TheChek News Flow

Apple sauce

Yesterday we presented a tidy theory of the Magnificent Seven’s underperformance in the past month or so. Considered as an asset class, the seven are the new defensives stocks — companies with great brands that can grow even in a slowing economy. But the market at the moment, far from playing defence, wants exposure to economic growth through cyclical stocks.

There’s a little problem, though: most of the decline in the seven comes down to one company. Apple is down 14 per cent since Christmas. Tesla is down 11 per cent, but is only a third of Apple’s weight in the S&P 500. The rest of the seven are meandering along, a bit above (Nvidia) or below (Alphabet) the performance of US big caps, generally.

We could repurpose our argument to be just about Apple, which very clearly is a defensive stock, rather than all the Magnificents. But there are specific things going on with the company that explain its even steeper decline.

In April of last year, after another bout of underperformance from the iPhone maker, we wrote a piece called “What’s wrong with Apple?” We considered six explanations:

We were not too impressed with any of those arguments then. Our instincts turned out to be good: the stock rose 50 per cent from when we wrote the piece to the end of the year. But now, all six worries have gotten worse.

Even after the recent sell-off, the stock’s price/earnings valuation is a third higher than it was back in April. The sales growth and AI issues have come together: consumers have not demonstrated wild enthusiasm for AI-enabled phones in general, and the perception that Apple is lagging behind Android on that tech has grown. This casts doubt on the idea that AI will drive a big iPhone upgrade cycle. As Craig Moffett of MoffettNathanson research sums up:

Not only have we not seen any sign of an upgrade cycle . . . we have seen growing evidence that consumers are unmoved by AI functionality (not just Apple’s but indeed everyone else’s as well). Meanwhile, fully agentic AI, the foundation of any real bull case for Apple, seems further away now than it did even five months ago.

The weakness of defensives we discussed yesterday. An ascendant and aggressive Trump increases the odds that Apple will not get a tariff exemption for iPhones (Edison Lee of Jefferies estimates 90 per cent of which are made in China), and makes it more likely that Chinese consumers will become more hostile to the company’s products.

The legal issues are perhaps the most acute risk. The judge in the Google search antitrust case has ruled the company’s payments to Apple for search traffic are illegal. These payments amount to perhaps $16bn or more a year, more than 10 per cent of Apple’s operating earnings in the US. CFRA Research’s Nicholas Rodelli put 60 per cent odds on the legal remedy cutting the payments by at least half.

Apple still looks expensive to us. Let us know what you think.

Readers of The Information are sharply divided on the future value of OpenAI, with roughly one-fourth expecting the value of the ChatGPT maker to soar past $500 billion in three years—and an almost equal number expecting it to plunge below $100 billion. OpenAI was most recently valued at $157 billion

Government is bracing for a possible renationalisation of the UK’s biggest water company

The UK government has approached consultancies about taking the role of special administrator in a sign that ministers are bracing themselves for the imminent renationalisation of Thames Water. 

Consultancies including Teneo, Interpath and EY are among the potential candidates to run a so-called special administration regime, according to people familiar with the process. A SAR is a temporary measure designed to keep services running, and suppliers and staff paid, in the event of a corporate collapse. 

“We are ready now, we could do one [SAR} today, if we had to,” said one official. “Incidentally, being ready for a SAR is also the strongest lever that we as government can have to make sure that another market-led, private-led solution is found.”

Thames Water is struggling under its £19bn debt pile and has warned that it will run out of cash in March unless the High Court signs off a controversial £3bn loan at a hearing in early February.

Another government official said that there had been “informal engagement” with certain consultancies over a special administrator role but no formal interview process.

Steve Reed, the environment secretary, said in October that he had “ruled out nationalisation”.

Officials insist that taking the company into SAR would not technically be a nationalisation despite it being a big state intervention. 

But plunging Thames Water into special administration may be unavoidable if the court blocks the loan agreed with its senior creditors, or if the company runs out of cash earlier than expected. The £3bn loan is controversial because it would carry an interest rate of 9.75 per cent as well as fees and incentives for the existing Thames Water management. 

The agreement is being challenged by a separate group of Thames Water’s junior creditors — which have proposed a cheaper deal — and by environmental campaigners who argue that the company would be better off in special administration.

The loan would buy the company time to raise at least £3bn of equity in a parallel process. Companies including Castle Water, Covalis and CKI Infrastructure are among investment groups lining up to put in potential bids for the utility.

Bidders and creditors are waiting to see if the company appeals to the Competition and Markets Authority over regulator Ofwat’s decision last month on the level by which water companies can hike customer bills over the next five years. Thames Water has not yet made a decision whether to appeal Ofwat’s decision to the CMA, according to people familiar with the matter.

Ofwat has said that Thames will be allowed to raise bills 35 per cent — much lower than the 59 per cent increase it had sought — taking average bills from around £436 now to £588 between now and 2030. 

The Department for Environment, Food and Rural Affairs, Thames Water and Ofwat did not respond to requests seeking comment.

EY, Teneo and Interpath declined to comment. 

In an update to the market on Wednesday, the company’s chief restructuring officer, Julian Gething, said: “Our plan delivers for customers and stakeholders by unlocking up to £3bn of new money and securing a total of £3.5bn of debt maturity extensions over the next two years and cash releases, so we can continue investing the billions of pounds required to improve our network’s resilience.

“We believe it is the only implementable solution to enable the equity investment required to provide stability and certainty in the longer term and will not impact customer bills.”

The government’s choice of administrator may be complicated by potential conflicts of interest. Teneo is already an adviser to Thames Water and has received £5mn in fees since August 2023. It had also received at least £60mn from running the special administration of collapsed energy supplier Bulb, according to the National Audit Office.

It has also written a report to the High Court supporting the senior creditors’ £3bn loan, while Interpath has written a separate report on behalf of the junior creditors.

Sir Dieter Helm, professor of economic policy at Oxford university, has argued that a SAR would enable Thames Water to focus on a restructuring and on delivering improvements, rather than on negotiating a deal with creditors.

Meta’s chief AI scientist, Yann LeCun, says that a “new paradigm of AI architectures” will emerge in the next three to five years, going far beyond the capabilities of existing AI systems.

LeCun also predicted that the coming years could be the “decade of robotics,” where advances in AI and robotics combine to unlock a new class of intelligent applications.

Speaking in a session dubbed “Debating Technology” at Davos on Thursday, LeCun said that the “flavor of AI” that we have at the moment — that is, generative AI and large language models (LLMs) — isn’t really up to all that much. It’s useful, sure, but falls short on many fronts.

“I think the shelf life of the current [LLM] paradigm is fairly short, probably three to five years,” LeCun said. “I think within five years, nobody in their right mind would use them anymore, at least not as the central component of an AI system. I think [….] we’re going to see the emergence of a new paradigm for AI architectures, which may not have the limitations of current AI systems.”

These “limitations” inhibit truly intelligent behavior in machines, LeCun says. This is down to four key reasons: a lack of understanding of the physical world; a lack of persistent memory; a lack of reasoning; and a lack of complex planning capabilities.

“LLMs really are not capable of any of this,” LeCun said. “So there’s going to be another revolution of AI over the next few years. We may have to change the name of it, because it’s probably not going to be generative in the sense that we understand it today.”

“World models”

This echoes sentiments that LeCun has espoused in the past. At the heart of this is what are coming to be known as “world models” that promise to help machines understand the dynamics of the real world. This includes having a memory, common sense, intuition, reasoning capabilities — traits far beyond that of current systems, which are mostly about pattern recognition.

Previously, LeCun has said this could still be some 10 years away, but today’s estimate brings things closer on the horizon. Though to what extent it will get to in that time frame isn’t exactly clear.

“LLMs are good at manipulating language, but not at thinking,” LeCun said. “So that’s what we’re working on — having systems build mental models of the world. If the plan that we’re working on succeeds, with the timetable that we hope, within three to five years we’ll have systems that are a completely different paradigm. They may have some level of common sense. They may be able to learn how the world works from observing the world and maybe interacting with it.”

“The decade of robotics”

As impressive as generative AI is, capable of passing the bar exam or unearthing new drugs, LeCun reckons that robotics could be a central component of the next wave of AI applications in such real-world scenarios.

Meta itself is doing some research work in the robotics realm, but so is the AI darling of the moment, ChatGPT-creator OpenAI. Earlier this month, new job listings emerged detailing a new OpenAI robotics team focused on “general-purpose,” “adaptive,” and “versatile” robots capable of human-like intelligence in real-world settings.

“We don’t have robots that can do what a cat can do — understanding the physical world of a cat is way superior to everything we can do with AI,” he said. “Maybe the coming decade will be the decade of robotics, maybe we’ll have AI systems that are sufficiently smart to understand how the real world works.”

Davos : World Economic Forum : Debating Technology - Video

--> https://dub.sh/Pqu0k6G

An OpenAI Math Test Raises Eyebrows

When OpenAI previewed o3, its latest reasoning model, in December, one of the ways the company demonstrated its strength was by showing off its 25.2% score on FrontierMath, a test of AI models’ math abilities that Epoch AI developed. (You might remember Epoch, the nonprofit research organization, from their work on AI scaling laws.)

That’s not the kind of grade I would have bragged about after taking a math test, but the problems in FrontierMath are extremely hard. Before the o3 announcement, the top scores were below 2%. But over the past weekend, it came to light that OpenAI had funded the development of the test and had access to the problems and their solutions, sparking skepticism about o3’s score, along with frustration from AI observers.

“It came off as being a totally independent benchmark that was going to be equal to all firms,” said Ozzie Gooen, Executive Director at the Quantified Uncertainty Research Institute, which develops Squiggle AI, a probabilistic software. “A lot of the frustration comes out from people thinking it was an independent thing while it was actually a proprietary set.”

Mathematicians who contributed to the project also expressed concern. “I try to evaluate all my research projects from the point of view of ‘do they contribute at least as much to safety as to capabilities’” of AI models, said one mathematician who contributed. They figured that Epoch had an industry partner, and they were not surprised to learn it was OpenAI, but the industry connection made the research seem less valuable, they said. “Is this even positive for humanity? It’s very unclear,” they said.

For its part, an OpenAI spokesperson said it did not use the benchmark to train its models—the equivalent of snagging a pop quiz from the teacher’s drawer to study in advance. But even without directly training its models on the solutions, access to FrontierMath could help OpenAI improve its models’ performance by making tweaks and checking whether their scores improved.

Another reason to moderate expectations for o3 is that 25% of the problems in FrontierMath are “easy,” meaning a high school math olympiad competitor could solve them. It’s possible this category made up the bulk of the problems that o3 was able to solve. That would still be impressive, but not new territory for AI.

Epoch revealed OpenAI’s involvement on December 20, the day OpenAI announced o3, and revised the FrontierMath paper to credit OpenAI’s involvement. Epoch’s contract prevented them from disclosing OpenAI’s involvement until around that time (this is not the first time OpenAI has faced concerns about restrictive NDAs).

Math Is Hard

Because AI models were already acing existing benchmarking tests, Epoch paid mathematicians $300 to $1,000 to come up with hundreds of new challenging math problems. Here’s one of the “high-medium difficulty” problems that those mathematicians came up with: Construct a degree 19 polynomial p(x) ∈ C[x] such that X := {p(x) = p(y)} ⊂ P1 × P1 has at least 3 (but not all linear) irreducible components over C. Choose p(x) to be odd, monic, have real coefficients and linear coefficient -19 and calculate p(19).

In case you’re curious, the answer is 1876572071974094803391179 (duh).

The full benchmark will eventually contain 300 problems, said Elliot Glazer, who leads the FrontierMath project. Epoch sent OpenAI the first 200 by early December and expects to finish collecting the last 100 within a month or so. At that point, Epoch will randomly choose 50 of those hundred problems to form a “hold out set” that Epoch will withhold from OpenAI so it can’t train its models on them.

Going forward, all benchmarks funded by industry players like OpenAI should retain a hold out set to preserve their ability to independently evaluate models from different companies, advised Glazer.

“We made a mistake in not being more transparent about OpenAI's involvement,” said Epoch’s co-founder and Associate Director Tamay Besiroglu in a statement. “We should have pushed harder for the ability to be transparent about this partnership from the start, particularly with the mathematicians creating the problems.”

OpenAI has already purchased 30 to 50 more problems from Epoch over the next three months, at $12,000 each, said a person who attended a recent talk by an Epoch employee. Epoch is paying contributors $7,500 for each of these extremely hard “tier 4” problems, which are intended to be so hard that solving all of them requires the skills of a full university math department, said Glazer.