Warning: contains finale spoilers for Dept. Q See it? Blink and you won’t have, but it was there for a good half second: a smile on the face of DCI Carl Morck. It’s a rare enough occurrence across nine episodes of this Netflix crime drama to make it worthy of note. Matthew Goode’s irascible detective frowned, scowled and spitted out sarcasm through his new department’s search for missing prosecutor Merritt Lingard, until the show’s very last moment, when he smiled. He’d earned it. Three months earlier, Morck’s cold case unit had rescued Lingard and used dirt dug up during the investigation to blackmail the Lord Advocate (the most senior law officer in Scotland) into a funding hike, a new work car, and fast-tracked DI status for Syrian copper and calmly proficient muscle Akram Salim. And now, Morck’s partner Hardy, formerly thought to have permanently lost the use of his legs after they were both shot in the line of duty, was walking again. It was a happy ending, or at least as happy as things get in a dank Edinburgh basement filled with unsolved case files detailing terrible crimes. With major Dept. Q finale spoilers, let’s dig into the revelations about Merritt’s kidnappers, what we learned about the Leith Park shooting, and more. Who Kidnapped Merritt Lingard and Why? Merritt’s kidnappers were Lyle and Ailsa Jennings, respectively the younger brother and mother of Harry Jennings, Merritt’s boyfriend as a teenager. She was being held on their property on her childhood island home of Mhòr, in a pressurised hyperbaric chamber inside a condemned building that was formerly part of Ailsa’s husband’s oceanography business Shorebird Ocean Systems – the logo for which was the cormorant symbol identified by Merritt’s brother William as on the hat of her stalker and kidnapper. Ailsa was a twisted and abusive mother to Harry and Lyle, as well as a murderer who killed her husband by burning the house down by flicking lighted matches at him while he slept. As a result, Lyle grew up to be a violent psychopath whose teenage crimes were covered up by the family and by the island’s police officer John Cunningham. John knew that Lyle was responsible for Merritt’s disappearance, but didn’t know she was still alive, having believed Lyle when he lied that Merritt had died by falling off the ferry from the mainland, just like Lyle’s brother Harry had done years earlier. Lyle and Ailsa blamed Merritt for Harry’s death because he jumped from the ferry while being pursued by John, who was chasing him because he was suspected of having severely beaten Merritt’s brother William into a coma during an interrupted robbery of the Lingard home. The robbery plan had been seeded by Merritt to steal her mother’s jewellery from her alcoholic father in order to sell for money to run away from home. In fact, Lyle was the one who had attacked William, being unhinged and thinking that he was protecting his brother. Harry and Ailsa covered up Lyle’s guilt and before his death, Harry took the blame. Lyle Jennings, Godhaven, and Sam Haig Lyle, who had a history of stalking, breaking-and-entering, and animal murder, spent time in youth psychiatric facility Godhaven, where he stalked another patient named Sam Haig. The pair fought, and Sam injured Lyle’s eye so severely that it was left permanently discoloured. Years later as an adult who had become an investigative journalist with no online presence, Sam sought Lyle out to apologise for the injury and invite him to come climbing at his favourite spot. Lyle used the information he had on Sam to assume his identity and stalk Merritt Lingard, who didn’t recognise him as an adult. He pretended to be investigating corruption at Merritt’s workplace, and targeted her, eventually starting a sexual relationship still pretending to be Sam. When Merritt told Lyle-as-Sam that she was leaving to go to Mhòr on the 10am ferry the next day, Lyle used that information to kidnap her during the crossing. Merritt’s brother William, who had been unable to speak since his teenage coma, had spotted Lyle on the ferry and because of Lyle’s distinctive hat, recognised him as the man who’d been stalking their house. He tried to communicate this to Merritt by throwing his own hat, and then struck her when he became afraid. It was when Merritt went to retrieve William’s hat that she was taken. Join our mailing list Get the best of Den of Geek delivered right to your inbox! Two days after Lyle kidnapped Merritt, he returned to the mainland to tie up loose ends and kill the real Sam Haig. He lured Sam to a remote spot, beat him to death, staged his corpse to look as though he had been climbing, and threw it off a cliff. Eventually, Morck and co. worked out that somebody else was pretending to be Sam Haig when they had evidence that Merritt had been with “Sam” at a hotel at the same time that the real Sam was having an affair with his friend’s wife at a different hotel. What Happened to Lyle and Ailsa Jennings? Merritt’s kidnappers both died – Ailsa by her own hand as she shot herself in her car before being taken into custody by the police, and Lyle by Akram’s hand, or rather, his trigger finger. After Lyle shot buckshot into Morck’s shoulder, Akram threw a knife at him, disarmed and shot him. Previously, Lyle had brutally killed Constable John Cunningham with a hammer when he learned that Merritt was still alive. Mark Bonnar’s character Stephen Burns had disallowed the appearance of a key witness in the Graham Finch murder trial after Burns’ daughter was run off the road by one of Finch’s goons as an intimidation tactic. Prisoner Kirsty Atkins was willing to testify that she had previously met Finch’s now-dead wife at a shelter for survivors of domestic violence, and that Finch routinely hit her. Because Burns wouldn’t allow her testimony, Finch was acquitted of his wife’s murder despite clearly having been guilty of it. Perhaps he also passed on the information about Kirsty Atkins to Finch’s lawyer, which led to Kirsty being viciously attacked and partially blinded in prison on Finch’s orders to keep her quiet. Who Is the Leith Park Shooter? We still don’t know who shot PC Anderson dead, and attempted to shoot Morck and Hardy dead at the Leith Park flat. If we can take the contents of Morck’s dream as gospel, then the shooting was done on the orders of Leith Park criminal kingpin Eugene Errington, whom we’ve yet to meet. According to Morck’s dream, in which he shot Errington’s muscle Charlie Bell in the head, Errington has a baby with Caroline Kerr – the witness who was intimidated into withdrawing her statement about the shooting. We also know, or at least have a pretty good idea, that PC Anderson was a corrupt officer working for Errington, and that he may even have been the one to stab the victim – thought to have been a police informant – in the head. Anderson was a top recruit pretending to be a klutz, Morck realised, in order to cover up the fact that his fingerprints were all over the victim’s flat. He also lied about being at the flat for a routine welfare check after being alerted by the victim’s daughter – that was a quick-thinking ruse to explain his presence when Morck and Hardy showed up there unexpectedly. The last we saw of the Leith Park case, Moira looked to be taking it away from its lead investigator and possibly closing it, unsolved, which may point to her also having been compromised by Errington. As Hardy was one of the shooting’s victims, surely that can’t be the case file that she gave to Hardy to investigate without Morck in the season’s closing moments? Dept. Q is streaming now on Netflix.

Reasoning tasks are a fundamental aspect of artificial intelligence, encompassing areas like commonsense understanding, mathematical problem-solving, and symbolic reasoning. These tasks often involve multiple steps of logical inference, which large language models (LLMs) attempt to mimic through structured approaches such as chain-of-thought (CoT) prompting. However, as LLMs grow in size and complexity, they tend to produce longer outputs across all tasks, regardless of difficulty, leading to significant inefficiencies. The field has been striving to balance the depth of reasoning with computational cost while also ensuring that models can adapt their reasoning strategies to meet the unique needs of each problem. A key issue with current reasoning models is the inability to tailor the reasoning process to different task complexities. Most models, including well-known ones like OpenAI’s o1 and DeepSeek-R1, apply a uniform strategy—typically relying on Long CoT across all tasks. This causes the “overthinking” problem, where models generate unnecessarily verbose explanations for simpler tasks. Not only does this waste resources, but it also degrades accuracy, as excessive reasoning can introduce irrelevant information. Approaches such as prompt-guided generation or token budget estimation have attempted to mitigate this issue. Still, these methods are limited by their dependence on predefined assumptions, which are not always reliable for diverse tasks. Attempts to address these issues include methods like GRPO (Group Relative Policy Optimization), length-penalty mechanisms, and rule-based prompt controls. While GRPO enables models to learn different reasoning strategies by rewarding correct answers, it leads to a “format collapse,” where models increasingly rely on Long CoT, crowding out more efficient formats, such as Short CoT or Direct Answer. Length-penalty techniques, such as those applied in methods like THINKPRUNE, control output length during training or inference, but often at the cost of reduced accuracy, especially in complex problem-solving tasks. These solutions struggle to achieve a consistent trade-off between reasoning effectiveness and efficiency, highlighting the need for an adaptive approach. A team of researchers from Fudan University and Ohio State University introduced the Adaptive Reasoning Model (ARM), which dynamically adjusts reasoning formats based on task difficulty. ARM supports four distinct reasoning styles: Direct Answer for simple tasks, Short CoT for concise reasoning, Code for structured problem-solving, and Long CoT for deep multi-step reasoning. It operates in an Adaptive Mode by default, automatically selecting the appropriate format, and also provides Instruction-Guided and Consensus-Guided Modes for explicit control or aggregation across formats. The key innovation lies in its training process, which utilizes Ada-GRPO, an extension of GRPO that introduces a format diversity reward mechanism. This prevents the dominance of Long CoT and ensures that ARM continues to explore and use simpler reasoning formats when appropriate. The ARM methodology is built on a two-stage framework. First, the model undergoes Supervised Fine-Tuning (SFT) with 10.8K questions, each annotated across four reasoning formats, sourced from datasets like AQuA-Rat and generated with tools such as GPT-4o and DeepSeek-R1. This stage teaches the model the structure of each reasoning format but does not instill adaptiveness. The second stage applies Ada-GRPO, where the model receives scaled rewards for using less frequent formats, such as Direct Answer or Short CoT. A decaying factor ensures that this reward gradually shifts back to accuracy as training progresses, preventing long-term bias toward inefficient exploration. This structure enables ARM to avoid format collapse and dynamically match reasoning strategies to task difficulty, achieving a balance of efficiency and performance. ARM demonstrated impressive results across various benchmarks, including commonsense, mathematical, and symbolic reasoning tasks. It reduced token usage by an average of 30%, with reductions as high as 70% for simpler tasks, compared to models relying solely on Long CoT. ARM achieved a 2x training speedup over GRPO-based models, accelerating model development without sacrificing accuracy. For example, ARM-7B achieved 75.9% accuracy on the challenging AIME’25 task while using 32.5% fewer tokens. ARM-14B achieved 85.6% accuracy on OpenBookQA and 86.4% accuracy on the MATH dataset, with a token usage reduction of over 30% compared to Qwen2.5SFT+GRPO models. These numbers demonstrate ARM’s ability to maintain competitive performance while delivering significant efficiency gains. Overall, the Adaptive Reasoning Model addresses the persistent inefficiency of reasoning models by enabling the adaptive selection of reasoning formats based on task difficulty. The introduction of Ada-GRPO and the multi-format training framework ensures that models no longer waste resources on overthinking. Instead, ARM provides a flexible and practical solution for balancing accuracy and computational cost in reasoning tasks, making it a promising approach for scalable and efficient large language models. Check out the Paper, Models on Hugging Face and Project Page. All credit for this research goes to the researchers of this project. Also, feel free to follow us on Twitter and don’t forget to join our 95k+ ML SubReddit and Subscribe to our Newsletter. NikhilNikhil is an intern consultant at Marktechpost. He is pursuing an integrated dual degree in Materials at the Indian Institute of Technology, Kharagpur. Nikhil is an AI/ML enthusiast who is always researching applications in fields like biomaterials and biomedical science. With a strong background in Material Science, he is exploring new advancements and creating opportunities to contribute.Nikhilhttps://www.marktechpost.com/author/nikhil0980/This AI Paper Introduces WEB-SHEPHERD: A Process Reward Model for Web Agents with 40K Dataset and 10× Cost EfficiencyNikhilhttps://www.marktechpost.com/author/nikhil0980/This AI Paper Introduces MMaDA: A Unified Multimodal Diffusion Model for Textual Reasoning, Visual Understanding, and Image GenerationNikhilhttps://www.marktechpost.com/author/nikhil0980/This AI Paper Introduces Differentiable MCMC Layers: A New AI Framework for Learning with Inexact Combinatorial Solvers in Neural NetworksNikhilhttps://www.marktechpost.com/author/nikhil0980/This AI Paper Introduces GRIT: A Method for Teaching MLLMs to Reason with Images by Interleaving Text and Visual Grounding

"AI is inevitable."That's a phrase that's rattled around my head for a month. Not willingly mind you. It's taken up lodging in my grey matter after hearing it in meetings, reading it in emails, and seeing it buffeted back and forth across Bluesky, LinkedIn, and Discord.It's not a convincing phrase. If you hear it from AI boosters it's easy to brush off as raw hype, and if you hear it from doomsayers it can lull you into a sense of fatalism. But as the philosopher Natasha Bedingfield told us in 2004, today is where the book begins, the rest is still unwritten. Nothing, for better or worse, is inevitable.But in those various calls another phrase—one you may have heard at your studio—has slipped past more unnoticed: "AI is just a tool. It can be used for good or evil, like any other tool."After all this is a business where we use tools for good or evil, right and wrong, correctly and incorrectly. We debate the effectiveness of Unity, Unreal, or Godot. We agonize over whether to use procedural versus hand-crafted content. We debate and discuss the topic so much that Game Developers Conference has a whole Tools Summit dedicated to craft of making game development software.Viewing generative AI through the neutral lens of tool assessment is natural—and I'll go so far as to say admirable—for our community. It's a method we use to get past hype and bombast, to try and take technology on its own terms and see how it fits our purposes. And as the 2025 GDC State of the Industry report tells us, some developers are adopting generative AI, plenty of them not bought in on the hype but through the act of seeking the right tool for the job.Related:But looking at generative AI as 'just a tool' is a deeply flawed lens. That phrase betrays a quiet cynicism (one we hear often from opponents of firearm regulation in the United Stats). Because nothing—not generative AI, not a firearm, not even a hammer—is "just a tool."The function of tools is influenced by their formConsider two tools found in many American households: the claw hammer and the handgun.Normally Game Developer restricts itself to the craft of making video games but I promise this is relevant. Guns are another tool where neutralizing rhetoric is deployed to downplay a tool's negative effects. I grew up in a gun-owning house in a gun-owning neighborhood in suburban Maryland. There were probably four handguns sitting in lockboxes across two rooms, a few rifles and shotguns in a vault in the basement, and one questionably legal World War I firearm tucked away in a closet. The NRA's mantra of "guns don't kill people, people kill people" was commonplace. A neighbor of mine laughed when I advocated for stronger regulations on gun ownership on the basis of "guns are meant to kill." "Guns aren't meant to kill," I recall him saying. "Cars can kill people. Does that mean cars are meant for killing?"His point boils down to this: The outcome of the tool's use is not worth considering when discussing regulation, only its potential use. A gun is a tool and the user has control over a tool is used.Cars are already tightly regulated and cost thousands of dollars, making his point moot, so we'll break down the construction of the claw hammer instead. We generally refer to hammers as being used to pound nails into wood, but I mainly use mine for hammering anchors into drywall because I'm a theater kid and was taught in crew to trust screws.In either case, the physical shape of the claw hammer dictates its most common purpose. The handle extends into a metal object that is blunt at one end, and clawed on the other. The design follows the swing of the human arm, transferring kinetic energy generated by the bicep, down the elbow, through the wrist, and into the blunt end.We also know that claw hammers are not useful for every form of transferring this energy. Variations on hammer design like the ball-peen hammer show how this basic purpose needs to be altered for different tasks. The shape and the material changes depending on the purpose. To sell more hammers, companies invest in better materials and affordances like rubber grips to make their use more comfortable.Like a firearm, hammers can be used as weapons. That same transference of force can be used to harm another living being. Video games sometimes place hammers in a players' loadout alongside guns, grenades, and weapons of war.Neither the hammer nor the firearm is "just" a tool. They are tools that are optimized for a purpose. We can study that purpose, and cast judgements about a tool's safety, merits, and need to be regulated based on that.But. The shape of the hammer is not an efficient way to inflict harm. This is supported by data from the FBI Crime Statistics survey, which gathers data filed by police departments that participate in assembling data. "Handgun" is the most common weapon used in homicides, and "knife/cutting instrument" ranks higher than "blunt objects." That's because handguns are an incredibly efficient means of wounding living beings.Let's break down the handgun the way we did the hammer. Handguns are assembled from an assortment of components that transfer the squeeze of a trigger into the strike of a hammer against a firing pin, which strikes the primer of a bullet's cartridge and sends it propelling out of the tube. Though some bullets seen in larger firearms are meant to penetrate metal, a handgun's bullet is envisioned and designed to cut through flesh.Image via Adobe Stock.These constraints make handguns efficient at few other tasks. In a pinch you could use the butt of a handgun as a hammer. I can't find any data about them being used for that purpose. I can only wander onto a construction site and count the number of firearms in toolboxes as a general sample size.Neither the hammer nor the firearm is "just" a tool. They are tools that are optimized for a purpose. We can study that purpose, and cast judgements about a tool's safety, merits, and need to be regulated based on that. Firearm advocates oppose this process through neutralizing language because it's difficult to dispute the correlation between the number of guns versus the number of murders and assaults with guns in a geographic area.Generative AI proponents sometimes regurgitate that language when defending this new technology. Because like the gun lobby, they don't want the purpose of generative AI decided by its outcomes, only its potential.What is that purpose? It may be the death of truth itself.Generative AI is broadly used to deceive through mimicryGenerative AI is a tool for deception.That's not what its biggest backers will tell you. It's broadly pitched as a tool for efficiency. But efficiency is hard to measure and easy to game. Deception is loud and obvious. Students are using it to cheat on papers. Scam calls with AI-generated voices are on the rise. The Department Human Health and Services published a study citing secretary Kennedy's unfounded health views that cites nonexistent studies, likely generated through AI. There was that cadre of YouTubers creating AI-generated fake movie trailers to attract clicks and make money off people who don't follow entertainment use. Apple marketed Apple Intelligence with advertisements showing people deceiving their neighbors, family, and coworkers. Activision Blizzard used generative AI to advertise games that don't exist.Now here's the rub: games—and all of entertainment—are also a form of deception. We use the phrase "magic circle" to describe how we attract players into our worlds. We use camera tricks, rendering technology, and even VO barks to simulate digital worlds. People engage with games, film, TV, books, and especially magic shows because on some level they want to be not just deceived, but lied to. AI has also been sold as technology that will let every player make their own perfect experience tailored for them by generating worlds, visual assets, and audio on the fly. But the best pitches I've heard for AI tend to "hide" the presence of the LLM, only mildly asking the player for prompts in order to accomplish behind-the-scenes computing tasks. These lies can make shared realities, not wholly distinct ones.That is the difference between telling lies to make virtual worlds and and telling lies to shape the real one. Lies in virtual worlds create shared realities. Lies in the real world tear them down.How appropriate that one such "shared reality," the Star Wars show Andor, recently warned us about the price we pay with treating AI as "just a tool." "The loss of an objective reality is perhaps the most dangerous," said the character Mon Mothma in a climactic speech decrying the whitewashing of a carefully executed genocide."When truth leaves us, when we let it slip away, when it is ripped from our hands, we become vulnerable to the appetite of whatever monster screams the loudest."Game Developers Conference and Game Developer are sibling organizations under Informa.

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More Alibaba Group has introduced QwenLong-L1, a new framework that enables large language models (LLMs) to reason over extremely long inputs. This development could unlock a new wave of enterprise applications that require models to understand and draw insights from extensive documents such as detailed corporate filings, lengthy financial statements, or complex legal contracts. The challenge of long-form reasoning for AI Recent advances in large reasoning models (LRMs), particularly through reinforcement learning (RL), have significantly improved their problem-solving capabilities. Research shows that when trained with RL fine-tuning, LRMs acquire skills similar to human “slow thinking,” where they develop sophisticated strategies to tackle complex tasks. However, these improvements are primarily seen when models work with relatively short pieces of text, typically around 4,000 tokens. The ability of these models to scale their reasoning to much longer contexts (e.g., 120,000 tokens) remains a major challenge. Such long-form reasoning requires a robust understanding of the entire context and the ability to perform multi-step analysis. “This limitation poses a significant barrier to practical applications requiring interaction with external knowledge, such as deep research, where LRMs must collect and process information from knowledge-intensive environments,” the developers of QwenLong-L1 write in their paper. The researchers formalize these challenges into the concept of “long-context reasoning RL.” Unlike short-context reasoning, which often relies on knowledge already stored within the model, long-context reasoning RL requires models to retrieve and ground relevant information from lengthy inputs accurately. Only then can they generate chains of reasoning based on this incorporated information.  Training models for this through RL is tricky and often results in inefficient learning and unstable optimization processes. Models struggle to converge on good solutions or lose their ability to explore diverse reasoning paths. QwenLong-L1: A multi-stage approach QwenLong-L1 is a reinforcement learning framework designed to help LRMs transition from proficiency with short texts to robust generalization across long contexts. The framework enhances existing short-context LRMs through a carefully structured, multi-stage process: Warm-up Supervised Fine-Tuning (SFT): The model first undergoes an SFT phase, where it is trained on examples of long-context reasoning. This stage establishes a solid foundation, enabling the model to ground information accurately from long inputs. It helps develop fundamental capabilities in understanding context, generating logical reasoning chains, and extracting answers. Curriculum-Guided Phased RL: At this stage, the model is trained through multiple phases, with the target length of the input documents gradually increasing. This systematic, step-by-step approach helps the model stably adapt its reasoning strategies from shorter to progressively longer contexts. It avoids the instability often seen when models are abruptly trained on very long texts. Difficulty-Aware Retrospective Sampling: The final training stage incorporates challenging examples from the preceding training phases, ensuring the model continues to learn from the hardest problems. This prioritizes difficult instances and encourages the model to explore more diverse and complex reasoning paths. QwenLong-L1 process Source: arXiv Beyond this structured training, QwenLong-L1 also uses a distinct reward system. While training for short-context reasoning tasks often relies on strict rule-based rewards (e.g., a correct answer in a math problem), QwenLong-L1 employs a hybrid reward mechanism. This combines rule-based verification, which ensures precision by checking for strict adherence to correctness criteria, with an “LLM-as-a-judge.” This judge model compares the semanticity of the generated answer with the ground truth, allowing for more flexibility and better handling of the diverse ways correct answers can be expressed when dealing with long, nuanced documents. Putting QwenLong-L1 to the test The Alibaba team evaluated QwenLong-L1 using document question-answering (DocQA) as the primary task. This scenario is highly relevant to enterprise needs, where AI must understand dense documents to answer complex questions.  Experimental results across seven long-context DocQA benchmarks showed QwenLong-L1’s capabilities. Notably, the QWENLONG-L1-32B model (based on DeepSeek-R1-Distill-Qwen-32B) achieved performance comparable to Anthropic’s Claude-3.7 Sonnet Thinking, and outperformed models like OpenAI’s o3-mini and Qwen3-235B-A22B. The smaller QWENLONG-L1-14B model also outperformed Google’s Gemini 2.0 Flash Thinking and Qwen3-32B.  Source: arXiv An important finding relevant to real-world applications is how RL training results in the model developing specialized long-context reasoning behaviors. The paper notes that models trained with QwenLong-L1 become better at “grounding” (linking answers to specific parts of a document), “subgoal setting” (breaking down complex questions), “backtracking” (recognizing and correcting their own mistakes mid-reasoning), and “verification” (double-checking their answers). For instance, while a base model might get sidetracked by irrelevant details in a financial document or get stuck in a loop of over-analyzing unrelated information, the QwenLong-L1 trained model demonstrated an ability to engage in effective self-reflection. It could successfully filter out these distractor details, backtrack from incorrect paths, and arrive at the correct answer. Techniques like QwenLong-L1 could significantly expand the utility of AI in the enterprise. Potential applications include legal tech (analyzing thousands of pages of legal documents), finance (deep research on annual reports and financial filings for risk assessment or investment opportunities) and customer service (analyzing long customer interaction histories to provide more informed support). The researchers have released the code for the QwenLong-L1 recipe and the weights for the trained models. Daily insights on business use cases with VB Daily If you want to impress your boss, VB Daily has you covered. We give you the inside scoop on what companies are doing with generative AI, from regulatory shifts to practical deployments, so you can share insights for maximum ROI. Read our Privacy Policy Thanks for subscribing. Check out more VB newsletters here. An error occured.

Last Surviving Grandson of President John Tyler, Who Took Office in 1841, Dies at 96 When Harrison Ruffin Tyler’s grandfather was born 235 years ago in 1790, George Washington had just become the nation’s first president John Tyler was 63 when his 13th child was born in 1853. That child, Lyon Gardiner Tyler Sr., was 75 when Harrison Ruffin Tyler was born in 1928. Heritage Art / Heritage Images via Getty Images Harrison Ruffin Tyler, grandson of the tenth American president John Tyler, died on May 25 at age 96. Though the cause of death has not been revealed, his health had deteriorated in recent years. He had been diagnosed with dementia and suffered several small strokes starting in 2012, reports the New York Times’ Robert D. McFadden. He died at his home in a retirement community in Richmond, Virginia, according to the Washington Post’s Andrew Jeong and Brian Murphy. After the death of his brother, Lyon Gardiner Tyler Jr., in September 2020, Harrison Ruffin Tyler was the last surviving grandson of John Tyler, who was born in 1790 and led the nation between 1841 and 1845. But how could someone born in 1790 still have—until very recently—living grandchildren? Even the president’s grandson acknowledged that the time frame was difficult to comprehend. “When you talk about my grandfather born in the 1700s, there is a disconnect there,” he told WTVR’s Scott Wise and Greg McQuade in 2012. The unusual timeline was the result of second marriages and late-in-life fatherhood for the former president and, later, one of his sons. John Tyler was 63 when his 13th child, Lyon Gardiner Tyler Sr., was born in 1853. Then, Lyon Gardiner Tyler Sr. was 75 when Harrison Ruffin Tyler was born in 1928. “Both my grandfather—the president—and my father were married twice,” he told New York magazine’s Dan Amira in 2012. “And they had children by their first wives. And their first wives died, and they married again and had more children.” With so many relatives to keep track of, he added, “it does get very confusing.” “When I was a child, I did know most of the descendants, but as you get more generations down the line, it’s hard to keep track of everybody,” he said.John Tyler was born just after George Washington became the fledgling nation’s first president. He pursued a career in politics, serving as Virginia’s governor, as well as a United States representative and senator. He became America’s vice president when William Henry Harrison was elected president in 1840. When Harrison died of pneumonia a month into his term, John Tyler became the first vice president to succeed a president who died in office. His ascension was controversial, with some federal lawmakers questioning the legitimacy of his claims to the presidency. Some detractors even took to calling him “His Accidency.” The issue was not officially settled until 1967, with the ratification of the 25th Amendment. As president, one of John Tyler’s biggest accomplishments was pursuing the annexation of Texas, which officially joined the Union in 1845 under President James K. Polk. After his stint in the White House, he retired to his plantation on Virginia’s James River. During the Civil War, he was elected to the Confederate legislature, but he died in 1862 before he could take office. During his lifetime, he had a record-setting 15 children—the most of any U.S. president. He was married twice: first to Letitia Christian, who became the first president’s wife to die in the White House in 1842, followed by Julia Gardiner, who also served as First Lady. Their fifth child was Lyon Gardiner Tyler Sr., who later fathered Harrison Ruffin Tyler. Born in 1928, Harrison Ruffin Tyler studied chemistry at the College of William & Mary and chemical engineering at Virginia Tech before founding a water treatment company called ChemTreat. He and his wife, Frances Payne Bouknight, who died in 2019, also spent many years restoring the family’s ancestral home, Sherwood Forest Plantation. The 1,600-acre property, built around 1730 and purchased by his grandfather during his presidency, is a National Historic Landmark. The couple also worked to preserve a nearby Union supply depot called Fort Pocahontas that had been constructed by a regiment of Black soldiers during the Civil War. They had three children together: Harrison Ruffin Tyler Jr., William Bouknight Tyler and Julia Gardiner Tyler Samaniego. Harrison Ruffin Tyler “will be missed immeasurably by those who survive him,” Annique Dunning, the executive director of Sherwood Forest, said in a statement, as reported by the Associated Press. “He will be remembered for his considerable charm, generosity and unfailing good humor by all who knew him.” Get the latest stories in your inbox every weekday.

This Shield concept is by PDR, an industrial design consultancy based in Cardiff, Wales. It aims to bring dignity to those who are required to live with catheters. Shield is a simple and low-cost redesign that utilises recycled single-use disposable plastic sterilisation tray wraps and drapes, commonly used in hospital operating theatres. Long term catheter use is a standard treatment that affects millions of people across Europe each year. Urine is collected externally in external drainage bag, typically held in a simple wireframe. These bags and frames cause many difficulties in handling and emptying and cause particular distress and stigmatisation through the visibility of up to 1.5L of urine and a very obvious hospital item at home.Overnight urine bag stands have remained largely unchanged for decades. It is often, though not exclusively older people, receiving care at home who are on a long-term catheter plan and use overnight bag stands. Embarrassment at both the display of bag contents and the practical challenges of removing the bag from the wireframe and emptying safely in the toilet is very real. Added to this are the very real issues of sustainability and waste, an issue of increasing awareness at hospitals around the world. There is little economic incentive for manufacturers to address this problem and hospitals have few options in addressing the problems described.Shield is a simple and low-cost redesign that utilises recycled single-use disposable plastic sterilisation tray wraps and drapes, commonly used in hospital operating theatre's. The design replaces a design that has not changed in 50 years, it provides a firm stable platform, an easy to empty and easy to handle drainage bag, and discreetly hides the contents from view. Shield is a simple product that provides stability and security to a 2L drainage bag whilst simultaneously allowing for more discreet use in a domestic setting. Use of recycled hospital waste material, specifically single-use disposable sterilisation tray wraps and drapes used in operating theatre's that would otherwise be incinerated are recycled to produces a sterile material in sheet and granular form with a unique surface finish that is suited to use in a more domestic environment. This ultra low-cost materials is processed to produce Shield in finishes that vary, creating unique finishes which are dependent on the colour and mix of the recycled polypropylene material. The product is stable in use and through its simple clean form unobtrusively sits next to a bed or chair at home. The bag with easy-grip, secure handle is designed for use by weakened and arthritic hands that may need to carry up to 2kg of fluid and lifts straight out to be easily carried to the bathroom before release with an easy turn large surface drainage tap. This approach avoids the current system of removing from wire lugs and attempting to carry, often in two hands a full bag that can leak and is difficult to empty without spilling.Shield provides a relatively simple solution to a patient-centred problem that has existed for over 50 years. The simple, clean snap together construction allows easy assembly by unskilled labour and its use of a recycled hospital waste material provided free by hospital trusts when approached from a circular economy perspective creates a highly functional material at very little additional energy consumption and cost. You can see more of PDR's work here.

Hiring managers aren’t convinced that master’s degree holders perform better than candidates with two years of work experience, but they are still willing to pay them more. That’s according to a recent survey of 1,000 U.S. hiring managers conducted by Resume Genius. Fifty-two percent of respondents said the performance of those who earn the credential is the same as those with a bachelor’s degree plus two years of work experience. Another 10% believe it’s worse. “A master’s degree simply isn’t experience; it’s knowledge,” explains Resume Genius career expert Nathan Soto. “So much practical know-how can only be learned by doing the job, and higher education doesn’t prepare people for work. It prepares them for even higher levels of academic study.” The survey also suggests a significant discrepancy in how different generations view the degree, with more than double the proportion of Gen Z hiring managers—29% in total—suggesting it leads to stronger performance, compared with just 13% of Boomers. “As Baby Boomers age out of the workforce, it suggests that the proportion of hiring managers who value master’s degrees is growing,” Soto says. Despite the broad skepticism over its value, 72% of hiring managers say they offer master’s degree holders higher salaries. In fact, one in four say they offer 20% more to candidates with the degree. “This may be reason enough to get one,” Soto says.  Wage premiums aren’t keeping up with tuition At the same time, Soto points out that the cost of higher education has more than doubled in the last 20 years. Coupled with a one- or two-year delay entering the workforce, he warns that getting a return on that investment is far from certain. “If you can afford to complete a master’s degree without incurring a crushing amount of debt, then there are real benefits,” he says. “However, people in fields without strict master’s degree requirements would be better off entering their chosen profession and then deciding whether or not a master’s degree is essential to their own professional advancement.” According to data from the Economic Policy Institute, those with a bachelor’s degree earn roughly 20% more in hourly wages than those without, and master’s degree holders earn an additional 20% on average. “Real hourly wages have grown both for workers with just a bachelor’s degree and a master’s degree by about the same amount—about 35% over the last 33 years,” says Economic Policy Institute senior economist Elise Gould, explaining that her data set goes back to 1992. “The differential between a master’s and a bachelor’s degree has been pretty constant as well. Workers with a master’s degree are paid about 20% more than a worker with only a bachelor’s degree.” Gould adds that during that time the cost of obtaining a degree has skyrocketed, but the relative earning potential has remained relatively unchanged. “The data are about averages. I think it is really about individuals, their resources, their time, what they’re going to study and their objectives,” she says. “All those factors should be taken into account.” Not all master’s degrees are made the same On average, a slight majority of master’s degrees pay for themselves over time, however, positive returns are far less likely compared to other degree types, and depend more heavily on factors like field of study, institution, and location. According to a 2024 study by the Foundation for Research on Equal Opportunity, or FREOPP, 57% of master’s degree earners experience a positive return on their investment, compared with 77% of bachelor’s, doctoral, and professional degree earners. “That’s actually overstating the performance of master’s degrees, because the positive ROI is heavily concentrated in nursing,” explains FREOPP founder Avik Roy. “The average master’s degree outside of nursing has a negative ROI.” According to the FREOPP study—which compares College Scorecard data against wage, occupation, and geography data published by the Census Bureau in the American Community Survey—there is a significant discrepancy in return potential between master’s programs.    Nearly 40% of MBA programs, for example, have negative returns, on average. However, there is a significantly higher chance of a positive ROI for those who attend a top-10 ranked business school. Technical programs, like engineering and computer science, were also more likely to offer a return on their investment than not. Humanities occupied the most spots on the other end of the spectrum, with film school standing out for offering the lowest median returns. “If you’re pursuing a master’s degree in nursing, computer science, or engineering, the likely return on that is very positive,” Roy says. “If you’re outside of those fields of study, buyer beware.”

May 31: Dragon Boat Festival in China and Taiwan (2025); World No Tobacco Day Bessarion 455 – Petronius Maximus, the ruler of the Western Roman Empire, was stoned to death by a mob as he fled Rome ahead of the arrival of a Vandal force that sacked the city. 1223 – Mongol invasion of Kievan Rus': Mongol forces defeated a Kievan Rus' army at the Battle of the Kalka River in present-day Ukraine. 1468 – Cardinal Bessarion (pictured) announced his donation of 746 Greek and Latin codices to the Republic of Venice, forming the Biblioteca Marciana. 1935 – A magnitude-7.7 earthquake struck Balochistan in British India, now part of Pakistan, killing between 30,000 and 60,000 people. 2013 – A tornado struck Central Oklahoma, killing eight people and injuring more than 150 others. Albertino Mussato (d. 1329)Joseph Grimaldi (d. 1837)Dina Boluarte (b. 1962)Mbaye Diagne (d. 1994) More anniversaries: May 30 May 31 June 1 Archive By email List of days of the year About