Metas decision to end its professional fact-checking program sparked a wave of criticism in the tech and media world. Critics warned that dropping expert oversight could erode trust and reliability in the digital information landscape, especially when profit-driven platforms are mostly left to police themselves.
What much of this debate has overlooked, however, is that today, AI large language models are increasingly used to write up news summaries, headlines, and content that catch your attention long before traditional content moderation mechanisms can step in. The issue isnt clear-cut cases of misinformation or harmful subject matter going unflagged in the absence of content moderation. Whats missing from the discussion is how ostensibly accurate information is selected, framed, and emphasized in ways that can shape public perception.
Large language models gradually influence the way people form opinions by generating the information that chatbots and virtual assistants present to people over time. These models are now also being built into news sites, social media platforms, and search services, making them the primary gateway to obtain information.
Studies show that large language models do more than simply pass along information. Their responses can subtly highlight certain viewpoints while minimizing others, often without users realizing it.
Communication bias
My colleague, computer scientist Stefan Schmid, and I, a technology law and policy scholar, show in a forthcoming accepted paper in the journal Communications of the ACM that large language models exhibit communication bias. We found that they may have a tendency to highlight particular perspectives while omitting or diminishing others. Such bias can influence how users think or feel, regardless of whether the information presented is true or false.
Empirical research over the past few years has produced benchmark datasets that correlate model outputs with party positions before and during elections. They reveal variations in how current large language models deal with public content. Depending on the persona or context used in prompting large language models, current models subtly tilt toward particular positionseven when factual accuracy remains intact.
These shifts point to an emerging form of persona-based steerabilitya models tendency to align its tone and emphasis with the perceived expectations of the user. For instance, when a user describes themselves as an environmental activist and another as a business owner, a model may answer the same question about a new climate law by emphasizing different, yet factually accurate, concerns for each of them. For example, the criticisms could be that the law does not go far enough in promoting environmental benefits and that the law imposes regulatory burdens and compliance costs.
Such alignment can easily be misread as flattery. The phenomenon is called sycophancy: Models effectively tell users what they want to hear. But while sycophancy is a symptom of user-model interaction, communication bias runs deeper. It reflects disparities in who designs and builds these systems, what datasets they draw from, and which incentives drive their refinement. When a handful of developers dominate the large language model market and their systems consistently present some viewpoints more favorably than others, small differences in model behavior can scale into significant distortions in public communication.
Bias in large language models starts with the data theyre trained on.
What regulation can and cant do
Modern society increasingly relies on large language models as the primary interface between people and information. Governments worldwide have launched policies to address concerns over AI bias. For instance, the European Unions AI Act and the Digital Services Act attempt to impose transparency and accountability. But neither is designed to address the nuanced issue of communication bias in AI outputs.
Proponents of AI regulation often cite neutral AI as a goal, but true neutrality is often unattainable. AI systems reflect the biases embedded in their data, training, and design, and attempts to regulate such bias often end up trading one flavor of bias for another.
And communication bias is not just about accuracyit is about content generation and framing. Imagine asking an AI system a question about a contentious piece of legislation. The models answer is not only shaped by facts, but also by how those facts are presented, which sources are highlighted and the tone and viewpoint it adopts.
This means that the root of the bias problem is not merely in addressing biased training data or skewed outputs, but in the market structures that shape technology design in the first place. When only a few large language models have access to information, the risk of communication bias grows. Apart from regulation, then, effective bias mitigation requires safeguarding competition, user-driven accountability and regulatory openness to different ways of building and offering large language models.
Most regulations so far aim at banning harmful outputs after the technologys deployment, or forcing companies to run audits before launch. Our analysis shows that while prelaunch checks and post-deployment oversight may catch the most glaring errors, they may be less effective at addressing subtle communication bias that emerges through user interactions.
Beyond AI regulation
It is tempting to expect that regulation can eliminate all biases in AI systems. In some instances, these policies can be helpful, but they tend to fail to address a deeper issue: the incentives that determine the technologies that communicate information to the public.
Our findings clarify that a more lating solution lies in fostering competition, transparency, and meaningful user participation, enabling consumers to play an active role in how companies design, test, and deploy large language models.
The reason these policies are important is that, ultimately, AI will not only influence the information we seek and the daily news we read, but it will also play a crucial part in shaping the kind of society we envision for the future.
Adrian Kuenzler is a scholar-in-residence at the University of Denver and an associate professor at the University of Hong Kong.
This article is republished from The Conversation under a Creative Commons license. Read the original article.
The Federal Communications Commission on Monday said it would ban new foreign-made drones, a move that will keep new Chinese-made drones such as those from DJI and Autel out of the U.S. market.
The announcement came a year after Congress passed a defense bill that raised national security concerns about Chinese-made drones, which have become a dominant player in the U.S., widely used in farming, mapping, law enforcement,ss and filmmaking.
The bill called for stopping the two Chinese companies from selling new drones in the U.S. if a review found they posed a risk to American national security. The deadline for the review was Dec. 23.
The FCC said Monday the review found that all drones and critical components produced in foreign countries, not just by the two Chinese companies, posed unacceptable risks to the national security of the United States and to the safety and security of U.S. persons.” But it said specific drones or components would be exempt if the Pentagon or Department of Homeland Security determined they did not pose such risks.
The FCC cited upcoming major events, such as the 2026 World Cup, America250 celebrations, and the 2028 Summer Olympics in Los Angeles, as reasons to address potential drone threats posed by criminals, hostile foreign actors, and terrorists.”
Michael Robbins, president and chief executive officer of AUVSI, the Association for Uncrewed Vehicle Systems International, said in a statement that the industry group welcomes the decision. He said it’s time for the U.S. not only to reduce its dependence on China but build its own drones.
Recent history underscores why the United States must increase domestic drone production and secure its supply chains,” Robbins said, citing Beijing’s willingness to restrict critical supplies such as rare earth magnets to serve its strategic interests.
DJI said it was disappointed by the FCC decision. While DJI was not singled out, no information has been released regarding what information was used by the Executive Branch in reaching its determination, it said in a statement.
Concerns about DJIs data security have not been grounded in evidence and instead reflect protectionism, contrary to the principles of an open market, the company said.
In Texas, Gene Robinson has a fleet of nine DJI drones that he uses for law enforcement training and forensic analyses. He said the new restrictions would hurt him and many others who have come to rely on the Chinese drones because of their versatility, high performance, and affordable prices.
But he said he understands the decision and lamented that the U.S. had outsourced the manufacturing to China. Now, we are paying the price, Robinson said. To get back to where we had the independence, there will be some growing pains. We need to suck it up, and lets not have it happen again.”
Also in Texas, Arthur Erickson, chief executive officer and co-founder of the drone-making company Hylio, said the departure of DJI would provide much-needed room for American companies like his to grow. New investments are pouring in to help him ramp up production of spray drones, which farmers use to fertilize their fields, and it will bring down prices, Erickson said.
But he also called it crazy and unexpected that the FCC should expand the scope to all foreign-made drones and drone components. The way it’s written is a blanket statement, Erickson said. There’s a global allied supply chain. I hope they will clarify that.
Didi Tang, Associated Press
The governor of Niigata on Tuesday formally gave local consent to put two reactors at the Kashiwazaki-Kariwa nuclear power plant in the north-central prefecture back online, clearing a last hurdle toward restarting the plant idled for more than a decade following the 2011 meltdowns at another plant managed by the same utility.
Gov. Hideyo Hanazumi, in his meeting with Economy and Industry Minister Ryosei Akazawa, conveyed the prefecture’s “endorsement” to restart the No. 6 and No. 7 reactors at the Kashiwazaki-Kariwa plant, accepting the government’s pledge to ensure safety, emergency response and understanding of the residents.
Restart preparations for No. 6 reactor have moved ahead and utility company TEPCO is expected to apply for a final safety inspection by the Nuclear Safety Authority later this week ahead of a possible resumption in January. Work at the other reactor is expected to take a few more years.
The move comes one day after the Niigata prefectural assembly adopted a budget bill that included funding necessary for a restart, supporting the governor’s earlier consent.
“It was a heavy and difficult decision,” Hanazumi told reporters.
Hanazumi also met with Prime Minister Sanae Takaichi, who also supports nuclear energy, and asked her to visit to observe the safety at the plant.
Japan once planned to phase out atomic power following the disaster at the Fukushima plant caused by an earthquake and tsunami. But in the face of global fuel shortages, rising prices and pressure to reduce carbon emissions, the government has reversed its policy and is now seeking to increase nuclear energy use by accelerating reactor restarts, extending their operational lifespan and considering building new ones.
Of the 57 commercial reactors, 13 are currently in operation, 20 are offline and 24 others are being decommissioned, according to the nuclear authorities.
The Kashiwazaki-Kariwa plant, which comprises seven reactors, is the world’s biggest. The plant has been offline since 2012 as part of nationwide reactor shutdowns in response to the March 2011 triple meltdowns at TEPCO’s Fukushima Daiichi plant.
Reactors No. 6 and 7 at Kashiwazaki-Kariwa had cleared safety tests in 2017, but their restart preparations were suspended after a series of safeguarding problems were found in 2021. The Nuclear Regulation Authority lifted an operational ban at the plant in 2023.
Its resumption again faced uncertainty following the Jan. 1, 2024, earthquake in the nearby Noto region that rekindled safety concerns among local residents about the plant and evacuation in case of a major disaster. The industry ministry sought an early resumption approval from Niigata two months later.
In Japan, a reactor restart is subject to the local community’s consent.
TEPCO, heavily burdened with the growing cost of decades-long decommissioning and compensation for residents affected by the Fukushima disaster, has been anxious to resume its only workable nuclear plant to improve its business. TEPCO has been struggling to regain public trust in safely running a nuclear power plant.
Aside from plant safety, experts say acceleration of reactor restarts also raises concern in a country without full nuclear fuel reprocessing or plans for radioactive waste management.
Mari Yamaguchi, Associated Press
The U.S. economy grew at a surprisingly strong 4.3% annual rate in the third quarter, the most rapid expansion in two years, as government and consumer spending, as well as exports, all increased.U.S. gross domestic product from July through September the economy’s total output of goods and services rose from its 3.8% growth rate in the April-June quarter, the Commerce Department said Tuesday in a report delayed by the government shutdown. Analysts surveyed by the data firm FactSet forecast growth of 3% in the period.However, inflation remains higher than the Federal Reserve would like. The Fed’s favored inflation gauge called the personal consumption expenditures index, or PCE climbed to a 2.8% annual pace last quarter, up from 2.1% in the second quarter.Excluding volatile food and energy prices, so-called core PCE inflation was 2.9%, up from 2.6% in the April-June quarter.Economists say that persistent and potentially worsening inflation could make a January interest rate cut from the Fed less likely, even as central bank official remain concerned about a slowing labor market.“If the economy keeps producing at this level, then there isn’t as much need to worry about a slowing economy,” said Chris Zaccarelli, chief investment officer for Northlight Asset Management, adding that inflation could return as the greatest concern about the economy.In a slow holiday trading week, U.S. markets on Wall Street turned lower following the GDP report, likely due to growing doubts that another Fed rate cut is coming next month.Consumer spending, which accounts for about 70% of U.S. economic activity, rose to a 3.5% annual pace last quarter, up from 2.5% in the April-June period.Consumption and investment by the government grew by 2.2% in the quarter after contracting 0.1% in the second quarter. The third quarter figure was boosted by increased expenditures at the state and local levels and federal government defense spending.Private business investment fell 0.3%, led by declines in investment in housing and in nonresidential buildings such as offices and warehouses. However, that decline was much less than the 13.8% slide in the second quarter.Within the GDP data, a category that measures the economy’s underlying strength grew at a 3% annual rate from July through September, up slightly from 2.9% in the second quarter. This category includes consumer spending and private investment, but excludes volatile items like exports, inventories and government spending.Exports grew at an 8.8% rate, while imports, which subtract from GDP, fell another 4.7%.Tuesday’s report is the first of three estimates the government will make of GDP growth for the third quarter of the year.Outside of the first quarter, when the economy shrank for the first time in three years as companies rushed to import goods ahead of President Donald Trump’s tariff rollout, the U.S. economy has continued to expand at a healthy rate. That’s despite much higher borrowing rates the Fed imposed in 2022 and 2023 in its drive to curb the inflation that surged as the United States bounced back with unexpected strength from the brief but devastating COVID-19 recession of 2020.Though inflation remains above the Fed’s 2% target, the central bank cut its benchmark lending rate three times in a row to close out 2025, mostly out of concern for a job market that has steadily lost momentum since spring.Last week, the government reported that the U.S. economy gained a healthy 64,000 jobs in November but lost 105,000 in October. Notably, the unemployment rate rose to 4.6% last month, the highest since 2021.The country’s labor market has been stuck in a “low hire, low fire” state, economists say, as businesses stand pat due to uncertainty over Trump’s tariffs and the lingering effects of elevated interest rates. Since March, job creation has fallen to an average 35,000 a month, compared to 71,000 in the year ended in March. Fed Chair Jerome Powell has said that he suspects those numbers will be revised even lower.
Matt Ott, AP Business Writer
Santa keeps delivering for quantum computing investors this year.
On Monday, shares of well-known quantum computing firms shot up by double digits, with D-Wave Quantum stock up almost 15% and Quantum Computing Inc. up 11%. Shares of IonQ Inc. and Rigetti Computing were likewise up roughly 10%.
The exact catalyst spurring those increases is unclear.
It may have initially been sparked in part by D-Waves Monday announcement that it would be attending the CES 2026 trade show next month.
The Palo Alto-based company plans to showcase its award-winning annealing quantum computing technology, hybrid quantum-classical solvers, and real-world customer use cases that are demonstrating measurable performance benefits, often beyond classical computing alone.
Quantum computing stocks have seen strong growth in 2025
Aside from that announcement, there may simply be ongoing excitement about the quantum space in general.
Publicly traded quantum computing firms have captivated investors over the past year or more, despite the speculative nature of the underlying technology that some say will transform the computer industry.
A June report published by McKinsey & Company dug into the appeal, saying that surging investment and faster-than-expected innovation could propel the quantum market to $100 billion in a decade.
It added that as quantum computing startups have received more funding from both public and private sources, the technology itself has started seeing more commercial deployment, and companies are also making progress in patenting the technology theyre developing.
Year-to-date growth for these stocks has been mostly impressive and in some cases eye-popping.
As of Tuesday morning, D-Wave shares are up 235% since January 1. IonQ shares are up 25%, and Rigetti shares are up 34%.
The outlier is Quantum Computing Inc., which has seen its stock price fall 35% year-to-date.
Will the end-of-year quantum rally last?
It’s unclear how long the holiday rally is going to last, but some profit-taking already seems to be underway.
As of early trading on Tuesday morning, D-Wave shares had fallen roughly 3%, while Rigetti was down around 1.58%. Shares in Quantum Computing Inc. IonQ were roughly flat.
We used to argue whether design was about aesthetics or about functionality. But in 2025, those conversations seemed downright quaint. Simpler debates for a simpler time.
Now were wondering if craft can survive the age of AI, and if well ever escape the politicization of every brand and object again.
For the December episode of our podcast By Design, I discussed these trends and more with Fast Company senior editor Liz Stinson.
We were joined by some of our brightest friends in the industry who shared their biggest own moments in design for the year, including Paola Antonelli (senior curator at MoMA), Cliff Kuang (FC Designs first editor and senior staff designer at Google), Forest Young (Global Design & AI Resident at Wolff Olins), and Elizabeth Goodspeed (editor-at-large at Its Nice That). Just try to guess who called out vibe coding, and who highlighted Sabrina Carpenters latest tour.
Tune in through Apple or Spotify, and please give us a few stars if you like it.
See you in 2026!
Across the country, a growing sentiment suggests the university degree is an artifact of a bygone era, a depreciating asset in an economy obsessed with speed. A recent Gallup poll confirms this shift, revealing that Americans confidence in the value of a college education has plummeted to a 15-year low.
Nowhere is this skepticism louder than in my own backyard. In Silicon Valley, the “skip college” mantra has evolved from a “hot take” to accepted wisdom. Fueled by the rise of generative AI, the logic is seductive: If artificial intelligence can code, write copy, and analyze data faster than a junior employee, why spend four years and a small fortune on skills a bot will master before you graduate?
It is a compelling argument. It is also fundamentally wrong.
As the CEO of an AI company, I witness the trajectory of automation daily. I see exactly what our models can do, and I recognize the massive disruption coming for knowledge work. Yet, my conclusion is the exact opposite of the current narrative.
As AI automates technical execution, the core purpose of the university sharpens. Far from making college obsolete, the AI revolution is making the benefits of higher education like wisdom, maturity, and the forging of mental models, the most critical economic differentiator a human can possess.
THE COMMODITY OF “HOW,” THE VALUE OF “WHY”
For the last two decades, higher education has been sold largely as vocational training. You go to school to learn a hard skill like computer science, accounting, or law, that you then trade for a salary. Under this transactional model, the skeptics are right. If college is just a place to download technical syntax into your brain, it is inefficient. AI is rapidly demonetizing the ability to simply do things.
However, the universitys true value was never entirely the how but it was always the why.
In an AI-native world, the technical barrier to entry is collapsing. Soon, natural language will be the only programming language required. When anyone can build an app, draft a legal brief, or design a product with a few prompts, execution becomes a commodity. The premium shifts to the ability to discern what to build, why it matters, and how it impacts the human ecosystem.
This requires a type of thinking that is rarely self-taught. It requires the kind of broad, interdisciplinary exposure that a university curriculum provides. We don’t need more people who can optimize a sorting algorithm; we need people who can debate the ethics of that algorithm, understand the sociological impact of its deployment, and navigate the geopolitical landscape it operates within.
COLLEGE AS SCAFFOLDING FOR THE MIND
Beyond the curriculum, the “skip college” contingent ignores the universitys profound developmental role. They view the four-year degree as a delay of adulthood. I view it as the necessary scaffolding for it.
The years between 18 and 22 are a neurological and psychological crucible. The brain is finalizing its development; identities are solidifying. The university environment provides a unique sandbox where young adults can collide with diverse philosophies, navigate complex social hierarchies, and fail in a relatively low-stakes environment.
When I hire for leadership roles, I rarely seek the fastest coder in the room. I seek resilience. I seek the ability to collaborate with dissenting voices and the maturity to navigate ambiguity. These are traits honed in lecture halls, seminar debates, and student organizations just as much as they are in internships.
THE SHELF-LIFE OF SKILLS VERSUS MINDSET
Critics often weigh the cost of tuition against the starting salary of a graduate’s first job. But in a world of accelerating technological velocity, the specific skills learned at 20 are often obsolete by 25. To skip college for a specific trade or tech stack is to bet one’s career on a snapshot in time.
A university education, particularly one grounded in the liberal arts and fundamental sciences, plays a longer game. It teaches you how to learn. It builds a mental operating system capable of updating itself.
Consider the “hallucination” problem in large language models. To effectively use these tools, a human must possess critical thinking skills robust enough to audit the machine. They need a foundational knowledge of history, logic, and science to discern when the AI is fabricating reality.
The worker who skips college risks becoming a passive consumer of AI output while the college graduate becomes its orchestrator. That is a difference in career trajectory that may not appear in year-one earnings, but compounds exponentially over a lifetime.
A CALL FOR A HUMAN RENAISSANCE
Silicon Valley loves efficiency. We love to optimize. And yes, the modern university is often inefficient, expensive, and bureaucratic. It is ripe for disruption and reform.
But lets not confuse the need for reform with the need for abolition. The “skip college” narrative is an oversimplification. It assumes that because machines are becoming more intelligent, humans can afford to be less educated.
The opposite is true. As we hand over more cognitive labor to AI, we free humans to operate at the peak of their intelligence. We are entering an era where philosophy, ethics, creative synthesis, and interpersonal leadership will be the most high-value skills in the global economy.
We should not encourage the next generation to skip the one institution dedicated to developing those traits. We should encourage them to go, but with a new purpose. Do not go to college just to get a job. Go to college to build the kind of complex, adaptable, and nuanced mind that no AI can replicate.
The future isn’t about competing with machines. It is about becoming more human. That is an education worth the investment.
Bhavin Shah is CEO and cofounder of Moveworks.
The U.S. Food and Drug Administration approved Novo Nordisk’s weight-loss pill on Monday, giving the Danish drugmaker a leg up in the race to market a potent oral medication for shedding pounds as it looks to regain lost ground from rival Eli Lilly.
The pill is 25 milligrams of semaglutide, the same active ingredient in injectable Wegovy and Ozempic, and will be sold under the brand name Wegovy. Novo already sells an oral semaglutide for type 2 diabetes, Rybelsus.
The approval could help spur a turnaround for Novo after a rocky year of sliding shares, profit warnings and slowing sales of its injectable Wegovy amid intense competition from Lilly and pressure from compounded versions.
U.S.-listed shares of Novo jumped 8% and Lilly fell 1% in extended trading after the approval announcement.
A 64-week, late-stage study showed participants who took 25 mg of oral semaglutide once daily lost an average of 16.6% of their body weight, compared with 2.7% for those on a placebo.
The pill was approved for chronic weight management in adults with obesity or overweight and at least one related health condition, broadening the potential patient pool at a time when insurers, employers and governments are wrestling with spiraling healthcare costs related to obesity.
It could help open the door to tens of millions of untapped patients in a global market, forecast to be worth some $150 billion a year by next decade.
“You’re going to see a huge uptake in the patient base as new indications open up and as oral versions hit the market,” said Anand Iyer, chief AI officer at telehealth firm Welldoc.
Novo is banking on the pill’s first-to-market advantage to revitalize sales in the U.S., where it has lost ground to Lilly. Lilly’s next-generation weight-loss pill orforglipron could be approved as soon as late March.
David Moore, Novo’s executive vice president of U.S. operations, said a daily pill could boost interest and uptake of the drug. Novo is manufacturing the pill in the United States in North Carolina and has been building up supplies of the pill “for some time” to ensure that it has “ample supply”, he said.
Some 40% of American adults are obese, U.S. government data shows, and around 12% say they currently take GLP-1 drugs, according to a poll published last month by health policy research organization KFF.
Novo had a first-to-market advantage with injectables, but initially struggled to meet explosive demand. Eventually, Lilly got ahead with its rival Zepbound, which now leads for weekly U.S. prescriptions.
Novo and analysts say a weight-loss pill would address injection hesitancy and expand access.
Analysts say pills could capture around a one-fifth share of the market by 2030, particularly among patients who prefer simpler and less invasive treatment options.
“The pills will not displace or replace the injections,” said Christopher Chrisman, a managing director and partner at consultancy BCG, adding some patients may prefer to continue with weekly injections.
“But pills offer clear advantages to some people. There’s travel convenience and no need for a fridge,” he added.
PRICING AGREEMENTS
Novo said the 1.5-milligram starting dose of the Wegovy pill will be available in early January.
Novo and Lilly had agreed to offer starter doses of their weightloss pills at $149 per month for the U.S. government Medicare and Medicaid health insurance programs and to cash-paying customers via the White House’s direct-to-consumer TrumpRx site.
Novo recently cut the cash price for Wegovy to $349 a month, from $499. U.S. list prices are about $1,000 per month or more.
Novo CEO Mike Doustdar said in November that people using weight-loss drugs show more “consumer-like” behavior than its traditional diabetes patients, acknowledging that the company needs to adapt to this and bring in new expertise.
Whether another semaglutide product can solve Novo’s current ills remains to be seen.
Novo’s oral semaglutide needs to be taken in the morning on an empty stomach, 30 minutes before eating, drinking or using any other oral medication. Lilly’s pill does not have those restrictions.
(Reporting by Maggie Fick, Patrick Wingrove, Mariam Sunny, Christy Santhosh and Mrinalika Roy; Editing by Adam Jourdan, Bill Berkrot, Rosalba O’Brien and Jamie Freed)
Maggie Fick and Mariam Sunny, Reuters
Markets are flat early Tuesday in holiday-thinned trading before head of the release of new data on how the U.S. economy fared in the third quarter.Futures for the S&P 500, the Dow Jones Industrial Average and Nasdaq are all essentially unchanged before the opening bell.Shares of the Danish pharmaceutical company Novo Nordisk jumped more than 7% overnight after U.S. regulators approved a pill version of the blockbuster weight-loss drug Wegovy, the first daily oral medication to treat obesity. Novo’s Wegovy is a GLP-1 drug that works like widely used injectables to mimic a natural hormone that controls appetite and feelings of fullness.Again touching new records, the price of gold rose 1.2% early Tuesday to $4,523.30 an ounce, adding to its consistent gains throughout the year. Silver rose 1.7%, to $69.71 an ounce.Oil prices edged higher early Tuesday after jumping more than 2% on Monday when the U.S. Coast Guard said it was pursuing another sanctioned oil tanker in the Caribbean.U.S. benchmark crude added 4 cents to $58.05 per barrel. The price of Brent crude, the international standard, gained 7 cents to $62.14 per barrel.Even after five straight days of gains, oil prices are down about 19% since the beginning of 2025 with demand lagging. U.S. factory conditions are weakening with activity readings hitting five-month lows, according to S&P Global.Markets in the U.S. will close early on Wednesday and remain closed on Thursday for the Christmas holiday. Yet several economic reports during the shortened week could shed more light on the condition and direction of the U.S. economy.The government on Tuesday releases the first of three estimates on gross domestic product, a reflection of how the broader U.S. economy fared in the third quarter. Also, the Conference Board will offer results from its December consumer confidence survey.Wednesday will bring a weekly update from the Labor Department on applications for jobless benefits, a proxy for U.S. layoffs.In Europe at midday Germany’s DAX edged 0.1% higher, while the CAC 40 in Paris slipped 0.2%. Britain’s FTSE 100 was unchanged.In Asian, Tokyo’s Nikkei 225 was flat at 50,412,87 and the dollar fell against the Japanese yen after officials in Tokyo warned they would intervene if the yen weakened further.The dollar traded at 155.95 yen, down from 157.04 yen late Monday. Instead of gaining after the Bank of Japan raised its key policy rate on Friday, the yen had weakened, drawing the usual objections from the Finance Ministry to larger than usual currency fluctuations.“The hint of currency intervention proved to be such a serious threat that the yen, which had been significantly oversold after the Bank of Japan meeting, rose from the ashes,” Alex Kruptsikevich of FXPro said in a commentary.The euro climbed to $1.1797 from $1.1762.Hong Kong’s Hang Seng gave up early gains to fall 0.1% to 25,774.14. The Shanghai Composite index edged 0.1% higher, to 3,919.98.South Korea’s Kospi added 0.3% to 4,117.32. Shipbuilder Hanwha Ocean’s shares jumped 12.5% after President Donald Trump said it would help build a new class of U.S. battleship at the Hanwha Philly shipyard.The S&P/ASX 200 in Australia jumped 1.1% to 8,795.70.In Taiwan, the Taiex advanced 0.6%, while India’s Sensex was nearly unchanged.
Elaine Kurtenbach and Matt Ott, AP Business Writers
In the fall of 2024, six college students joined forces to start an AI company together. Five of them had met while studying computer science, computer engineering, and electrical engineering at Georgia Tech in Atlanta. The sixth, its CEO, was pursuing a degree in childhood and adolescent development at Sacramento State, with an eye on becoming a grade-school teacher. That wasnt the only thing that made him an outlier. He also happened to have been in the tech industry for well over 30 yearslonger than his fellow founders had been alive.
The Georgia Tech students are Ian Boraks, Jacob Justice, Drake Kelly, Ella McCheney, and Abhinav Vemulapalli, all of whom happen to be 21. The Sac State student/tech veteran is Bill Nguyen, whose past startups amount to a guided tour of Silicon Valley trends over the years, from push technology to unified messaging to digital music to social networking to telehealth. Their new company, Olive.isOlive for shortis developing technology to make AI better at grasping the full meaning of spoken communications, as conveyed by elements, such as inflection and dialect, that current models may gloss over. It plans to offer its tech as a service for enriching AI-powered applications in education and other areas.
Olives name references the companys ambitious hope of fostering better understandingan olive branch, if you willbetween humans and machines. Its still in the process of researching and developing its AI model, and has raised $5 million in seed funding from education-focused venture capital firm Owl Ventures and Georgia Tech. The unusual founding team was a selling point to Owl, which also backed one of Nguyens previous ventures, Hazel Health.
As students themselves, the Georgia Tech founders are deeply connected and have a lot of recency with the ideal cohort of potential users that are going to benefit from all this technology, says Owls Lyman Missimer. But Bill is giving this team the full kind of Silicon Valley hustle out in the middle of Atlanta.
More than words can say
When I first met Nguyen in 2006, he was already a Silicon Valley vet and burbling with enthusiasm for a company hed founded called Lala, which helped people trade CDs through the mail. (It later moved into music streaming and was acquired by Apple in 2009.) His knack for high-energy pitchmanship helped his next company, the location-based, photo-centric social network Color, raise $41 million from firms such as Sequoia. It was ultimately best known for crashing and burning, as detailed in a 2011 Fast Company feature by Danielle Sacks.
Today, Nguyen is as exuberant as ever when discussing Olives goals and origin story, and doesnt seem to have aged nearly 20 years since our earliest encounter. As he explains it, AI models for turning speech into text, such as OpenAIs Whisper, have gotten uncannily good at correctly transcribing the literal meaning of what they hear. Yet the words we choose hardly convey our intent all by themselves. Elements such as inflection matter, tooand are sometimes absolutely crucial to understanding what someone is trying to say.
There’s definitely a lot in human conversation that gets chopped off by LLMs, says Nguyen. For example, if you ask me a question and I go, Yeeeeeees?he infuses the word with uncertaintyit’s not really a Yes. But an automated speech recognition system will basically truncate all that nuance, get rid of it, and just put it as a Yes.
If existing LLMs struggle with some of the subtleties of how we talk to each other, its at least in part because theyve been trained on material thats publicly available in vast quantities, such as podcasts. Such recordings probably sound really, really clean and they’re great audio, says Nguyen. But that’s not how we actually converse.
Nguyens interest in this current limitation of AI is intertwined with his long-standing passion for education. Years before he went back to school to become a teacher himselfhes halfway to earning his degreehe cofounded a public charter school near Lake Tahoe. As a result, he learned that few doctors in the area accepted Medicaid, greatly limiting student access to healthcare. That helped catalyze Hazel Health, which provides telehealth services through K-12 schools. It now serves 5,000 of them in 19 states.
Bill Nguyen and Ella McChesney, two of Olives six cofounders. [Photo: Olive.is]
The Hazel experience left Nguyen attuned to the real-world challenges schools face as they adopt technology. He provides an example relating to speech recognition. In a school district, one of the things that they have to focus on is the ability to understand when to do an intervention for a student around reading, he says. In theory, AI might help by analyzing audio of them speaking. But only if it understands what theyre saying, regardless of whether a student has a Mexican-American vernacular, African-American English vernacular, or Hawaiian vernacular.
To complicate matters, Children are especially hard [for AI to understand], because they have very limited vocabulary, says cofounder McChesney. So in order to express themselves, they find more creative ways to use words. And so what we’ve seen is that that can mean that models misinterpret them more, which can have negative consequences, especially when teachers are trying to leverage these tools to help them bring better experiences to the classroom.
The glimmer of opportunity in the idea of training AI models using audio that reflects how real people talkespecially studentsled to Olives founding. How Nguyen, based in Tahoe, ended up collaborating with a bunch of young techies in Atlanta is a story in itself. At first, he noodled on the idea with Justice, who is his son as well as a fellow Olive founder. As they forged ahead, the project expanded to include more people from Justices social circle.
McChesney, whose credentials include high-school work at the Department of Defense and four years interning at Lockheed Martin, hadrecently returned from a study trip to Korea when she joined the effort, right as Nguyen was prepping to pitch the company to investors. I got a text while I was in Costco from Drake, and he’s like, Bill wants your résumé, send it in the next 10 minutes, she remembers. Which would’ve been great if I wasn’t in a Costco with my phone at 5% and no cell service, because Costco is a giant steel box. She Airdropped her CV to a friend, who sent it to Nguyen just in time.
Olives iPhone app, as seen in the process of analyzing audio. [Screenshot: Olive.is]
The Atlanta-based cofounders do much of their partnering with Nguyen over Discord, though they quickly ran into the limitations of typed messages as a form of collaboration. In their minds, that only underlined the richness of verbal communications and the importance of teaching AI to comprehend it.
We’d always end up doing late-night calls, because that was the easiest way to communicate amongst ourselves, and the easiest way to really get our ideas across and understand what people are saying, says McChesney. There’s no ambiguity, the way there is in a lot of these text messages, and we can iterate faster. That really inspired what we’re trying to do here.
Its all in the data
To overcome current AI models limitations when it comes to capturing how humans express themselves verbally, Olive had to start with better data. More specifically, it had to start with raw audio of people talking to each other in unscripted situations. Our whole idea was if we can get really clean data sets, if we don’t remove any of the information, if we train a model that actually retains all of this context, then we can solve these mission-critical cases, says Nguyen.
More specifically, it decided to start with audio recordings of students engaged in conversations with professionals such as teachers and therapistsrecorded, it stresses, with the participants permission and awareness that they could be used for training.
As the company was finding sources of such material, Nguyens background at Hazel Health came in handy. We worked with school districts, we worked with universities, he says. The data set is pretty extensive now. Its north of 40,000 hours.
The company also built an iPhone app of its own, which I tried in pre-release form. Taking advantage of the beefy AI capabilities of Apples newest smartphone chips, it builds an understanding of the users needs by applying Olives models to verbal input. All processing is done on the phone, and input isnt used for training purposes.
Olive doesnt see this app replacing other AI tools so much as enhancing them. For instance, you could talk into the Olive app at length about an app youd like to create, then have it turn your verbal meanderings into a product requirements document to feed into a vibe-coding platform. Youre using your voice to have a more engaging conversation and actually hash it out, says McChesney. That’s what makes this so cool.
However, Olive isnt building its business around this iPhone app. Nor does it intend to provide fully-baked applications based on its technology. Instead, it plans to offer its AI model as a cloud-based service. Other companies will be able to use it as a technological layer in their own creations, providing them with a deeper understanding of speech than theyd get by relying entirely on existing voice models.
Along with possible uses in educationranging from tutoring to helping scale up college counselingOlive is targeting hiring, healthcare, and finance as areas where it hopes to find customers. These are all high stakes, and theyre all regulated in terms of what you can do, says Nguyen.
Theyre also all places where the limitations of existing AI may introduce harm that Olive hopes to overcome through better, fairer comprehension of a wider range of communications styles. You want AI access to be more equitable, says McChesney. You want everyone to be able to leverage these tools, because these tools are inherently part of our workforce.
The companys home page is currently devoted to a sobering blog post, Covert Racism: The Voice Inside the Machine. Heavily footnoted, it cites research that shows how prejudices are baked into AI in ways that can be difficult to detect even if its creators are actively trying to combat bias.
The part of it that’s mind-blowing to me is people are not getting jobs, Nguyen says. People are getting declined on loans. People are having adverse health effects. And no one knows why.
Olives potential to steer AI in a better direction might be particularly relevant in education, where the technology is still in the process of finding applications and the company has a shot at being foundational. Every major new technological shift is developed, built, and scaled, and then 10 years later it finally finds its way into education, says Owl Ventures Missimer. When we saw what Bill and team were building, we knew that the edtech market couldn’t wait 10 years for this type of technology, especially in a time where voice is becoming such a larger part of the technological stack.
With that in mind, Owl is helping to introduce Olive to other companies in its portfolio of education startups. They include Amira Learning, a 2025 Fast Company Next Big Things in Tech honoree that offers a suite of AI and neuroscience-based reading aids.
Given that Olives strategy is to provide its AI model as an ingredient for other companies products, those kinds of relationships have everything to do with its long-term fate. For now, it remains tiny, though its already grown to nine people. Nguyen says hes reveling in the hands-on experience of running something so tiny.
At his previous, larger startups, I, as the founder, was pretty separated from the actual engineering process, says Nguyen, who is not an engineer by background. But now Im not. I’m in the code base. I check it every day. I know what’s happening with it.
Once again, he exudes enthusiasm. Once again, hes working on something that taps into the tech industrys current obsession. Nguyen, who dubbed himself the Don Quixote of startups in Sacks 2011 article, may not be destined to run Olive forever. (Did I mention his intention to become a grade-school teacher?) But if this startup takes flight, having helping his youthful cofounders get it off the ground will be a legacy in itself.
