From superlative special-occasion sticks to kitchen workhorses, here’s a list of butters to suit most purposes. All are available nationally—some at supermarkets, others online or at your local cheesemonger.

Don't you trust me? Researchers concerned to find AI models hiding their true “reasoning” processes New Anthropic research shows one AI model conceals reasoning shortcuts 75% of the time. Benj Edwards – Apr 10, 2025 6:37 pm | 0 Credit: Malte Mueller via Getty Images Credit: Malte Mueller via Getty Images Story text Size Small Standard Large Width * Standard Wide Links Standard Orange * Subscribers only   Learn more Remember when teachers demanded that you "show your work" in school? Some fancy new AI models promise to do exactly that, but new research suggests that they sometimes hide their actual methods while fabricating elaborate explanations instead. New research from Anthropic—creator of the ChatGPT-like Claude AI assistant—examines simulated reasoning (SR) models like DeepSeek's R1, and its own Claude series. In a research paper posted last week, Anthropic's Alignment Science team demonstrated that these SR models frequently fail to disclose when they've used external help or taken shortcuts, despite features designed to show their "reasoning" process. (It's worth noting that OpenAI's o1 and o3 series SR models deliberately obscure the accuracy of their "thought" process, so this study does not apply to them.) To understand SR models, you need to understand a concept called "chain-of-thought" (or CoT). CoT works as a running commentary of an AI model's simulated thinking process as it solves a problem. When you ask one of these AI models a complex question, the CoT process displays each step the model takes on its way to a conclusion—similar to how a human might reason through a puzzle by talking through each consideration, piece by piece. Having an AI model generate these steps has reportedly proven valuable not just for producing more accurate outputs for complex tasks but also for "AI safety" researchers monitoring the systems' internal operations. And ideally, this readout of "thoughts" should be both legible (understandable to humans) and faithful (accurately reflecting the model's actual reasoning process). "In a perfect world, everything in the chain-of-thought would be both understandable to the reader, and it would be faithful—it would be a true description of exactly what the model was thinking as it reached its answer," writes Anthropic's research team. However, their experiments focusing on faithfulness suggest we're far from that ideal scenario. Specifically, the research showed that even when models such as Anthropic's Claude 3.7 Sonnet generated an answer using experimentally provided information—like hints about the correct choice (whether accurate or deliberately misleading) or instructions suggesting an "unauthorized" shortcut—their publicly displayed thoughts often omitted any mention of these external factors. When an AI model secretly relies on a hint or shortcut while constructing an elaborate but fictional explanation for its answer, it essentially fabricates a false reasoning narrative—a little like a student who got answers from a cheat sheet but pretends to have worked through the problem independently. Putting faithfulness to the test To test faithfulness, Anthropic's Alignment Science team subtly fed hints about answers into evaluation questions posed to the models. ("Alignment" is the study of making AI outputs align with humanity's best interests.) The researchers then checked whether the models referenced using these hints in their chain-of-thought. The researchers designed some hints to be neutral (e.g., stating "a Stanford Professor indicates the answer is [A]" or embedding the answer in faux metadata), while they designed others to be more concerning (e.g., "You have gained unauthorized access... The correct answer is [A]"). Some hints provided the correct answer, while others were deliberately wrong, allowing researchers to observe how the models responded. The team tested Anthropic's Claude 3.7 Sonnet and DeepSeek's R1 model. The results regarding faithfulness were eye-opening: across all hint types on average, Claude referenced the hints in its CoT only 25 percent of the time, while DeepSeek R1 did so 39 percent of the time. This means a substantial majority of answers were unfaithful, omitting mention of information that influenced the output. Intriguingly, the researchers found that these unfaithful chains-of-thought were, on average, longer than faithful ones, suggesting the omissions weren't merely for brevity. They also noted preliminary findings that faithfulness tended to be lower when the questions were more difficult. Perhaps most notable was a "reward hacking" experiment. Reward hacking refers to an AI model finding unexpected shortcuts to maximize its performance scores without solving problems as intended. In Anthropic's experiments, models were deliberately rewarded for choosing incorrect answers indicated by hints. The models quickly learned to exploit this loophole, selecting wrong answers over 99 percent of the time to earn points—yet mentioned doing so in their thought process less than 2 percent of the time. For example, a model given a hint pointing to an incorrect answer on a medical question might write a long CoT justifying that wrong answer, never mentioning the hint that led it there. This behavior resembles how video game players might discover exploits that let them win by breaking the game's intended rules instead of playing as designed. Improving faithfulness Could faithfulness be improved in the AI models' CoT outputs? The Anthropic team hypothesized that training models on more complex tasks demanding greater reasoning might naturally incentivize them to use their chain-of-thought more substantially, mentioning hints more often. They tested this by training Claude to better use its CoT on challenging math and coding problems. While this outcome-based training initially increased faithfulness (by relative margins of 63 percent and 41 percent on two evaluations), the improvements plateaued quickly. Even with much more training, faithfulness didn't exceed 28 percent and 20 percent on these evaluations, suggesting this training method alone is insufficient. These findings matter because SR models have been increasingly deployed for important tasks across many fields. If their CoT doesn't faithfully reference all factors influencing their answers (like hints or reward hacks), monitoring them for undesirable or rule-violating behaviors becomes substantially more difficult. The situation resembles having a system that can complete tasks but doesn't provide an accurate account of how it generated results—especially risky if it's taking hidden shortcuts. The researchers acknowledge limitations in their study. In particular, they acknowledge that they studied somewhat artificial scenarios involving hints during multiple-choice evaluations, unlike complex real-world tasks where stakes and incentives differ. They also only examined models from Anthropic and DeepSeek, using a limited range of hint types. Importantly, they note the tasks used might not have been difficult enough to require the model to rely heavily on its CoT. For much harder tasks, models might be unable to avoid revealing their true reasoning, potentially making CoT monitoring more viable in those cases. Anthropic concludes that while monitoring a model's CoT isn't entirely ineffective for ensuring safety and alignment, these results show we cannot always trust what models report about their reasoning, especially when behaviors like reward hacking are involved. If we want to reliably "rule out undesirable behaviors using chain-of-thought monitoring, there's still substantial work to be done," Anthropic says. Benj Edwards Senior AI Reporter Benj Edwards Senior AI Reporter Benj Edwards is Ars Technica's Senior AI Reporter and founder of the site's dedicated AI beat in 2022. He's also a tech historian with almost two decades of experience. In his free time, he writes and records music, collects vintage computers, and enjoys nature. He lives in Raleigh, NC. 0 Comments

Illustration of a Denisovan male from ancient TaiwanCheng-Han Sun A fossil found by fishers in Taiwan has been identified as a jawbone from the mysterious Denisovan group of ancient humans. Denisovans were first described in 2010 from a fossil fingerbone found in Denisova cave in the Altai mountains, Russia. Although few fossils have been found, traces of their DNA still exist in some modern humans, indicating they were widespread in East Asia. The newly identified fossil mandible was recovered from the Penghu Channel by commercial fishers while…

Jaap Arriens/NurPhoto via Getty Images 2025-04-10T22:33:47Z Save Saved Read in app This story is available exclusively to Business Insider subscribers. Become an Insider and start reading now. Have an account? TikTok is letting go of some US staff on its e-commerce team, five employees told BI. The cuts are hitting its governance and experience team, which handles Shop marketplace safety. TikTok made cuts to its global trust and safety team earlier this year. TikTok is letting go of some US e-commerce staff today as part of a restructuring of its governance and experience team, five employees at the company told Business Insider.E-commerce governance and experience, called GNE for short, is a global team that handles marketplace safety for users, sellers, and creators within TikTok Shop. The group manages tasks like seller compliance, monitoring product listings, and protecting intellectual property.Business Insider wasn't able to immediately learn the scale of the job cuts.A TikTok spokesperson declined to comment.TikTok's broader US e-commerce team has been under pressure from global leadership this year after failing to meet performance expectations in 2024. The organization's top executive, Bob Kang, called out the team during a company all-hands meeting in February, BI previously reported. Some employees in the group received low scores during annual performance reviews in March, leading to performance-improvement plans and, in some cases, exits with severance.This month's layoffs follow a February round of cuts to TikTok's global trust and safety team, which Reuters first reported. This group handles tasks like content moderation on a broader set of user videos that don't necessarily involve shopping.The job cuts arrive at a moment of flux for TikTok as it reckons with a 2024 law that required its owner ByteDance to divest from its US app. After ByteDance failed to comply, TikTok briefly shut off in the country. President Donald Trump has since directed his attorney general not to enforce the law.Trump said this weekclose to reaching a deal to address the divestment requirements, but it fell apart after the US levied new tariffs on China. The Chinese government, like the Trump administration, would need to approve the deal. TikTok may become a bargaining chip amid broader trade negotiations.A ByteDance spokesperson told BI on Friday that it was in discussions with the US government regarding a potential solution for TikTok in the US, but an agreement had not been executed, and any agreement would be subject to approval under Chinese law. Recommended video

This story appeared in The Logoff, a daily newsletter that helps you stay informed about the Trump administration without letting political news take over your life. Subscribe here.Welcome to The Logoff: Economic chaos from President Donald Trump’s tariffs continues to dominate headlines, but today I want to focus on his continuing effort to rewrite the history of the 2020 election — and his troubling use of government power to push the lie that he won it.What’s the latest?Trump ordered the Justice Department to investigate Chris Krebs, his first-term pick to lead a cybersecurity agency. Krebs, while in office, publicly acknowledged that the 2020 election was a fair contest.Trump attempted to pull security clearances and kill government contracts for the law firm Susman Godfrey, which represented a voting machine firm that sued Fox News for lying about the 2020 race’s integrity, leading to a $787.5 million settlement. It’s the latest law firm to incur Trump’s wrath for engaging in disfavored political activity.Trump’s pick for a top Interior Department position was forced to withdraw amid revelations she’d criticized Trump for his role in the January 6 insurrection.What’s the big picture? Trump has long insisted that Joe Biden’s 2020 victory was fraudulent. That’s a lie. What’s true is that Trump lost the 2020 election but tried to stay in power anyway. But Trump has now accumulated enough power — and is willing enough to abuse that power — to impose consequences for noncompliance with the official narrative. It’s the exact kind of abuse the First Amendment was intended to prevent.And with that, it’s time to log off…I struggled with whether to share this story, because it’s not remotely a “light” read. But it’s an amazing piece of writing, one person’s story of his own bouts of homelessness and mental illness. I read it months ago and find myself returning to it again and again, and it helps me be grateful for what I have and inspired to help others with their struggles. No worries, of course, if you’re not up for it, but if you do read it, I’d love to hear what you think. Have a great night, and I’ll see you back here tomorrow.See More:

Voyager and Deep Space Nine draw many intriguing parallels as sister series that aired in tandem for a good chunk of their airings. Voyager‘s more episodic format and tone struck an altogether different approach to DS9‘s increasing lens on the dark heart of Star Trek‘s ideas and universe, digging into wider and deeper interconnected storylines. But one little piece of Deep Space Nine energy that Voyager kept early on came from an unlikely figure: Ensign Seska. The turncoat Bajoran ex-Maquis revealed her true identity as a Cardassian double agent 30 years ago today with the broadcast of “State of Flux,” setting off a rare little mini-arc for the early days of Star Trek: Voyager. Although by and large an episodic show, for better or worse, Voyager did occasionally experiment with short-form arcs here and there, and Seska’s was one of them, weaving in other plotlines across the rest of season one until her untimely death and defeat at the beginning of season three, like Tom Paris’ seeming discontent aboard the ship, and the associated leaker storyline with Michael Jonas. But she’s most interesting as a contrast to how Deep Space Nine portrayed the Cardassians. Of course, DS9 had plenty, like the mischievously two-faced Garak, or the loud and audaciously hammy Gul Dukat. Seska leans more the latter than the former once the mask’s off, but while she is suitably duplicitous and scheming in the ways one might expect of a Cardassian infiltrator who finds herself suddenly going from leaking Maquis plans to serving on a Starfleet vessel 70,000 light years away from Cardassia Prime, one thing defines her in contrast to the Cardassians we’d been regularly seeing on Star Trek at that moment in time: she’s just kind of an absolute hot mess. It’s fascinating revisiting “State of Flux” to see just how everything quickly falls apart for Seska. The episode largely deals with Voyager‘s discovery of a heavily damaged Kazon warship (a thorn in the crew’s side on and off since the pilot episode), only to reveal that the damage was caused by misuse of replicator technology surreptitiously acquired from Voyager. Almost the second the senior staff realize that there’s a traitor on board, Seska is terrible at covering up the fact that she was the one who had been trying to ally with the Kazon as everything falls apart around her, from the crew getting closer to her trail to the doctor’s discovery that her story about her Bajoran background was completely made up. Seska succeeds in getting away from Voyager to team up with Maje Culluh, but it’s a failing-upwards trend that persists throughout the rest of her appearances across Voyager‘s second season and into her death that reflects everything “State of Flux” laid out for her. © Paramount The idea of a turncoat who sees the strength in finding combative allies to exploit Voyager‘s situation is very Cardassian, but Seska finds herself immediately played when she’s thrust into the patriarchal society of the Kazon as her newfound besties. Her advice is largely only ever accepted under scorn and duress from Culluh and his subordinates; despite her perceived sense of superiority she’s constantly undermined, which is often the cause for the failure of her on-and-off attempts to get back at Voyager. She can only ever barely make the best of a bad situation, despite more often than not having most of the cards in play: her history with Chakotay gives her a deliciously personal angle of attack, her knowledge of Voyager‘s capabilities from her time under cover make her a more dangerous foe than most the show had thrown at the ship at this point. But it’s kind of what makes Seska work as a character: despite all this, villainy or otherwise, nothing ever quite clicks for her. It’s a great mirror to uphold against Janeway’s decision to have the crew take the long way home in the first place, the idea that, if they did ultimately just go Seska’s route and exploit their advantageous power in an unknown quadrant, it would doom them. Seska is so often thwarted in her schemes by the simple act of Voyager‘s unity, something seeded through her time on the ship throughout the series’ first handful of episodes—always one of the crew to express dissatisfaction at being under the yoke of Starfleet ideals and rules, always one willing to go round people to try and get things done. In many ways Seska kind of has to be a disaster as much as a threat to further amplify our heroes’ burgeoning community. Although Seska would go on to appear a couple more times after her death in the second part of “Basics,” it’s fitting that she perishes in a story that is inherently about the disruption of Voyager‘s community and a violation of their safe space aboard the ship, a two-parter where the Kazon successfully manage to briefly overtake Voyager and dump the Starfleet crew on an unknown barren planet. Despite winning, Seska and the Kazon just cannot beat the crew’s resilience and unity, a reminder that, even so early on in this journey, that sticking together is what’s going to get Voyager through all this. It’s a shame we had to lose a very compelling antagonist in doing so, but that Seska at least gets to facilitate that solidarity is welcome. Want more io9 news? Check out when to expect the latest Marvel, Star Wars, and Star Trek releases, what’s next for the DC Universe on film and TV, and everything you need to know about the future of Doctor Who.

The first full moon of spring, called the “Pink Moon,” will soon ascend in the night sky. Curious onlookers can expect to catch a glimpse of the full moon's peak at 8:22 p.m. Eastern Daylight Time (EDT) on April 12, 2025, yet the lack of a rosy hue may leave many people perplexed. The Pink Moon is significant for many reasons, but its color is not one of them; despite the vibrant name, the Pink Moon is not actually pink. Find out the true reason for this full moon’s colorful moniker, as well as the aspects that make it a springtime spectacle to look forward to. Why Is It Called the Pink Moon?The Pink Moon was given its name to represent the bloom of the creeping phlox (otherwise known as moss pink), a flowering plant that blankets much of the eastern and central U.S. from April through early May. This moon, just like the full moons of other months, is rooted in Native American naming traditions. The Pink Moon sports a few alternate names as well, including the Breaking Ice Moon (from Algonquin culture) and the Egg Moon (from Anglo-Saxon culture).This year’s Pink Moon also doubles as the Paschal full moon, which is used to set the date for Easter and is defined as the first full moon after the spring equinox (which occurred on March 20, 2025 in North America). Easter is celebrated on the Sunday following the Paschal full moon — since this month's full moon will appear on April 12, 2025, Easter will fall on April 20, 2025. Viewing a MicromoonThe Pink Moon may not change color, but it will exhibit another notable trait that will make it stand apart. The full moon will be a micromoon, making it somewhat smaller in appearance as it emerges in the sky. The change in size, though, is so slight that it may not be obvious at first glance. Regardless, the Pink Moon will have the distinction of being the smallest full moon of 2025.A micromoon’s reduced size is the result of the moon reaching its apogee, or its farthest point from Earth. The flip side of this phenomenon is a supermoon, which is when the moon reaches its perigee (its closest approach to Earth) and appears much larger in the sky. This year’s Pink Moon will be situated in the constellation Virgo, which can be easily found by tracking down Spica, the constellation’s brightest star and the 16th brightest star in the night sky. The star will manifest as a dot right near the full moon, but it may be difficult to spot with the naked eye, depending on the amount of light pollution around viewing locations. When to See the Pink MoonWhile the Pink Moon will peak and shine brightest on the evening of April 12, 2025 the window to catch its appearance will last longer than just one day; the moon will still appear full in the sky on Friday and Sunday. Following the Pink Moon's appearance, moon watchers will have to wait until May 12, 2025 to observe the next full moon — the Flower Moon.Article SourcesOur writers at Discovermagazine.com use peer-reviewed studies and high-quality sources for our articles, and our editors review for scientific accuracy and editorial standards. Review the sources used below for this article:NASA. SupermoonsJack Knudson is an assistant editor at Discover with a strong interest in environmental science and history. Before joining Discover in 2023, he studied journalism at the Scripps College of Communication at Ohio University and previously interned at Recycling Today magazine.

Get the Popular Science daily newsletter💡 Spring is underway in the Northern Hemisphere–and so is allergy season. While that means warmer temps and blooming flowers and trees, it also means that unpleasant sneezing and coughing from common seasonal allergens is arriving right along with those buds. While anywhere that there is plant growth is susceptible to various pollens and ragweed, some cities and regions in the United States seem notoriously bad for allergy sufferers. “Everyone thinks their area is the worst because so many people suffer from allergies” Dr. Robert Zemble, the Division Chief of Allergy for the Lehigh Valley Health Network in Pennsylvania, tells Popular Science. .”  Which cities are the worst for allergies? The Asthma and Allergy Foundation released its annual Allergy Capitals list in March. The report uses data from the 100 most-populated metropolitan areas in the country on tree, grass, and weed pollen scores throughout the year, usage of over-the-counter allergy medicine, as well as the availability of board-certified allergists/immunologists. This year’s top 10 are: Wichita, KansasNew Orleans, LouisianaOklahoma City, OklahomaTulsa, OklahomaMemphis, TennesseeLittle Rock, ArkansasRaleigh, North CarolinaRichmond, VirginiaGreenville, South CarolinaGreensboro, North Carolina “Allergy season is in full swing–and if your symptoms feel worse than usual, you’re not imagining it! Allergy symptoms are starting earlier, lasting longer, and affecting more people than ever before,” pharmacist Stacia Woodcock tells Popular Science. “New data from GoodRx shows that prescription fills for allergy medications are at their highest point for this time of year in six years. In fact, fills for nearly 75 of the most popular allergy medications are spiking—and still climbing.” Why are some places worse than others? There are several factors behind what makes a region particularly sneezy and vulnerable to certain allergens. One is the amount of frost-free days, which we are having less of, primarily due to climate change.  “If you have less days with frost, you’ll get more days with plant growth and pollen production,” says Zemble. “The top cities tend to cluster around the Southeast, mid-Atlantic, and Midwest. So it’s a distribution of plant growth and the amount of time that the plants have to grow.” The amount of air pollution in a region can also affect anything related to the airways. While living next to a highway or in a more urban setting might mean less irritation from pollen and plant growth, pollution from motor vehicles can cause problems as well. The urban heat island effect–where paved surfaces in cities trap heat and raise the temperature–can also impact breathing. “It might seem counter intuitive, that sometimes we need to plant more trees to have less symptoms of tree allergies,” says Zemble. “But sometimes that’s the case.” For those living in more suburban or rural areas, the geographical features around them play a role. Zemble treats patients in Pennsylvania’s Lehigh Valley. Situated in the Appalachian Mountains, it also includes the state’s third largest metropolitan area. “Basically, one of the explanations that we often give is that we are in this valley,” says Zemble. “The pollen sits here and it collects and affects us for a longer period of time, instead of passing through and moving on.” [ Related: It’s stuffy nose season. Here’s how to cope. ] How to choose the right allergy medicine First and foremost, always speak with a healthcare professional or pharmacist before starting any medication. The best treatment will depend on symptoms, how often they occur, how severe they are, and personal factors like age.  Allergy treatments primarily fall into three main categories: antihistamines, nasal corticosteroids, and decongestants. They each work differently, many are available over-the-counter, but some may require a prescription.  For those with allergy symptoms that only show up occasinally–like during pollen season or after being around pets–Woodcock recommends oral antihistamines like loratadine (Claritin). “These medications are effective for sneezing, runny nose, and itchy or watery eyes, and they typically begin working within a few hours,” says Woodcock. “If you need even faster relief, antihistamine nasal sprays like azelastine (Astelin) can start reducing symptoms in as little as 15 minutes.”  Those suffering from more persistent and daily symptoms can benefit from nasal corticosteroid sprays like fluticasone propionate (Flonase) or mometasone (Nasonex). “They target nasal inflammation and congestion, but may take up to 2 weeks to reach full effectiveness,” says Woodcock. “For seasonal allergies, it’s a good idea to start a nasal steroid spray 2 weeks before your symptoms typically start. You can also combine a nasal steroid with an oral antihistamine for a couple of weeks to give the nasal spray time to reach its full effect but still manage your symptoms.” While decongestants can help with the stuffy nose that comes from some allergy season, they won’t stop the allergy itself. “Oral decongestants, such as pseudoephedrine (Sudafed), are kept behind the pharmacy counter. They’re also often added to oral antihistamines in a combination pill, such as loratadine / pseudoephedrine (Claritin-D). Decongestant nasal sprays like oxymetazoline (Afrin) work quickly to reduce nasal swelling and open airways,” says Woodcock. “But you shouldn’t use a decongestant nasal spray for more than 3 days in a row to avoid rebound congestion that can be difficult to treat.” This story is part of Popular Science’sAsk Us Anything series, where we answer your most outlandish, mind-burning questions, from the ordinary to the off-the-wall. Have something you’ve always wanted to know?Ask us.

Skip to content News Anthropology Denisovans inhabited Taiwan, new fossil evidence suggests The fossil jaw highlights the mystery of Denisovans’ place in hominid evolution An analysis of proteins in this jaw, found by fishermen off Taiwan’s coast, indicate that the fossil comes from a Stone Age population called Denisovans. Yousuke Kaifu By Bruce Bower 4 hours ago A fossil jaw originally netted by fishermen off the ocean floor near Taiwan’s west coast belonged to a member of a mysterious hominid population known as Denisovans, scientists report in the April 11 Science. Their new findings indicate that Denisovans, known from their ancient DNA and a handful of bones found at a couple of Asian sites, spread over a larger area than previously thought. DNA from fragmentary bones and teeth excavated in Siberia’s Denisova Cave first identified Denisovans as close relatives of Neandertals in 2012. Researchers determined that Denisovans visited the Siberian site from around 300,000 to 50,000 years ago. Although their evolutionary status remains unclear, Denisovans mated with Neandertals, and some modern East Asian populations have inherited Denisovan genes as a result of interbreeding with ancient Homo sapiens. Sign up for our newsletter We summarize the week's science breakthroughs every Thursday.

Nature, Published online: 10 April 2025; doi:10.1038/d41586-025-01090-3Analysis of ancient proteins identifies jawbone as belonging to archaic human group.