per day in June 2023 to 4.1mn in June 2024, according to a report published by the Kyiv School of Economics (KSE) on Monday.

The trend comes as the US, Canada, Japan and European allies increasingly targeted global insurers and ship owners in a bid to crack down on Moscow’s ability to generate revenues for its war in Ukraine. They also added to the sanctions list companies and individual vessels associated with the Russian shadow fleet. 

“Sanctions on tankers have been quite effective but the designation campaign has been too limited to actually rein in Russia’s shadow fleet,” said Benjamin Hilgenstock, one of the authors of the KSE report.


He added that sanctions should be used “systematically” to enforce a requirement for adequate oil spill insurance and, thereby, “address the serious and urgent environmental threat stemming from the shadow fleet”.

Many of these vessels regularly navigate busy European waters, including the Baltic Sea, the Danish Straits and the Strait of Gibraltar, increasing the risk of environmental disasters for the EU and neighbouring countries.

KSE proposes the establishment of “shadow-free” zones in European waters to mitigate those risks. Otherwise a disaster is just “waiting to happen on Europe’s doorstep,” the report argues. “The weak link in the regulatory framework, together with the dramatically expanded role of shadow tankers in the Russian oil trade, means that a major environmental disaster is only a question of time.”

In June 2024, 70 per cent of Russia’s seaborne oil was transported by the shadow fleet, which Russia is estimated to have spent $10bn assembling, according to the KSE. This included 89 per cent of Russia’s total crude oil shipments, most of which traded above the $60 per barrel price cap since mid-2023, and 38 per cent of Russia’s oil product exports.

By assembling this fleet, Moscow severed ties with price cap coalition countries that pressured global insurance companies to comply with the sanctions regime, reducing Russia’s options to primarily domestic insurers.

This raised significant concerns about such coverage's quality, reliability, and scope. The combination of the tankers’ age — which averages 18 years — and lack of adequate insurance makes these vessels extremely hazardous, the KSE report argues.

The risks are further heightened by the fleet’s use of grey-listed flag states anonymous ownership structures. They often include multiple intermediaries such as British accountants and Dubai-based companies hidden through layers of corporate entities, the FT's recent investigation found.

KSE report authors argued that in the event of trouble, European states could face clean-up costs running into billions of euros.

Several accidents involving shadow ships associated with Russia have already occurred. This March, a 15-year-old shadow tanker Andromeda Star collided with another vessel near Denmark. No oil was spilt as it was on the way to Russia, unloaded.

Over the past two years, four Russian shadow fleet vessels have lost engine power, including incidents in the Dardanelles and the Danish Straits.

Shadow fleet vessels used to carry oil from other sanctioned sellers also experienced engine breakdowns, highlighting maintenance issues and explosions. In May 2023, a 27-year-old Gabon-flagged ship with 700,000 barrels capacity used to transport Iranian oil suffered a large explosion near Indonesia. It was empty at the time.

Several shadow fleet vessels have been involved in oil spills, with some fleeing the scene after causing environmental damage. In 2019, the 23-year-old Ceres I, previously involved in the Iran oil trade, collided with another tanker near Singapore, turned off its signal, and tried to flee before being caught by Malaysia’s coastguard.

FT : Europe searches for scale in race to boost ammunition production

Europe searches for scale in race to boost ammunition production
Efforts to supply Ukraine and replenish inventories are hampered by bottlenecks and under-investment, say industry experts

The war in Ukraine has an unlikely new frontline. A new munitions factory in Queensland, Australia, jointly owned by Germany’s Rheinmetall and a local contractor NIOA, is churning out tens of thousands of artillery shells for Kyiv. 

Work at the Maryborough factory, the first such facility to be built in Australia since the second world war, has barely stopped since it opened two years ago. Although it took time to secure the workforce, “now we are flying,” said Robert Nioa, chief executive of NIOA. 

The company, which exports the shells to Germany where they are then filled with explosives by Rheinmetall, plans to increase annual production at the facility by 25 per cent to around 55,000 shells next year. Nioa says it could produce more than 100,000 shells a year with more capital investment. 

Like NIOA, other defence companies globally have boosted output of everything from ammunition to rocket motors and missiles to replenish national stockpiles which have been depleted as governments have shipped arms to Ukraine.

Some European arms producers have made progress over the past two and a half years, often investing ahead of securing government contracts. Tom Waldwyn, research associate for defence procurement at the International Institute for Strategic Studies, said in some instances European companies had expanded artillery ammunition production “by as much as 10 times what they were producing before the war”. 

Rheinmetall has plans to increase output of Nato-standard 155mm ammunition from around 100,000 rounds before February 2022 to 1.1mn rounds per year from 2027. Sweden’s Saab said the capacity of its ground combat business, which includes ammunition, had doubled to 200,000 units a year in recent years — and it is on course to double that again to 400,000 in the near future. 

Thales UK, the British subsidiary of the French defence and technology group, has similarly increased capacity at its facilities in Belfast, Northern Ireland where it builds the Starstreak short-range air defence system and assembles the Saab NLAW anti-tank system. The company plans to double capacity over the next two years and then double again by 2028. 

But overall, efforts are still being hampered by supply chain constraints, and industry stakeholders say that much greater investments are needed if Europe is to be able withstand Russia’s aggression while also refilling depleted stockpiles.

Some key bottlenecks remain, notably for raw materials such as cotton linters which are needed to produce nitrocellulose used in artillery shells and other explosives. 

There are “significant supply chain issues with munitions supply for Australia and the Allied effort,” said Nioa, including “nitrocellulose . . . which goes into propellant”. 

Rheinmetall said it had increased the “safety stock” for certain raw materials such as “cotton linters and armoured steel to around three years of annual production” in order to de-risk its supply chains. 

Norway’s government said last week it would invest close to NKr1bn (£70mn) to boost production of critical explosives and rocket motors as part of a plan for its defence industry.

It will co-finance a feasibility study with Chemring Nobel, the Norwegian subsidiary of Britain’s Chemring, to assess the development of an explosives facility. The funding also includes plans for a new rocket motor production line for Nammo, co-owned by the Norwegian and Finnish governments. The company produces rocket motors for air defence missiles among other things. 

Rocket motors, said Vegard Sande, director of large calibre systems at Nammo, “are not the only bottleneck for increasing the number of missiles for air defence but it’s one of the critical components where we have to increase capacity”.

Although companies have now secured contracts from governments to boost capacity, ramping up production takes time. “It takes two to three years from an investment decision before you see a production increase,” said Sande. 

And as Ukraine confronts its third winter of war, the industry says more needs to be done to make up the gulf between supply and demand.

Jan Pie, secretary-general of ASD, the industry trade body, said that production levels were still short of what was required, especially following decades of under-investment in Europe’s industrial base.

“If the objective is to provide adequate support to Ukraine to withstand Russia’s aggression and to build up sufficient military capacities to deter Russia from further aggression, we have to procure and to produce in Europe much more and much faster,” he said.

Russia, on the other hand, is operating on a war footing. Thanks to support from North Korea, Moscow can currently expend around 10,000 rounds of ammunition per day with no fear of depleting its stockpiles, according to a recent report by the Kiehl Institute for the World Economy. At a similar rate of fire, Germany would use up a year’s worth of its entire ammunition production within 70 days, it said. 

The Russians, said Guntram Wolff, senior fellow at Bruegel think-tank and co-author of the study, “have large factory lines . . . massive assembly lines, where the numbers that are churned out are frighteningly high”. 

The west needed to step up, he added. “This is about building industrial capacities at scale and being able to produce for years at scale.”

But in order to deliver at scale — and over time — executives said the industry needs much longer contracts to invest. 

“If we are going to raise the necessary capital and do heavy investments based on order intake we need a 10-year perspective,” said Nammo’s Sande. 

Micael Johansson, chief executive of Saab, echoed the point. Industry, he said, “needs more long-term commitment from nations as it is taking on risk ramping up capacity without guaranteed long-term contracts”.

FT : Corporate debts mount as credit funds let borrowers defer payments

Corporate debts mount as credit funds let borrowers defer payments
Use of payment-in-kind loan terms is growing as companies struggle with heavy leverage and high interest rates

A growing list of cash-strapped companies have turned to their lenders at private credit funds for relief in recent months, seeking to conserve capital by delaying payments on their debt.

The rate at which companies are opting to increase their principal balance instead of paying cash, known as “payment-in-kind” or PIK, edged higher during the second quarter, according to a recent report from rating agency Moody’s. These types of loans have a catch: while they provide temporary relief, they often come with a higher interest rate on a mounting debt load as the deferred payments pile up. 

The publicly traded private credit funds that the rating agency keeps tabs on reported the highest levels of PIK income since it began tracking the data in 2020 — though the income is paper profit and it is not clear how much of the gains will actually be realised.

The growth in these types of loans is one signal of stress in corporate America even as the broader economy expands, particularly for businesses that were leveraged to the hilt by their private equity owners and are now struggling with those interest burdens.

“PIK was born out of necessity and is something that people in the market viewed as a temporary situation,” said Sheel Patel, a partner at King & Spalding, referring to recent high interest rates. “Even though the interest rate environment has changed, I think that sponsors will continue . . . trying to keep it in their documents.”

While the Federal Reserve’s move to cut interest rates in September is the first step in alleviating pressure on borrowers, rates are expected to remain far higher than the rock-bottom level they hit in the immediate aftermath of the Covid-19 pandemic, when private equity firms went on a debt-fuelled buyout binge.

Companies could afford the interest burden then. But as rates surged above 5 per cent, interest costs for many highly levered businesses began to eat up most of their cash. That financial distress is showing up, in part, through the swell of PIK income reported by credit funds, fuelling worries that some borrowers are wrestling to stay afloat.

The shift to PIK borrowing is just one risk being borne by the burgeoning private credit industry, where asset managers lend directly to businesses. The loans — while risky — can generate lucrative returns to the lenders who are willing to provide the capital.

Moody’s estimated that 7.4 per cent of the income reported by private credit funds was in the form of PIK during the most recent quarter. Analysts at Bank of America pegged the figure at 9 per cent and said its analysis showed that these funds had gone one step further: 17 per cent of the loans they hold give the borrower an option to pay at least part of their interest with more debt going forward, even if they are not doing so now.


In the second quarter, Blue Owl’s technology fund reported that 23.6 per cent of the income it was earning was in the form of PIK. It was followed closely by Prospect Capital at 18.6 per cent, New Mountain Finance at 17.7 per cent and Ares Management’s ARCC — one of the largest of the funds — at 15.4 per cent.

Prospect did not respond to a request for comment. New Mountain declined to comment.

Beauty Industry Group, a maker of hair extensions backed by private equity group L Catterton, was among the companies seeking to reduce its cash interest burden this year. Its lender, Blue Owl, agreed to take roughly a fifth of its interest payments in PIK. In exchange for the option, the company’s overall interest bill went up.

The same was true for Avalign Technologies, which sought flexibility from Ares this year. The manufacturer of medical implants amended its loan so that, while it would pay more over time, it could put off roughly 30 per cent of its interest payments.

While PIK income is counted as income each quarter, the funds do not receive cash payments until the loan is refinanced or matures. That can create a liquidity crunch for funds, which are required to pay out 90 per cent of their income to investors, even when they have not received cash on those debts.


The uptick in PIK income itself can make investment income at these funds look more attractive, even though the companies bearing the debts are struggling — and the funds are not yet getting paid in full.

PIK is not always a worrying sign, said Clay Montgomery, a Moody’s analyst. Some funds offer PIK to allow healthy businesses to direct their cash towards expansion plans. But it can be difficult for investors to discern when PIK is being extended to give a lifeline during a time of financial stress, or ambition.

For investors, understanding the difference is crucial. Lenders said that if built into a loan at the start, PIK did not indicate stress. Ares said that more than 90 per cent of second quarter PIK income at one of its funds was structured at the start of the investment. Blue Owl said more than 90 per cent of the loans in its technology fund that can defer payment were structured that way from the start.

Because PIK income compounds at a higher rate than cash interest payments, it can be a boon to lenders. But by the same token, once it starts rising, it can become too onerous for some companies to pay back. It can also be used as a tool to avoid impairing debt until a later date, with some investors worried it allows companies to kick the can on what will still end with a default and resulting loss.

Khoros, a software company owned by Vista Equity Partners, is one such company that struggled to repay its PIK interest bill as its business deteriorated. Earlier this year it began to defer the entirety of its debt payments — at an interest rate of more than 16 per cent — and several lenders marked it as a troubled loan.

Vista and L Catterton declined to comment. Linden Capital Partners, the owner of Avalign, did not respond to a request for comment.

WSJ : Used EVs Sell for Bargain Prices Now, Putting Owners and Dealers in a Bind

Used EVs Sell for Bargain Prices Now, Putting Owners and Dealers in a Bind
Sharp drop in the value of preowned electric vehicles stands in contrast to the broader used-car market

Electric cars have gone from pricey purchases to some of the biggest bargains on the used-car lot, as resale values for the vehicles have tumbled.

Two years ago, some used electric-car models were selling for as much or more than new ones, because of a supply-chain crisis that resulted in a broader car shortage.

Now, the dynamic has flipped. Dealer lots are full of unsold EVs, and car companies—largely led by Tesla—have slashed prices on new models in an effort to sell them.

The discounts on new EVs have helped prop up flagging sales. Financing deals elevated Tesla’s global deliveries in the third quarter, reversing declines earlier in the year.

But the flood of discounts has caused prices for previously owned electric vehicles to plunge, adding to the challenges confronting the auto industry as its big bet on battery technology continues to sputter.

In September, the average selling price of a three-year-old electric vehicle was about $28,400, less than that of a gas-engine vehicle of the same age and a 25% drop from the start of 2023, according to car-shopping website Edmunds. The sharp fall in the price of preowned EVs stands in contrast to the broader used-car market, where values have stayed steady during the same period, according to the firm’s data.

The sharp decline in prices of used EVs could potentially broaden their appeal to budget-conscious buyers, analysts say. At the same time, many current owners who paid a premium for their cars are now upside down on their loans, owing more than the car is worth.

Among the hardest hit are some Tesla owners.

The world’s top seller of EVs last year cut prices by as much as one-third on some new models in the U.S., leading other automakers to respond with their own markdowns. On top of that, rental-car company Hertz in January dumped a large chunk of its Tesla fleet into the used-car market, further depressing values, analysts and dealers say.

Within the past year, the average list price for a used Model 3 and Model Y—Tesla’s most popular models—dropped about 25%, according to data from car-listing website CarGurus.

“They kept reducing the price of the cars, which killed the used-EV market,” said Christian Lange, the owner of a 2018 Tesla Model 3, which he had purchased brand new. In early 2023, the vehicle was valued at around $35,000, roughly equivalent to what he owed on his loan.

Normally, buyers gradually build equity in their vehicles as they make payments. But the reverse happened to Lange, because sharp price reductions on new Teslas made his vehicle worth less as well. At the start of this year, the value of Lange’s Tesla had dropped so much it was worth $10,000 less than what he owed on his loan.

Frustrated with Tesla, Lange said he traded his Model 3 for a Kia EV9. Tesla didn’t return a request for comment.

As buyer interest in EVs has continued to subside this year, automakers have become more aggressive with their sales promotions for new models, offering cheap leases and low-interest-rate financing deals that are sapping demand from the used-car lot.

Some brands are lowering payments on new EVs by applying a $7,500 tax credit available for such purchases directly to the lease deal.

As a result, the average monthly payment on an electric-car lease has fallen from $950 at the start of last year to $582 in August, according to Edmunds. That figure would be roughly comparable to what one would pay each month on a loan for a $28,000 used EV, said Ivan Drury, the firm’s director of insights.

“The biggest competition for one- or two-year-old used vehicles is across the lot. It is the brand-new version,” Drury said, referring to previously owned EVs.

Leasing has quickly become the most popular method of buying a new EV. At the start of last year, 16% of electric-vehicle sales at dealerships were leases, according to Edmunds. Today, the figure is nearly 80%.

Auto executives admit they misjudged the market, and that miscalculation is taking a toll on the used-EV prices. To limit supplies, car companies are pulling back on their ambitions, including reducing production and delaying expansion plans. Some are also taking precautions with their EV leases to preserve resale values.

“We don’t want to end up in a situation where we’re just producing to a target, and the demand isn’t there,” said General Motors Chief Financial Officer Paul Jacobson on a call with analysts in July. “You’ve got residual value implications which stay with you for a long, long time.”

Prices in the preowned market have also been hurt by a new federal tax credit for used-EV purchases. The $4,000 credit, introduced early last year, only applies to electric cars selling for under $25,000, a requirement that has led some dealers to cut the price to get under that amount, analysts say.

Dave Katarski, operations chief for Feldman Automotive Group, a large dealer group in Michigan, said his stores have been selling more than 200 used electric cars a month, a significant increase from a year ago.

The used-EV tax credit has helped stimulate interest among consumers, along with an influx of cheap Teslas previously owned by Hertz, he added. “We’re seeing a ton of people flock to them when they can afford them,” Katarski said.

There are also some troubling signs. An Edmunds study of EV trade-in values showed that in August, owners on average owed about $10,000 more than the car was worth—up from about $8,000 at the start of 2023.

The abrupt rise in electric-car leasing could also have longer-term implications, said Kevin Roberts, head of industry insights at CarGurus.

While helping to alleviate inventory buildups now, it is likely to result in a flood of gently used EVs returning to the market in two to three years. This potential wave could put more downward pressure on prices, Roberts said. Automakers and lenders, who own the leased vehicles, would be on the hook for any losses incurred by steeper-than-expected drops in resale values.

“There’s no positive way to put this,” said Edmunds’s Drury, adding that 77% of EVs from dealerships were leased. “The problem is those are coming back.”

TechCrunch : The promise and perils of synthetic data

The promise and perils of synthetic data

Is it possible for an AI to be trained just on data generated by another AI? It might sound like a harebrained idea. But it’s one that’s been around for quite some time — and as new, real data is increasingly hard to come by, it’s been gaining traction.
Anthropic used some synthetic data to train one of its flagship models, Claude 3.5 Sonnet. Meta fine-tuned its Llama 3.1 models using AI-generated data. And OpenAI is said to be sourcing synthetic training data from o1, its “reasoning” model, for the upcoming Orion.

But why does AI need data in the first place — and what kind of data does it need? And can this data really be replaced by synthetic data?
The importance of annotations

AI systems are statistical machines. Trained on a lot of examples, they learn the patterns in those examples to make predictions, like that “to whom” in an email typically precedes “it may concern.”

Annotations, usually text labeling the meaning or parts of the data these systems ingest, are a key piece in these examples. They serve as guideposts, “teaching” a model to distinguish among things, places, and ideas.

Consider a photo-classifying model shown lots of pictures of kitchens labeled with the word “kitchen.” As it trains, the model will begin to make associations between “kitchen” and general characteristics of kitchens (e.g. that they contain fridges and countertops). After training, given a photo of a kitchen that wasn’t included in the initial examples, the model should be able to identify it as such. (Of course, if the pictures of kitchens were labeled “cow,” it would identify them as cows, which emphasizes the importance of good annotation.)

The appetite for AI and the need to provide labeled data for its development have ballooned the market for annotation services. Dimension Market Research estimates that it’s worth $838.2 million today — and will be worth $10.34 billion in the next ten years. While there aren’t precise estimates of how many people engage in labeling work, a 2022 paper pegs the number in the “millions.”

Companies large and small rely on workers employed by data annotation firms to create labels for AI training sets. Some of these jobs pay reasonably well, particularly if the labeling requires specialized knowledge (e.g. math expertise). Others can be backbreaking. Annotators in developing countries are paid only a few dollars per hour on average without any benefits or guarantees of future gigs.

A drying data well
So there’s humanistic reasons to seek out alternatives to human-generated labels. But there are also practical ones.
Humans can only label so fast. Annotators also have biases that can manifest in their annotations, and, subsequently, any models trained on them. Annotators make mistakes, or get tripped up by labeling instructions. And paying humans to do things is expensive.

Data in general is expensive, for that matter. Shutterstock is charging AI vendors tens of millions of dollars to access its archives, while Reddit has made hundreds of millions from licensing data to Google, OpenAI, and others.
Lastly, data is also becoming harder to acquire.

Most models are trained on massive collections of public data — data that owners are increasingly choosing to gate over fears their data will be plagiarized, or that they won’t receive credit or attribution for it. More than 35% of the world’s top 1,000 websites now block OpenAI’s web scraper. And around 25% of data from “high-quality” sources has been restricted from the major datasets used to train models, one recent study found.

Should the current access-blocking trend continue, the research group Epoch AI projects that developers will run out of data to train generative AI models between 2026 and 2032. That, combined with fears of copyright lawsuits and objectionable material making their way into open data sets, has forced a reckoning for AI vendors.

Synthetic alternatives
At first glance, synthetic data would appear to be the solution to all these problems. Need annotations? Generate ’em. More example data? No problem. The sky’s the limit.
And to a certain extent, this is true.
“If ‘data is the new oil,’ synthetic data pitches itself as biofuel, creatable without the negative externalities of the real thing,” Os Keyes, a PhD candidate at the University of Washington who studies the ethical impact of emerging technologies, told TechCrunch. “You can take a small starting set of data and simulate and extrapolate new entries from it.”
The AI industry has taken the concept and run with it.
This month, Writer, an enterprise-focused generative AI company, debuted a model, Palmyra X 004, trained almost entirely on synthetic data. Developing it cost just $700,000, Writer claims — compared to estimates of $4.6 million for a comparably-sized OpenAI model.

Microsoft’s Phi open models were trained using synthetic data, in part. So were Google’s Gemma models. Nvidia this summer unveiled a model family designed to generate synthetic training data, and AI startup Hugging Face recently released what it claims is the largest AI training dataset of synthetic text.
Synthetic data generation has become a business in its own right — one that could be worth $2.34 billion by 2030. Gartner predicts that 60% of the data used for AI and an­a­lyt­ics projects this year will be syn­thet­i­cally gen­er­ated.
Luca Soldaini, a senior research scientist at the Allen Institute for AI, noted that synthetic data techniques can be used to generate training data in a format that’s not easily obtained through scraping (or even content licensing). For example, in training its video generator Movie Gen, Meta used Llama 3 to create captions for footage in the training data, which humans then refined to add more detail, like descriptions of the lighting.
Along these same lines, OpenAI says that it fine-tuned GPT-4o using synthetic data to build the sketchpad-like Canvas feature for ChatGPT. And Amazon has said that it generates synthetic data to supplement the real-world data it uses to train speech recognition models for Alexa.
“Synthetic data models can be used to quickly expand upon human intuition of which data is needed to achieve a specific model behavior,” Soldaini said.

Synthetic risks
Synthetic data is no panacea, however. It suffers from the same “garbage in, garbage out” problem as all AI. Models create synthetic data, and if the data used to train these models has biases and limitations, their outputs will be similarly tainted. For instance, groups poorly represented in the base data will be so in the synthetic data.
“The problem is, you can only do so much,” Keyes said. “Say you only have 30 Black people in a dataset. Extrapolating out might help, but if those 30 people are all middle-class, or all light-skinned, that’s what the ‘representative’ data will all look like.”
To this point, a 2023 study by researchers at Rice University and Stanford found that over-reliance on synthetic data during training can create models whose “quality or diversity progressively decrease.” Sampling bias — poor representation of the real world — causes a model’s diversity to worsen after a few generations of training, according to the researchers (although they also found that mixing in a bit of real-world data helps to mitigate this).

Keyes sees additional risks in complex models such as OpenAI’s o1, which he thinks could produce harder-to-spot hallucinations in their synthetic data. These, in turn, could reduce the accuracy of models trained on the data — especially if the hallucinations’ sources aren’t easy to identify.
“Complex models hallucinate; data produced by complex models contain hallucinations,” Keyes added. “And with a model like o1, the developers themselves can’t necessarily explain why artefacts appear.”
Compounding hallucinations can lead to gibberish-spewing models. A study published in the journal Nature reveals how models, trained on error-ridden data, generate even more error-ridden data, and how this feedback loop degrades future generations of models. Models lose their grasp of more esoteric knowledge over generations, the researchers found — becoming more generic and often producing answers irrelevant to the questions they’re asked.
Image Credits:Ilia Shumailov et al.
A follow-up study shows that oher types of models, like image generators, aren’t immune to this sort of collapse:
Image Credits:Ilia Shumailov et al.
Soldaini agrees that “raw” synthetic data isn’t to be trusted, at least if the goal is to avoid training forgetful chatbots and homogenous image generators. Using it “safely,” he says, requires thoroughly reviewing, curating, and filtering it, and ideally pairing it with fresh, real data — just like you’d do with any other dataset.

Failing to do so could eventually lead to model collapse, where a model becomes less “creative” — and more biased — in its outputs, eventually seriously compromising its functionality. Though this process could be identified and arrested before it gets serious, it is a risk.
“Researchers need to examine the generated data, iterate on the generation process, and identify safeguards to remove low-quality data points,” Soldaini said. “Synthetic data pipelines are not a self-improving machine; their output must be carefully inspected and improved before being used for training.”

OpenAI CEO Sam Altman once argued that AI will someday produce synthetic data good enough to effectively train itself. But — assuming that’s even feasible — the tech doesn’t exist yet. No major AI lab has released a model trained on synthetic data alone.

At least for the foreseeable future, it seems we’ll need humans in the loop somewhere to make sure a model’s training doesn’t go awry.

FT : Pakistan security services pressured utilities over government power deal

Pakistan security services pressured utilities over government power deal
Talks that led to expected state savings of nearly $1.5bn described as ‘more an execution than a negotiation’

Pakistan’s powerful security services used heavy pressure to coerce five local utility companies to end electricity supply contracts with the government early, according to people familiar with the talks.

Pakistan’s power ministry has said agreements that were announced on Thursday to end the contracts will save the cash-strapped government Rs411bn ($1.48bn) and help it cut electricity prices for households and businesses.

Prime Minister Shehbaz Sharif’s office said the power companies had “prioritised national interest over personal interest” and “voluntarily agreed” to the terminate their contracts.

But energy sector businesspeople said the agreement with the five publicly listed “independent power producers” followed weeks of pressure from security services.

“We will go to any measure even beyond our imaginations to get the issue settled,” one military officer told an energy executive in a text message seen by the Financial Times. “Time has come to give a final blow to such IPPs.”

Senior executives were called to meetings with senior security officials, according to three people in the energy industry familiar with the conversations. Nadeem Anjum, head of the Inter-Services Intelligence, Pakistan’s powerful spy agency, attended some of the meetings before he retired in late September, they said.

One businessman involved in the process said the talks had been more an “execution than a negotiation”. Security service and government officials threatened to investigate energy investors’ ventures in other sectors if they did not comply with the government’s demands, said the businessman, who like others familiar with the talks asked not to be identified because of their sensitivity.

“Coercion and threats worked. At the end, all sponsors and investors are human and take decisions to ensure their physical and business interests’ wellbeing,” he said.

“The negotiations took place in a cordial and constructive environment, and the allegations of harassment are completely unfounded and baseless,” Pakistan’s power ministry said in statement. The Pakistan Armed Forces also denied any use of threats or intimidation.

The share prices of the five utilities slumped over the past month as investors anticipated the premature demise of their contracts.

Hub Power Company, the country’s largest energy producer, agreed to end early a contract under which the government had committed to buy electricity from one of its power plants until 2027.

In a statement to Pakistan’s stock exchange on Thursday, Hub Power, which is also a partner for a number of Chinese ventures in the country including electric vehicle giant BYD, said its decision was made “in the greater national interest”.

By close of trade on Friday, shares in Hubco had fallen more than 30 per cent since September 18, while those in Lalpir Power, another utility that agreed to end its contract early, were down 32 per cent.

In order to end widespread electricity shortages a decade ago, the Pakistani government used promises of sovereign-backed, dollar-indexed returns as well as purchase commitments to attract billions of dollars from lenders into the country’s power sector.

The move eased crippling blackouts. However, power tariffs in Pakistan have more than doubled over the past three years, as the heavily indebted government cut subsidies and passed capacity payments for about 40,000MW of installed generating capacity — much of it sitting idle — on to consumers.

The surge in electricity bills to some of the highest levels in the region turned independent power producers into public villains and spawned protests demanding their lucrative contracts be cancelled.

In August, Sharif appointed a task force co-ordinated by a military general to find solutions to the country’s spiralling power costs.

Awais Leghari, Pakistan’s power minister, told the Financial Times that the government and the power companies held multiple talks to revise the terms of the agreements and to take into account the companies’ objections.

There was a shared understanding between the parties that a solution was needed “to keep the entire power sector from going bankrupt”, he said, adding: “In spite of the termination of the contracts, they [the power companies] will have still made far higher returns than they would have in any other country.”

He has said the government is still negotiating with other power producers to revise their contracts.

The tough tactics are the latest sign of the creeping influence of Pakistan’s military in managing the crisis-stricken country’s turbulent economic affairs, analysts say.

“Power sector debts are ruining the country’s finances . . . and the military didn’t trust that the civilians, with their own ties to the power industry, could get a deal done,” said Ayesha Siddiqa, author of Military Inc, a book on the military’s business affairs, and a senior fellow at King’s College, London.

But analysts warned the state’s approach risked deterring investors from taking part in the government’s planned privatisation of Pakistan’s debt-laden flag-carrier airline and power distribution companies.

“This gain has come at the cost of breaking investors’ trust,” said Uzair Younus, a principal at The Asia Group consultancy in Washington, adding he believed the savings would be much less than the government expected.

“However, the military will chalk this up as a success, meaning that they’ll increase their interventions even more in the months to come,” Younus said.

FT : US deploys advanced antimissile system in Israel

US deploys advanced antimissile system in Israel
Rare move will involve sending American forces to operate THAAD battery for defence against further Iranian missile attacks


The US is sending an advanced anti-missile system to Israel and 100 troops to operate it in a rare move that will boost its ally’s defences as Benjamin Netanyahu’s government plans retaliatory strikes against Iran.

President Joe Biden authorised the Pentagon to send a Terminal High-Altitude Area Defense (THAAD) battery to “bolster Israel’s air defences” following two missile barrages from Iran earlier this year, the US defence department said on Sunday.

Washington will also send approximately 100 soldiers to operate the missile system within Israel, according to a US defence official. The THAAD, a ground-based system designed to shoot down ballistic missiles, can defend a larger area than the more common Patriot system.

The move comes ahead of expected Israeli retaliation against Iran’s ballistic missile attack against the Jewish state on October 1. Iran fired 180 missiles in response to the assassination of Hizbollah leader Hassan Nasrallah late last month and of Hamas’s political leader Ismail Haniyeh in Tehran in July.

Washington has backed Israel’s right to retaliate but Biden has been urging Netanyahu not to hit Iranian nuclear sites or its oil infrastructure. Israeli defence minister Yoav Gallant has vowed his government’s response will be “deadly, precise and — above all — surprising”.

Biden and Netanyahu spoke on the phone on Thursday, a conversation the White House said involved an “honest” and “productive” discussion of retaliation plans.

Many of Iran’s missiles were intercepted, but the assault was regarded as far more severe than a more telegraphed attack in April.

During that round of tit-for-tat exchanges, limited damage was caused and the Biden administration helped contain the escalation as Israel responded with a missile attack on a military base near the Iranian city of Isfahan.

The scale and size of Israel’s response to this month’s barrage will determine how Iran reacts, analysts say, with the fear being that the region is sliding to all-out war, with counterstrike followed by counterstrike.

Israel’s options include targeting the Islamic republic’s nuclear plants, considered the most extreme scenario, its oil infrastructure or military facilities.

Since Hamas’s October 7 attack on Israel triggered the war in Gaza and a wave of regional hostilities last year, Washington has dispatched additional US troops and military equipment, including warships, fighter jets and air defences, to help defend Israel and deter Iran.

Major General Pat Ryder, Pentagon spokesperson, said the THAAD deployment underscored America’s “ironclad commitment to the defence of Israel”, and was part of wider regional changes made in recent months.

Israel has its own sophisticated anti-missile defences, including its Iron Dome, David’s Sling and Arrow systems, but US support is considered vital to protecting the country from large missile and drone barrages from Iran.

The US last sent a THAAD battery after the October 7 attack last year having previously deployed one to Israel in 2018 for training and a military exercise.