University humanities departments are at risk; AI is turning homework assignments into prompting exercises; and institutions from the Ivy League on down are reevaluating the basic principles of higher education. And yet The Wall Street Journal reports one of Harvard's hottest courses is a demanding, digital-free deep dive into great works of literature and philosophy. Law professor Anton Sorkin notes the "human craving for formation through friction" at work, while founder Sara Holoubek sees "a good reminder of something America tends to forget: to be human is to be well-rounded."
AI's accelerating growth collided with mounting questions about its economics this week. Anthropic's IPO filing and Big Tech's earnings put the industry's massive infrastructure commitments, profitability and long-term sustainability under the spotlight.
McKinsey, Bain and Boston Consulting Group are offering higher starting pay to new consultants next year, according to salary data compiled by industry watcher Management Consulted and confirmed by multiple outlets. The reported base salaries are $200,000 for MBA hires and $120,000 for hires without an MBA — gains of $8,000 to $10,000 not including bonuses. The move represents the first reported increase since 2022 — but the gains still trail inflation-fueled price jumps that have happened over the same period.
Halloween is a $13.5 billion holiday built around handing strangers free candy.
On paper, economists should hate it.
Research (NRF) says Americans will spend $13.5B this year. $4.4B on costumes. $4.3B on decorations. $4.1B on candy. Another $920M putting dogs in hot dog costumes. That's about $115 a head for one night.
But spending isn't the same as benefit. Most of that money would've gone somewhere else anyway. A 12-foot skeleton in October is a patio heater that didn't get bought in November. Retail wins. The economy mostly shuffles.
Then there are the costs. In the UK alone, roughly 7 million costumes get binned every year, and 83% of the material is oil-based plastic. That's about 2,000 tonnes of polyester witches headed for landfill. And your candy money is shrinking: Bankrate figures it buys about 40% less candy by weight than it did in 2020.
So on a spreadsheet? Halloween's a wash at best.
Except the spreadsheet is measuring the wrong thing.
The classic economics knock on Christmas is that gifts destroy value. Your aunt spends $50 on a sweater you'd pay $20 for. Halloween mostly dodges that trap. You pick your own costume. You decorate your own porch. Nobody's guessing what you want.
And the real output was never the candy. It's the kid who's been planning a dinosaur costume since July. It's neighbors who ignore each other 364 days a year standing on the sidewalk comparing jack-o'-lanterns.
GDP counts the pumpkin. It doesn't count the grin.
Can you put a price on fun? Sure. $115 is the floor. People pay it voluntarily, which means the night's worth more than that to them. Everything above the receipt is value nobody books.
Now the funny part. We call it the most American holiday there is. It isn't. It's Irish and Scottish, carried over by immigrants, then supersized and shipped back out.
Germany barely noticed it before the 1990s, and now it shares Oct 31 with Reformation Day. Japan got it through Tokyo Disneyland, kept the costumes, skipped the trick-or-treating, and turned Shibuya into a street party so big the ward banned nighttime street drinking year-round. This year they'll fine you ¥2,000 for dropping a wrapper.
That's the American export model in miniature. The world grabs the fun, ignores the instructions, and deals with the bill later.
So is Halloween worth it? As a stimulus package, no. As a once-a-year excuse for an entire country to be ridiculous together, it's probably the best $115 most people spend all year.
What's your house's Halloween budget, and could you defend it to an economist?
Apple CEO Says the Company Doesn't 'Want to Ship AI for AI's Sake'
In a recent interview with Wired's Steven Levy, Apple CEO John Ternus pushed back against criticism of the company's standing in the AI race.
Over the past two years, Apple has frequently been portrayed as lagging in artificial intelligence, particularly after repeated delays to a more capable version of Siri — a feature that only finally launched this summer. Ternus, a longtime Apple executive who officially took over as CEO at the start of September, acknowledged that "everyone has an opinion," but maintained, "I really like where we're at."
He said Apple is "absolutely going to leverage AI" in its upcoming products, but drew a line at adopting the technology simply to keep pace with rivals. "We don't want to ship AI for AI's sake," Ternus explained. "When we're doing it right, people won't have to think about whether it was AI that made this happen or more traditional software. What matters is, did we build something that they really love?"
When Levy pressed further into Apple's business more fundamentally — suggesting that intelligent agents might one day make apps obsolete — Ternus pushed back on the idea. While conceding such a shift "could make life really easy, I suppose," he argued that building products remains "a deeply human and personal thing, something that we love to do."
The abrupt decision to dissolve one-time legal powerhouse Weil, Gotshal and Manges has rattled Big Law — and social media. While insider play-by-plays and early autopsies start to make the rounds online, affected Weil litigators and alums on LinkedIn are sharing their memories and perspectives. In the meantime, rival firms are swooping in to grab top talent, and top law schools and their alumni networks are helping students and recent grads figure out next steps.
Burger King's attempts to revive the brand have included everything from modernizing restaurants to giving customers a phone number for President Tom Curtis. According to parent company Restaurant Brands International, it's working: U.S. same-store sales grew 8.4% in the third quarter, and Burger King recently restored its No. 2 ranking among burger chains. For all the sleek remodels and clever marketing, Bloomberg reporter Daniela Sirtori says Curtis' approach "also comes down to something much simpler: asking people, 'Hey, how's that burger?' and taking the answer seriously."
Big Tech concluded a bruising week as disappointing earnings and mounting concerns over AI spending fueled a market rout. The Nasdaq tumbled 2.9% Friday, with Intel plunging 29% and Amazon slipping 1.4% in the latest post-earnings selloffs. Those declines followed a $500 billion wipeout in Alphabet's market value amid fears that AI could undermine its search business. The selloff exposes a dividing line in the AI boom: Financial staying power. On LinkedIn, BlackRock’s Wei Li notes many tech firms are "spending more than they are making," pushing free cash flow into negative territory.
AI Investors Are Racing to Insure the Biggest Market Bet in History
With trillions riding on the artificial-intelligence trade, investors, traders and tech executives alike are hunting for high-stakes protection
Wall Street has placed a multitrillion-dollar bet on artificial intelligence. Now, some investors are desperate to take the other side.
The AI trade has been a lasting winner so far. Excitement about AI's potential to transform the economy has lifted stocks from tech heavyweights to utilities and chip makers, while driving a speculative frenzy in other corners of the market.
But some of those optimistic forecasts recently got a reality check, with big and small players alike failing to meet the market's expectations. Now, investors are sizing up how to protect themselves against the risk of a larger unraveling.
"We're seeing really significant demand from folks looking to protect downside," said Stephen Solaka, a senior vice president at Gateway Investment Advisers.
Jitters Are Starting to Show
The nervousness is appearing in several ways:
- **Lending costs**: The yields lenders demand to finance big data-center projects and chip purchases
- **Neocloud troubles**: Plunging stock and bond prices at heavily indebted AI "neocloud" companies like CoreWeave
- **Tech selloffs**: Steep drops from darlings including Alphabet and Intel, which just recorded some of their biggest one-day declines ever
- *Market-wide pain*: The tech-heavy Nasdaq lost 2.9% Friday, capping its worst month since March 2025
Solaka oversees portfolios for individuals who often have concentrated bets in individual stocks, including the tech giants that have powered markets in recent years. Increasingly, he said, his clients—some of whom are former employees of those very technology companies—are looking for something else: insurance.
"These people have made so much money, they're quitting to hedge their stock," Solaka said, adding that some companies have restrictions around employees' trades.
How Investors Are Hedging
Investors and tech executives alike are turning to options and credit-default swaps for protection.
Put options—which give investors the right to sell stock at a specific price at a future date—are growing in volume. Contracts tied to the neocloud CoreWeave surged to the highest level on record this month and have stayed elevated, Cboe Global Markets data show.
Some of the most popular bets tied to CoreWeave have been put options expiring in 2028, paying out if the shares plunged to $35. The shares hovered at around $55 this week.
Meanwhile, the cost to protect against a debt default in hyperscalers Amazon, Alphabet, Microsoft, Meta and Oracle widened over the past month to the highest levels going back to 2021, according to bond data provider Solve.
"There were no exceptions, suggesting broad market concern about the sector rather than idiosyncratic credit events at individual companies," said Eugene Grinberg, a co-founder of Solve.
The recalibration in Oracle credit-default spreads was especially stark, widening by more than 1 percentage point.
Hedging the AI Build-Out Itself
The AI build-out is expected to grow so large that traders are cooking up other ways to hedge its risks.
Quant trading firm Jump Trading recently launched a trading desk for computing power. The nascent operation has been taking advantage of wide price differences across providers—the types of inefficiencies that have been tougher to come by in mainstream markets such as stocks and options. It has also been structuring over-the-counter derivatives, including for those who want to hedge against a big collapse in prices going forward.
"Folks are becoming more discerning about which participants are winners and losers, and asking tougher questions," said Dave Olsen, president of Jump. "People are drilling deeper into the economic models."
Here's How a Trade Would Work
A neocloud or AI infrastructure developer that rents out Nvidia chips might buy a put option to hedge against the possibility that the hourly rental rates they pocket will plunge if demand drops.
It would have the right to sell that compute power at a specific price by a stated date, buffering itself from a steeper hit to profits. Jump would agree to take the other side. Jump could also negotiate a cash-settled contract, simply paying the provider in cash. Or the trading firm could agree to rent the compute, or find someone else to rent it to for a profit.
The cost to rent Nvidia B300 chips has surged around 56% this year to $7.56 an hour, according to a Silicon Data index. Jump, for example, could agree to buy if prices fell to $4 an hour.
Olsen and his traders are closely scrutinizing the creditworthiness of those they cut deals with, and the likelihood that they might default on their obligations.
A Growing Derivatives Web
Big exchanges and trading firms are also looking to launch a web of derivatives tied to compute, as it is known in the industry.
Susquehanna International Group, one of the largest trading firms on Wall Street, has fielded interest—both directly and through intermediaries such as brokers—from data centers and privately held neoclouds seeking to hedge their exposure to semiconductor chips, said Nick Volbi, head of institutional prediction markets at the trading firm. If compute prices sank because of decreased demand or increased supply, the trades would pay out, hedging an option buyer against losses in other parts of its business.
That includes put option trades tied to a compute index, which can serve as insurance to assuage lenders at banks, he said.
"I'm going to go to a potential lender and say, 'Hey, if I buy this structure and I get this form of insurance, can you now give me financing? And what would that financing look like?'" Volbi said.
There has been some thin trading in prediction markets tied to data-center moratoriums in certain states. Kalshi offers trades on which states might halt new data-center developments before 2028, including New Hampshire, Michigan and Maine. So far, though, trades have been sparse.
Olsen said hedging strategies as a whole are still dwarfed by the AI trade. Many remain bullish.
"I don't think my house is going to catch on fire, but I still have insurance," Olsen said. "You don't need to be bearish to want some insurance."
It was a fractious week for new media, with Instagram's re-embrace of long-form video spurring big debates around monetization for creators, while the booming microdrama market could be seeing its first signs of growing pains. Jay-Z selected himself as the halftime performer for next February's Super Bowl. Skydance gave its on-edge workforce yet another thing to worry about. Idris Elba unveiled a free AI training program for filmmakers in Africa. Anthropic got into legal trouble with Elvis. And the world bid farewell to Sir Patrick Stewart.
Binance says it has recovered most of the assets stolen this week in the largest-ever cryptocurrency hack, which the FBI has linked to North Korea. In a LinkedIn post, co-CEO Richard Teng said the company replenished 90% of the $6.3 billion in stolen cryptocurrency, and will cover the remaining 10%: "Our users will not bear the financial impact of this incident," Teng said. Binance suspended all withdrawals Friday, and Teng says the company is working to allow customers to retrieve funds by Sunday.
$6.3 BILLION stolen from Binance. The most important question isn’t just how hackers got in. It’s how they were able to move the money.
The recent Binance cyberattack should be getting the attention of every financial institution, technology provider, and security executive.
According to Binance’s preliminary investigation, attackers may have initially compromised an engineer’s laptop, gained access to internal infrastructure, and initiated unauthorized withdrawals.
Further findings suggest that attackers exploited Ethereum’s account-delegation functionality to bypass normal approval procedures.
The FBI has attributed the attack to North Korea’s TraderTraitor hacking operation.
Binance has committed to covering the losses and reports that 90% of the affected assets have already been replenished.
But there’s a much bigger conversation we need to have.
What happens when the systems we trust to protect our money are themselves compromised?
Organizations invest enormous resources into preventing unauthorized access.
And they should.
But preventing someone from getting into a system is only half the challenge.
The other half is making sure that even when access is compromised, unauthorized actions cannot proceed unchecked.
That means having independent safeguards that can:
• Evaluate transactions before execution.
• Enforce financial limits that cannot simply be bypassed.
• Detect unusual behavior and suspend suspicious activity.
• Require additional authorization for high-risk actions.
• Maintain independently verifiable records of what happened.
This is especially important as financial institutions increasingly introduce autonomous AI agents capable of making decisions and initiating transactions.
At NextGen Global Trading Labs (NGTL), these are the types of challenges we’re addressing through independent behavioral governance and transaction enforcement.
To be clear, this was not reported as an AI-agent attack, and the full technical investigation is still ongoing.
But the underlying lesson extends far beyond cryptocurrency.
Trust should never depend entirely on the same system that has the authority to execute an action.
The next generation of financial infrastructure needs more than security around its systems.
It needs independent accountability at the moment decisions become actions.
Here’s a question for financial and technology leaders:
If an attacker compromises an authorized system in your organization, what independently prevents that system from executing a transaction it should never have been allowed to make?
Meta Built Humanoid Prototype With Chinese Startup To Ramp Up Robotics Push
Starbucks explores personalized digital menus at the drive-thru
Starbucks is exploring ways to recognize participating customers on arrival and make payment easier.
Starbucks is still exploring personalized digital menus at its drive-thrus, and their timing remains unclear. CEO Brian Niccol described recognizing participating customers on arrival and making payment easier. Separately, the chain plans to begin testing background AI models in 2027 to help baristas enter orders more accurately and efficiently. It also plans to add AI to app search and discovery; conversational ordering remains a possibility.
FBI attributes $6 billion-plus Binance theft to North Korea
Binance has set aside company funds to cover losses and promised customers full reimbursement.
The FBI linked the theft of more than $6 billion from Binance to North Korean actors it calls TraderTraitor. It urged crypto firms to block transactions tied to laundering addresses and reported that stolen assets were being converted primarily into Ethereum. Binance says it has funds to cover 90% of the missing assets and has pledged full reimbursement. That funding is separate from recovering stolen crypto.
Startup says AI agent injected therapy into mouse's brain
An AI agent chose a gene therapy and steered the needle that delivered it into the brain of a living, epileptic mouse, according to Exin Therapeutics CEO Gabriel Ocana Santero. Anthropic's most advanced model would not do the injection, though. Claude Opus 5.5 helped plan the experiment but refused to guide the needle, so the team switched to Claude Sonnet 5, WIRED reported on October 31.
The San Francisco startup uses AI to analyze brain activity and develop drugs for epilepsy and neurodegenerative diseases. It built the experiment to test what happens when agents get direct control of lab equipment instead of only designing studies. "We've realized that we need to connect these agents to a lab to get the maximum value from them," Santero told WIRED.
How the procedure worked
The agent got a spreadsheet of 19 candidate gene therapies, a motorized surgery rig with cameras, and one goal: reduce seizures. It chose the therapy, the brain region and the injection coordinates, and it set the volume and speed. According to The Decoder, it picked a therapy that strengthens an inhibitory signaling pathway. It worked out the coordinates by averaging values from three papers, and it logged 32 observations through the camera while watching for bleeding. Santero said on X that the agent injected 250 nanoliters.
People still handled key steps. A veterinary surgeon did the craniotomy to open the skull, humans had to approve the injection plan, and an emergency button could cut power to the rig's motors.
Refusals and a glitch
At the moment of injection, Opus 5.5 said "something on my output is telling me I shouldn't do this," Santero recalled. He called that "very interesting". The Decoder reported that Opus 5 also refused, warning that stopping halfway with a needle in an anesthetized animal's brain would be worse than not starting at all.
During the procedure, a glitch threw off the rig's coordinate system. The Sonnet agent noticed that its target position was wrong but would not change the equipment code, because its instructions banned that. WIRED reported that the team used a separate Opus 5 agent to fix the code without telling it a live surgery was underway, which stretched a roughly 45-minute procedure to just over an hour. The Decoder instead said Santero fixed the code by hand. Anthropic did not respond to WIRED's request for comment.
Results still pending
The mouse is recovering well, but the team has not yet measured whether the therapy reduces its seizures. The results have not been peer-reviewed, and a technical report is expected. Exin plans to test Grok and other models next. It is also exploring agent-controlled craniotomies but wants to be sure they are safe before trying them in live animals.
"I am not aware of an agent previously choosing a treatment and a brain target and then delivering it into a living mammal," Santero said. The experiment has also raised ethical concerns among some observers.