Technologies

Google’s AI Overviews Explain Made-Up Idioms With Confident Nonsense

The latest meme around generative AI’s hallucinations proves you can’t lick a badger twice.

Language can seem almost infinitely complex, with inside jokes and idioms sometimes having meaning for just a small group of people and appearing meaningless to the rest of us. Thanks to generative AI, even the meaningless found meaning this week as the internet blew up like a brook trout over the ability of Google search’s AI Overviews to define phrases never before uttered.

What, you’ve never heard the phrase «blew up like a brook trout»? Sure, I just made it up, but Google’s AI overviews result told me it’s a «colloquial way of saying something exploded or became a sensation quickly,» likely referring to the eye-catching colors and markings of the fish. No, it doesn’t make sense.

The trend may have started on Threads, where the author and screenwriter Meaghan Wilson Anastasios shared what happened when she searched «peanut butter platform heels.» Google returned a result referencing a (not real) scientific experiment in which peanut butter was used to demonstrate the creation of diamonds under high pressure.

It moved to other social media sites, like Bluesky, where people shared Google’s interpretations of phrases like «you can’t lick a badger twice.» The game: Search for a novel, nonsensical phrase with «meaning» at the end.

Things rolled on from there.

This meme is interesting for more reasons than comic relief. It shows how large language models might strain to provide an answer that sounds correct, not one that is correct.

«They are designed to generate fluent, plausible-sounding responses, even when the input is completely nonsensical,» said Yafang Li, assistant professor at the Fogelman College of Business and Economics at the University of Memphis. «They are not trained to verify the truth. They are trained to complete the sentence.»

Like glue on pizza

The fake meanings of made-up sayings bring back memories of the all too true stories about Google’s AI Overviews giving incredibly wrong answers to basic questions — like when it suggested putting glue on pizza to help the cheese stick.

This trend seems at least a bit more harmless because it doesn’t center on actionable advice. I mean, I for one hope nobody tries to lick a badger once, much less twice. The problem behind it, however, is the same — a large language model, like Google’s Gemini behind AI Overviews, tries to answer your questions and offer a feasible response. Even if what it gives you is nonsense.

A Google spokesperson said AI Overviews are designed to display information supported by top web results, and that they have an accuracy rate comparable to other search features.

«When people do nonsensical or ‘false premise’ searches, our systems will try to find the most relevant results based on the limited web content available,» the Google spokesperson said. «This is true of search overall, and in some cases, AI Overviews will also trigger in an effort to provide helpful context.»

This particular case is a «data void,» where there isn’t a lot of relevant information available for the search query. The spokesperson said Google is working on limiting when AI Overviews appear on searches without enough information and preventing them from providing misleading, satirical or unhelpful content. Google uses information about queries like these to better understand when AI Overviews should and should not appear.

You won’t always get a made-up definition if you ask for the meaning of a fake phrase. When drafting the heading of this section, I searched «like glue on pizza meaning,» and it didn’t trigger an AI Overview.

The problem doesn’t appear to be universal across LLMs. I asked ChatGPT for the meaning of «you can’t lick a badger twice» and it told me the phrase «isn’t a standard idiom, but it definitely sounds like the kind of quirky, rustic proverb someone might use.» It did, though, try to offer a definition anyway, essentially: «If you do something reckless or provoke danger once, you might not survive to do it again.»

Read more: AI Essentials: 27 Ways to Make Gen AI Work for You, According to Our Experts

Pulling meaning out of nowhere

This phenomenon is an entertaining example of LLMs’ tendency to make stuff up — what the AI world calls «hallucinating.» When a gen AI model hallucinates, it produces information that sounds like it could be plausible or accurate but isn’t rooted in reality.

LLMs are «not fact generators,» Li said, they just predict the next logical bits of language based on their training.

A majority of AI researchers in a recent survey reported they doubt AI’s accuracy and trustworthiness issues would be solved soon.

The fake definitions show not just the inaccuracy but the confident inaccuracy of LLMs. When you ask a person for the meaning of a phrase like «you can’t get a turkey from a Cybertruck,» you probably expect them to say they haven’t heard of it and that it doesn’t make sense. LLMs often react with the same confidence as if you’re asking for the definition of a real idiom.

In this case, Google says the phrase means Tesla’s Cybertruck «is not designed or capable of delivering Thanksgiving turkeys or other similar items» and highlights «its distinct, futuristic design that is not conducive to carrying bulky goods.» Burn.

This humorous trend does have an ominous lesson: Don’t trust everything you see from a chatbot. It might be making stuff up out of thin air, and it won’t necessarily indicate it’s uncertain.

«This is a perfect moment for educators and researchers to use these scenarios to teach people how the meaning is generated and how AI works and why it matters,» Li said. «Users should always stay skeptical and verify claims.»

Be careful what you search for

Since you can’t trust an LLM to be skeptical on your behalf, you need to encourage it to take what you say with a grain of salt.

«When users enter a prompt, the model just assumes it’s valid and then proceeds to generate the most likely accurate answer for that,» Li said.

The solution is to introduce skepticism in your prompt. Don’t ask for the meaning of an unfamiliar phrase or idiom. Ask if it’s real. Li suggested you ask «is this a real idiom?»

«That may help the model to recognize the phrase instead of just guessing,» she said.

Technologies

Kevin Weil and Bill Peebles exit OpenAI as company continues to shed ‘side quests’ | TechCrunch

Kevin Weil and Bill Peebles are leaving OpenAI as the company shuts down Sora and folds its science team, signaling a sharp pivot away from consumer moonshots toward enterprise AI.

OpenAI is losing two of the architects of its most ambitious moonshots. Kevin Weil, who led the company’s science research initiative, and Bill Peebles, the researcher behind AI video tool Sora, both announced their departures on Friday. The exits come as OpenAI consolidates around enterprise AI and its forthcoming “superapp.”

The departures follow OpenAI’s decision to cut back on “side quests,” including customer-facing bets like Sora and OpenAI for Science. Sora, which was losing an estimated $1 million per day in compute costs, was shut down last month.

OpenAI for Science was the internal research group behind Prism, an AI-powered platform that promised to accelerate scientific discovery. It’s being absorbed into “other research teams,” according to Weil’s social media post announcing the news.

“It’s been a mind-expanding two years, from Chief Product Officer to joining the research team and starting OpenAI for Science,” Weil wrote. “Accelerating science will be one of the most stunningly positive outcomes of our push to AGI.”

The team had a short and bumpy road after its formal announcement in October 2025. Weil deleted a tweet claiming GPT-5 had solved 10 previously unsolved Erdős mathematical problems, but that claim fell apart immediately when the mathematician who runs the website erdosproblems.com called it out.

Weil’s departure comes a day after his team released GPT-Rosalind, a new model to accelerate life sciences research and drug discovery.

In a social media post announcing his departure, Peebles credited Sora with igniting a “huge amount of investment in video across the industry,” and argued that the kind of research that produced the video tool requires space away from the company’s mainline roadmap.

“Cultivating entropy is the only way for a research lab to thrive long-term,” he wrote.

OpenAI is also losing Srinivas Narayanan, its chief technology officer of enterprise applications, Wired reports. Narayanan reportedly announced the news internally that he was leaving to spend more time with family.

This article was updated to include the departure of Srinivas Narayanan.

Topics

StrictlyVC kicks off the year in SF. Get in the room for unfiltered fireside chats with industry leaders, insider VC insights, and high-value connections that actually move the needle. Tickets are limited.

Sam Altman’s World Initiative Expands Human Verification to Tinder and Beyond

Sam Altman’s World initiative expands its human verification technology, starting with a global rollout on Tinder and introducing new features like Concert Kit to combat scalpers and deepfakes.

At a popular spot near the San Francisco waterfront, Sam Altman’s verification project World marked its latest phase and ambitious growth. The initiative begins by partnering with Tinder.

Tools for Humanity (TFH), the firm driving the World project, revealed on Friday that it will embed its verification technology into dating platforms, event ticketing networks, corporate entities, email services, and various other sectors of daily life.

«The world is approaching incredibly advanced AI, which is accomplishing remarkable things,» Altman noted while addressing a full room at The Midway. «However, we are moving toward an era where AI-generated content will surpass human-created material,» he continued. «I am certain many of you [have experienced] moments where you question, ‘Am I communicating with an AI or a real person, or what is the ratio, and how can I verify?’»

World (previously known as Worldcoin) sets itself apart from other identity verification services by enabling the confirmation that a genuine, living individual is accessing a digital platform while maintaining their privacy. This relies on sophisticated cryptographic methods (specifically, «zero-knowledge proof-based authentication»). The result: The organization is developing what it terms «proof of human» solutions, which are systems designed to confirm human presence in an environment increasingly populated by AI agents and automated bots.

Its primary verification instrument is a spherical device named the Orb, which captures a user’s eye patterns to generate a distinct, anonymous cryptographic code (referred to as a verified World ID). This code can then be utilized to access World’s services, though individuals may also use the World application without possessing an Orb.

Altman’s speech on Friday was concise (TFH’s co-founder and CEO, Alex Blania, was missing due to unexpected hand surgery, according to Altman). He subsequently passed the presentation to World’s chief product officer, Tiago Sada, and his colleagues.

Sada detailed that World is introducing the latest iteration of its application (the previous release was unveiled during a December gathering), alongside numerous new technology integrations.

World has been working for a while to introduce a verification system for dating applications — particularly Tinder. Last year, Tinder initiated a World ID trial program in Japan. This trial reportedly succeeded, prompting World to announce that Tinder would roll out its verification integration across global markets, including the U.S. The system adds a World ID badge to the profiles of users who complete its verification steps, confirming their authenticity as real individuals.

World is also targeting the entertainment sector with a new feature called Concert Kit, allowing musicians to set aside specific ticket quantities for World ID-verified attendees. This aims to protect fans from scalpers who frequently employ automated ticket-purchasing bots to secure seats. Concert Kit works with major ticketing platforms like Ticketmaster and Eventbrite, and the company is highlighting it through collaborations with 30 Seconds to Mars and Bruno Mars — both of whom intend to utilize it for their upcoming tours.

The gathering featured numerous additional announcements, including those focused on corporate clients. A Zoom/World ID verification integration aims to counter a perceived deepfake risk in business calls, and a Docusign partnership is designed to ensure

Technologies

From Acquisition Talks to Rivals: How Stripe and Airwallex’s Paths Diverged

Once on the verge of a $1.2 billion acquisition by Stripe, Airwallex founder Jack Zhang rejected the deal to pursue a long-term vision, now positioning the company as a formidable rival in the global payments infrastructure space.

Jack Zhang, a 34-year-old entrepreneur who had been leading his startup for three and a half years, found himself in a pivotal meeting with Michael Moritz, a prominent investor from Sequoia Capital. Invited to Moritz’s San Francisco residence, which offered stunning views of the Golden Gate Bridge, Zhang was presented with an offer: Stripe intended to acquire Airwallex for $1.2 billion. At that moment, Airwallex was generating approximately $2 million in annualized revenue, making the valuation seem incredibly lucrative. Moritz emphasized that Patrick Collison, Stripe’s founder, was a visionary leader, suggesting the acquisition could lead to extraordinary growth. Zhang spent two weeks in San Francisco grappling with the decision, eventually agreeing to the deal.

Yet, he soon flew back to Australia, nearly 8,000 miles away. Reflecting on the decision, Zhang explained, ‘I had to delve into my core motivations for building Airwallex. I was only three and a half years into the venture, which had grown exponentially in 2018. I had just begun to experience the thrill of entrepreneurship, which is what I had always dreamed of.’

Two of his co-founders opposed the acquisition, which influenced his choice. However, Zhang cited a clearer moment of clarity when he looked at the whiteboard in his office. The unfinished vision remained: to create financial infrastructure enabling businesses to operate globally as if they were local entities.

This decision appears increasingly justified. Airwallex now reports over $1.3 billion in annualized revenue, growing 85% annually, and processes nearly $300 billion in transaction volume. Zhang attributes this success to the deliberate challenges they faced.

Zhang’s journey began in Qingdao, China, and he moved to Melbourne at 15 with minimal English, living with a host family. After his family’s financial struggles, he worked multiple jobs to fund his computer science degree at the University of Melbourne, including bartending, dishwashing, gas station shifts, and farm work. He later worked in trading code development at an Australian investment bank, a role that paid well but lacked personal fulfillment.

Secure your next funding round, hire key talent, and seize emerging opportunities at TechCrunch Disrupt 2026. Join 10,000+ founders, investors, and tech leaders for three days of 250+ tactical sessions, valuable connections, and groundbreaking innovation. Register now to save up to $410.

Before founding Airwallex, Zhang launched approximately 10 ventures, including a magazine at 14, a real estate development firm, import-export businesses dealing in wine and olive oil between Australia and Asia, textiles in the opposite direction, and a burger chain.

The concept for Airwallex emerged while Zhang ran a Melbourne coffee shop. When attempting to pay suppliers in Brazil, Indonesia, and Guatemala, co-founder Max Li observed payments vanishing into correspondent banking systems, often flagged or frozen by U.S. intermediary banks enforcing OFAC sanctions. ‘This prompted me to investigate correspondent banking and SWIFT systems to build our own global money movement network,’ Zhang noted.

That vision has scaled significantly. Airwallex now holds nearly 90 financial licenses across 50 markets, far exceeding Stripe’s estimated half. Acquiring these licenses required immense effort; in Japan, it took seven years. In some emerging markets, the company acquired shell companies with outdated licenses and rebuilt their technology from scratch.

‘You can’t just vibe-code an integration with Mexico’s central bank,’ Zhang remarked. ‘Access requires a secure room and biometric scans.’ These licenses are not merely regulatory formalities. In Japan, for example, Stripe and Square must transfer funds immediately to merchants’ bank accounts, whereas Airwallex, holding a fund transfer operator license, retains funds within its ecosystem. This allows customers to issue bank accounts, cards, and spend locally without funds leaving the platform.

The foreign exchange advantages are significant. A U.S. merchant settling transactions in Australian dollars avoids the 2% to 3% conversion fees typically charged by processors like Stripe to move funds back to U.S. dollars. Instead, they can use local balances to pay vendors, manage payroll, and cover digital marketing at interbank rates.

‘You no longer operate like a traditional U.S. company,’ Zhang explained. ‘You function as a global entity without the need to physically establish offices worldwide.’ This strategic approach, which Zhang calls the ‘path of maximum resistance,’ has created competitive barriers. ‘It took us six and a half years to reach $100 million in annual recurring revenue,’ Zhang stated. ‘But after that, it took just over three years to hit a billion.’ The competitive logic, in his telling, is clear.