Business
Column: These Apple researchers just showed that AI bots can't think, and possibly never will
See if you can solve this arithmetic problem:
Oliver picks 44 kiwis on Friday. Then he picks 58 kiwis on Saturday. On Sunday, he picks double the number of kiwis he did on Friday, but five of them were a bit smaller than average. How many kiwis does Oliver have?
If you answered “190,” congratulations: You did as well as the average grade school kid by getting it right. (Friday’s 44 plus Saturday’s 58 plus Sunday’s 44 multiplied by 2, or 88, equals 190.)
You also did better than more than 20 state-of-the-art artificial intelligence models tested by an AI research team at Apple. The AI bots, they found, consistently got it wrong.
The fact that Apple did this has gotten a lot of attention, but nobody should be surprised at the results.
— AI critic Gary Marcus
The Apple team found “catastrophic performance drops” by those models when they tried to parse simple mathematical problems written in essay form. In this example, the systems tasked with the question often didn’t understand that the size of the kiwis have nothing to do with the number of kiwis Oliver has. Some, consequently, subtracted the five undersized kiwis from the total and answered “185.”
Human schoolchildren, the researchers posited, are much better at detecting the difference between relevant information and inconsequential curveballs.
The Apple findings were published earlier this month in a technical paper that has attracted widespread attention in AI labs and the lay press, not only because the results are well-documented, but also because the researchers work for the nation’s leading high-tech consumer company — and one that has just rolled out a suite of purported AI features for iPhone users.
“The fact that Apple did this has gotten a lot of attention, but nobody should be surprised at the results,” says Gary Marcus, a critic of how AI systems have been marketed as reliably, well, “intelligent.”
Indeed, Apple’s conclusion matches earlier studies that have found that large language models, or LLMs, don’t actually “think” so much as match language patterns in materials they’ve been fed as part of their “training.” When it comes to abstract reasoning — “a key aspect of human intelligence,” in the words of Melanie Mitchell, an expert in cognition and intelligence at the Santa Fe Institute — the models fall short.
“Even very young children are adept at learning abstract rules from just a few examples,” Mitchell and colleagues wrote last year after subjecting GPT bots to a series of analogy puzzles. Their conclusion was that “a large gap in basic abstract reasoning still remains between humans and state-of-the-art AI systems.”
That’s important because LLMs such as GPT underlie the AI products that have captured the public’s attention. But the LLMs tested by the Apple team were consistently misled by the language patterns they were trained on.
The Apple researchers set out to answer the question, “Do these models truly understand mathematical concepts?” as one of the lead authors, Mehrdad Farajtabar, put it in a thread on X. Their answer is no. They also pondered whether the shortcomings they identified can be easily fixed, and their answer is also no: “Can scaling data, models, or compute fundamentally solve this?” Farajtabar asked in his thread. “We don’t think so!”
The Apple research, along with other findings about the limitations of AI bots’ cogitative limitations, is a much-needed corrective to the sales pitches coming from companies hawking their AI models and systems, including OpenAI and Google’s DeepMind lab.
The promoters generally depict their products as dependable and their output as trustworthy. In fact, their output is consistently suspect, posing a clear danger when they’re used in contexts where the need for rigorous accuracy is absolute, say in healthcare applications.
That’s not always the case. “There are some problems which you can make a bunch of money on without having a perfect solution,” Marcus told me. Recommendation engines powered by AI — those that steer buyers on Amazon to products they might also like, for example. If those systems get a recommendation wrong, it’s no big deal; a customer might spend a few dollars on a book he or she didn’t like.
“But a calculator that’s right only 85% of the time is garbage,” Marcus says. “You wouldn’t use it.”
The potential for damagingly inaccurate outputs is heightened by AI bots’ natural language capabilities, with which they offer even absurdly inaccurate answers with convincingly cocksure elan. Often they double down on their errors when challenged.
These errors are typically described by AI researchers as “hallucinations.” The term may make the mistakes seem almost innocuous, but in some applications, even a minuscule error rate can have severe ramifications.
That’s what academic researchers concluded in a recently published analysis of Whisper, an AI-powered speech-to-text tool developed by OpenAI, which can be used to transcribe medical discussions or jailhouse conversations monitored by correction officials.
The researchers found that about 1.4% of Whisper-transcribed audio segments in their sample contained hallucinations, including the addition to transcribed conversation of wholly fabricated statements including portrayals of “physical violence or death … [or] sexual innuendo,” and demographic stereotyping.
That may sound like a minor flaw, but the researchers observed that the errors could be incorporated in official records such as transcriptions of court testimony or prison phone calls — which could lead to official decisions based on “phrases or claims that a defendant never said.”
Updates to Whisper in late 2023 improved its performance, the researchers said, but the updated Whisper “still regularly and reproducibly hallucinated.”
That hasn’t deterred AI promoters from unwarranted boasting about their products. In an Oct. 29 tweet, Elon Musk invited followers to submit “x-ray, PET, MRI or other medical images to Grok [the AI application for his X social media platform] for analysis.” Grok, he wrote, “is already quite accurate and will become extremely good.”
It should go without saying that, even if Musk is telling the truth (not an absolutely certain conclusion), any system used by healthcare providers to analyze medical images needs to be a lot better than “extremely good,” however one might define that standard.
That brings us to the Apple study. It’s proper to note that the researchers aren’t critics of AI as such but believers that its limitations need to be understood. Farajtabar was formerly a senior research scientist at DeepMind, where another author interned under him; other co-authors hold advanced degrees and professional experience in computer science and machine learning.
The team plied their subject AI models with questions drawn from a popular collection of more than 8,000 grade school arithmetic problems testing schoolchildren’s understanding of addition, subtraction, multiplication and division. When the problems incorporated clauses that might seem relevant but weren’t, the models’ performance plummeted.
That was true of all the models, including versions of the GPT bots developed by OpenAI, Meta’s Llama, Microsoft’s Phi-3, Google’s Gemma and several models developed by the French lab Mistral AI.
Some did better than others, but all showed a decline in performance as the problems became more complex. One problem involved a basket of school supplies including erasers, notebooks and writing paper. That requires a solver to multiply the number of each item by its price and add them together to determine how much the entire basket costs.
When the bots were also told that “due to inflation, prices were 10% cheaper last year,” the bots reduced the cost by 10%. That produces a wrong answer, since the question asked what the basket would cost now, not last year.
Why did this happen? The answer is that LLMs are developed, or trained, by feeding them huge quantities of written material scraped from published works or the internet — not by trying to teach them mathematical principles. LLMs function by gleaning patterns in the data and trying to match a pattern to the question at hand.
But they become “overfitted to their training data,” Farajtabar explained via X. “They memorized what is out there on the web and do pattern matching and answer according to the examples they have seen. It’s still a [weak] type of reasoning but according to other definitions it’s not a genuine reasoning capability.” (the brackets are his.)
That’s likely to impose boundaries on what AI can be used for. In mission-critical applications, humans will almost always have to be “in the loop,” as AI developers say—vetting answers for obvious or dangerous inaccuracies or providing guidance to keep the bots from misinterpreting their data, misstating what they know, or filling gaps in their knowledge with fabrications.
To some extent, that’s comforting, for it means that AI systems can’t accomplish much without having human partners at hand. But it also means that we humans need to be aware the tendency of AI promoters to overstate their products’ capabilities and conceal their limitations. The issue is not so much what AI can do, but how users can be gulled into thinking what it can do.
“These systems are always going to make mistakes because hallucinations are inherent,” Marcus says. “The ways in which they approach reasoning are an approximation and not the real thing. And none of this is going away until we have some new technology.”
Residents of Monterey Park voted overwhelmingly to ban data centers on election day, making the San Gabriel Valley city the first in the nation to do so by public vote.
As of Wednesday, 86% of votes were in favor of Measure NDC, the city ban, according to the Los Angeles County registrar-recorder/county clerk.
Other cities and towns have passed moratoriums on data centers, as a wave of opposition sweeps the country. But the Monterey Park vote can only be overturned by another ballot measure, making it the most permanent data center ban in a jurisdiction.
Monterey Park’s City Council had already banned data centers by ordinance, after a proposed 247,000-square-foot data center met an outpouring of public anger and concern. The developer withdrew that plan.
That facility would have been less than 500 feet away from the nearest home, and would have used three times the electricity of the entire 60,000-person city. Residents said it would have caused noise and air pollution and driven up electricity rates.
“This ensures long-lasting protections for current and future generations,” Amy Wong, co-founder of the group San Gabriel Valley Progressive Action, said of the vote. “It means that future city councils cannot overturn a data center ban, even if data center developers wanted to spend money to fund pro-data center candidates.”
The measure had no formal opposition. The developer of the proposed facility, investment firm HMC StratCap, said it wouldn’t engage in the ballot fight when it withdrew in March.
The Data Center Coalition, an industry trade group, expressed disappointment in the vote.
“It sends a signal that the area is closed for business, both for data centers and for other significant economic development projects,” state policy director Khara Boender said.
“It deprives local residents of the opportunity to compete for jobs and investment, while also causing the area to relinquish substantial long-term economic investment, high-wage jobs, and critical tax revenue to neighboring areas or other states.”
SGV Progressive Action worked with hyperlocal groups including No Data Center Monterey Park to rally support for the measure.
The group is now focused on stopping data center proposals in the City of Industry and fighting a move by City of Industry, Santa Fe Springs, Vernon and City of Commerce to welcome data centers and other industry with fast-tracked permitting and tax incentives.
City of Industry, in the San Gabriel Valley, and Vernon, south of downtown L.A., are primarily industrial areas, each with around 300 permanent residents. They are employment centers, and tens of thousands of workers commute in daily.
There has been little vocal opposition to data centers among the few residents of these cities. Wong said the protest is primarily coming from the surrounding neighborhoods.
“If a data center gets built in City of Industry, residents across the region would bear the brunt of pollution and increased utility costs,” Wong said, noting that it is surrounded by 16 other cities and unincorporated communities.
Data center proposals have been limited in California compared to Virginia, Texas, Georgia, Illinois and Arizona, which sit at the center of a recent boom in hyperscaler facilities to power artificial intelligence.
California has the third-most data centers in the country, with 300, but high electricity rates, expensive land and regulatory hurdles mean that fewer, and smaller, facilities are currently planned than in other hotspots.
That doesn’t mean opposition hasn’t been fierce. In Coachella and Imperial County, residents are showing up in droves to protest local proposals.
In the San Gabriel Valley, Montebello, El Monte and Baldwin Park have all enacted temporary moratoriums, and Alhambra recently banned data centers as part of a zoning code update.
Wong said she hoped the ballot measure vote would galvanize the opposition. “The vote is a testament to the people power of our region,” she said. “Our region is worth protecting, and we won’t let data centers determine our future.”
A rule intended to prevent rent gouging in the wake of the Eaton and Palisades fires has lapsed in Los Angeles County, possibly exposing some renters to hikes.
The executive order that blocked rent increases was issued by Gov. Gavin Newsom amid the devastating wildfires last year. Under the order, landlords couldn’t increase rents by more than 10% above their prefire levels.
The rule, which was supposed to be temporary and was repeatedly extended, ended Friday after a vote to extend it again failed to garner enough votes. Supervisor Lindsey Horvath, whose district includes Pacific Palisades, sounded the alarm in a motion to extend price protections that failed to pass at the Board of Supervisors’ May 19 meeting.
“These price gouging protections continue to be necessary as construction and rebuilding continue, and as thousands of people remain displaced,” the motion said. “Families which signed short-term leases could face drastic price increases of 50% or more without further price gouging protection.”
Los Angeles County is home to more than 1 million rental properties, though not all of them needed protection from the new rule. There are already stricter rent increase caps for many residences, depending on the location, type and age of the building. Despite the rent control in the region, the people of Los Angeles pay among the highest rents in the country.
It is uncertain whether renters will face rapidly rising rents now that the protection has lapsed. But some real estate experts and policymakers said there was no need for the temporary rule that was part of the governor’s state of emergency.
Supervisors Kathryn Barger, Janice Hahn and Holly Mitchell abstained from voting on the motion to extend the protection, while Supervisors Hilda Solis and Horvath supported it.
“I abstained because I did not see sufficient evidence to justify extending this emergency ordinance, nor did I see evidence to eliminate it entirely,” Hahn said.
Barger’s office said she supported allowing the protections to sunset while waiting to see whether new information emerged.
“Market data already shows countywide rents are only about 2% above pre-emergency levels and rental inventory has grown,” Barger representative Helen E. Chavez Garcia said. “The Supervisor is also mindful of the burden these ongoing protections place on small property owners throughout the county.”
Mitchell did not immediately respond to a request for comment.
There haven’t been steep rent hikes in neighborhoods within three miles of the Palisades fire, according to a Times analysis of data from Zillow, the property listing company.
In ZIP Codes within three miles of the Palisades fire, rent increased 4.8% from December 2024 to April 2025. In areas around the Eaton fire, which destroyed swaths of Altadena, rent jumped 5.2% in the same period.
In L.A. County, ZIP Codes farther from the fires saw only about a 2% increase.
A landlords representative, Jesus Rojas of the Apartment Owners Assn. of Greater Los Angeles, told the supervisors during public comment at the meeting that the county’s rent-gouging rules have “long outlived the emergency they were intended to address” and are now being “wrongfully used to harm thousands of rental housing providers throughout the county.”
“There is no proof that multifamily rental housing providers are hugely increasing rents for impacted homeowners,” Rojas said.
Indeed, there are strong signs that the property market in the Los Angeles area has at last begun to cool.
L.A. metro-area rent prices recently fell to a four-year low, with the median rent slipping to $2,167 in December.
Meanwhile, condominium sales had their slowest start of the year in decades. Condo sales in Los Angeles have plummeted to a 20-year low, with fewer than 2,000 units sold in January and February — the worst start to the year since 2005.
Newsom defended the price-gouging protections shortly after they went into effect.
“In the days following the Los Angeles firestorms, we worked quickly to protect Los Angeles survivors from any form of exploitation,” he said in February 2025. “The state has the tools in place to not only block price gouging during this emergency, but also to prosecute bad actors.”
The Los Angeles County Department of Consumer and Business Affairs said it received more than 2,000 complaints after the fires, alleging that retailers and landlords were taking advantage of people put in hardship by their losses, and sent out more than 2,000 cease-and-desist letters to businesses and landlords for alleged price gouging, said Morine Merritt, who oversees department investigations into consumer and real estate fraud.
“Close to 90% of the complaints that we received involved allegations of rent increases,” Merritt said in an interview. Now that the fire-related protections have expired, existing laws and “regular market conditions determine price increases for goods and services, including rents,” she said.
Crackdowns on fire-related rent gouging have been rare, said Chelsea Kirk of the activist organization the Rent Brigade, which analyzed L.A. County’s rental market in the year after the fires. It reported 18,360 potential examples of price gouging in listings but said that few lawsuits had been filed by authorities so far.
Last week, Rent Brigade announced what it said was the first private civil lawsuit brought by a family that claimed to be rent-gouged in the aftermath of the wildfires. Plaintiffs Randall and Candy Renick, whose Altadena home was damaged, said they were charged nearly three times the maximum permitted rate for nearly 10 months. They seek restitution of $96,000 plus civil penalties and attorneys’ fees.
The rental market has probably stabilized since the fires, Kirk said, but other families may still be “locked into illegal rents” that they agreed to pay when they were in a rush to find housing after they were displaced.
Dear Mr. Pelley:
I meant what I said in my letter last week to the 60 Minutes team: joining 60 Minutes is the honor of my career and I am grateful to be working alongside the people who have contributed to the most important television journalism brand this country has ever produced. While I’m new to 60 Minutes, I’ve devoted my career to investigative journalism and storytelling. I started this job excited to collaborate and to benefit from the wisdom and experience of the 60 Minutes veterans, with you among them. For that reason, one of the first things I did in my new role was call you to talk and invite you to dinner. It is a profound disappointment that you rejected that overture and chose ambush instead. Yesterday, you hijacked my first meeting with staff to disparage me, my qualifications, and my intentions with remarkable incivility and contempt. I welcome a diversity of viewpoints and respectful debate among the team, but this was nothing of the sort. Yesterday’s performative display of hostility enacted in front of the staff instead of in a civil, private conversation-demonstrated that you have no interest in contributing to the future success of the show, or approaching my new tenure with a mind open to collaboration and progress. I am here to deliver first-in-class news programming, not to make headlines about newsroom drama. I am eager to work alongside those who share this goal.
Despite yesterday’s misconduct, I had hoped that in sitting down with you today we could find a path forward together. You made clear that you are not interested in such a path.
Your antipathy to the future of the show has come through loud and clear. And I have heard you. I therefore write on behalf of CBS News, Inc. (“CBS”) to inform you that your employment with CBS is terminated for cause effective immediately. Enclosed is your formal termination letter.
Sincerely,
Nick Bilton
Executive Producer, 60 Minutes
-
San Francisco, CA6 minutes agoSan Francisco family devastated as they face nearly 90% rent increase
-
Dallas, TX11 minutes agoWings’ top pick Azzi Fudd hosts clinic as Cash App donates to Dallas nonprofit
-
Miami, FL18 minutes agoPatients left scrambling for care after Miami-Dade woman accused of operating an unlicensed surgery recovery center
-
Boston, MA21 minutes agoClover plans to reopen some locations after sudden closure, thanks to an anonymous investor
-
Denver, CO26 minutes agoNew report finds Denver metro home buyers and sellers experiencing ‘unattainability fatigue’
-
Seattle, WA33 minutes agoSeattle mayor grilled over public safety, affordability, CCTV
-
San Diego, CA36 minutes agoAutomated license plate readers and public surveillance cameras are coming to Imperial Beach
-
Milwaukee, WI41 minutes agoFriends, family gather to remember gunshot victim Pepe Sikisi-Belle Jr.
