Menu Home News Hardware Gaming Mobile Finance Deals Reviews How To Wccftech Mobile Apple Announced Its AirTags Four Years Ago Today, And While These Trackers Are A Helpful Tool In Locating Missing Items, Their Launch Has Deterred Crime, To An Extent Omar Sohail • Apr 20, 2025 at 11:50am EDT The extensive Find My network developed by Apple alongside the AirTags has given the masses enough reason to grab several of these trackers to keep an eye on their belongings, and to monitor their more expensive possessions, such as cars, which are more susceptible to theft. In the four years that the company announced these circular-shaped devices, the AirTags have been a godsend for thousands, while also stirring up some controversy from those who believe these are tools that will fuel the flames for stalkers, while also putting a major dent in competitors’ business. Overall, AirTags has managed to provide the means to users to curb crime, just as long as they seek a proper law enforcement channel, but stalking concerns will remain active During Apple’s preview of the AirTags, the obvious issue that others would have are the innumerable stalking attempts, as with these trackers, almost anyone or anything could be located. At $29 apiece, these would barely pinch the finances, but Apple has made every effort it can to allay these nefarious activities. For instance, a device running iOS or iPadOS will alert the owner if someone else’s AirTag is with them, with the feature also ported to Android smartphones. Additionally, a beeping sound will begin playing, warning others of its presence. Unfortunately, there are times when people who voice their concerns against AirTags are correct in their assessment, as there have been instances where multiple individuals felt that their safety had been compromised, such as when Australian tourists found the trackers in their luggage, forcing them to alter their holiday plans due to stalking fears. In a related incident, a woman informed the authorities that a man was pursuing her, claiming that he was her husband. Luckily, the person kept receiving alerts on her device from the nearby AirTag to prepare her for the impending danger. A class-action lawsuit was also filed against Apple’s AirTags, claiming that these tools were an ideal addition for stalkers, but we would like to assume that where technology exists to provide genuine help to people, there are always those who will continue to misuse it. Whatever potential threats these AirTags can conjure up, one can agree that these devices have contributed more towards keeping crime at bay as much as possible. In fact, these trackers have become so effective in preventing car thefts that Colorado police were giving them out to citizens for free, as it would make life significantly easier for both parties. There was also an occasion where a carpenter successfully recovered machinery worth $5 million that had been stolen, with a family vehicle also located by the police as a result of having AirTags hidden in it. Unfortunately, not all incidents related to crime have had a pleasant outcome, but that is the result of those who pursue thieves without informing law enforcement. Such was the case when a 61-year-old tragically met her demise while tracking down criminals using the AirTag. Then again, at $79.99 for a 4-pack on Amazon, should you ever have bad luck befall you, such as the ones mentioned above, you can rest assured that AirTags will become a barrier between a successful crime and a failed one. Subscribe to get an everyday digest of the latest technology news in your inbox Follow us on Topics Sections Company Some posts on wccftech.com may contain affiliate links. We are a participant in the Amazon Services LLC Associates Program, an affiliate advertising program designed to provide a means for sites to earn advertising fees by advertising and linking to amazon.com © 2025 WCCF TECH INC. 700 - 401 West Georgia Street, Vancouver, BC, Canada

The ubiquity of anime and manga in this era is indisputable. As more and more creators who had their childhoods in the late 90s and early 2000s find themselves creating in the same spheres they once worshiped as kids, like animation or comics, we're starting to see some very interesting works of art that are unique in their clear inspiration by some of the most iconic pieces of entertainment media the world has ever seen.

After 10 episodes, HBO’s The Last of Us has finally gotten to the issue being talked around since the show first dropped: What happens between Joel (Pedro Pascal) and Abby (Kaitlyn Dever). The scene is very close to what plays out in the game — certainly too close for those who weren’t looking forward to watching it happen again. But the show version makes a crucial change to the setup of the scene, something that makes Abby’s actions feel all the more galling, both for this episode and for the future of the show.  [Ed. note: This post contains spoilers for season 2 episode 2 of The Last of Us, as well as discussion of the corresponding events in the game, The Last of Us Part 2. That’s vague, but if none of this so far has rung a bell and you want to keep it that way, this is your warning!] Whether you know the story of The Last of Us Part 2 or not, it’s not hard to see that Abby’s numbers — and strategizing to get Joel back to the cabin — mean Joel’s in trouble. She’s already said she wants him dead and she wants to take her time with it, and once she puts together who he is, the dominoes seem set.  But The Last of Us season 2 makes a smaller change earlier on that sets off this chain reaction of bad — literally, considering it all starts when Abby sets off an avalanche. The resulting horde leads to two major things happening in episode 2: Jackson under siege from undead, and Abby torturing and killing Joel. And by making it clearer that Abby is the one setting off this chain reaction, the show deepens the conflict between her and Joel.  As in the game, the two meet after Joel rescues Abby when she’s almost certainly lost to the zombie horde. But in the show, the audience is clued into what Joel means to Abby, or, more specifically, what Abby means to Joel: a bloodthirsty vendetta. The irony of all it is so much clearer, which is exactly what co-showrunners Craig Mazin and Neil Druckmann were hoping for when they set up Abby’s background much earlier in the show than the second game does.  “Because you’re not experiencing the story through gameplay, there is not only an opportunity, but almost a requirement now to adapt it somewhat differently,” Mazin told Polygon. “When you meet [Abby] you don’t know why she’s doing what she’s doing, but you are her. And because you’re moving her around, because you’re trying to keep her safe, because you’re trying to keep her alive, you have a connection with her that matters.” With that connection missing in the show, Abby (and her role in the story) has to feel a lot different. And the result so far has been that she feels like more of a tragic threat compared to the game’s frenetic fury — to both Joel and even the future of those around him.  You could look at this as simply plugging a plot hole in the narrative; Abby is introduced earlier, so there are stakes that the audience is aware of when she’s running for her life outside Jackson. When she finds herself on the end of a zombie avalanche, it’s not just because the game needs to drop us into the middle of a fraught situation, but rather because she caused it.  But episode 2 takes the weight of her actions further. Her bloodthirstiness causes the avalanche, and unleashes the zombie horde not just on herself but on Jackson, which is forced to batten down the hatches. It’s a small change, but an important one — Abby’s attack of Joel in the games nestles itself as simply being a frantic aside in between gameplay as Ellie, but in the show it’s counterbalanced by a largely peaceful community coming under threat. The brutality of Abby’s violence and the zombies breaking through White Walker-style are directly paralleled here. The literal snowball effect of her choices leads to hundreds of people she doesn’t know getting attacked. And the result seems to clearly communicate the stakes of even direct confrontation: No matter how focused the revenge seems to be, the implications of it ripple out until they weaken the very future of the society around you.  In a story this dark and resolutely committed to exploring violence as an expression of love, that perspective is important. The Last of Us as a franchise can be largely single-minded as it trudges toward bleakness. Joel’s death is certainly a big part of that, and a bigger part in the coldness to come. But with these early changes, The Last of Us is also getting laced with a potent reminder that revenge comes at a cost — even if it goes far beyond anything Joel, Abby, or even Ellie feel is at risk. 

Spoilers follow for The Last of Us season two, episodes one and two HBO’s The Last of Us showed viewers in season one that it would lean heavily on the source video games for major plot points and general direction of the season while expanding on the universe, and season two has followed that to the most extreme end possible. Episode two sees Tommy and Maria lead the town of Jackson Hole against a massive wave of Infected, the likes of which we haven’t seen in the show (or video games) yet. This was a complete invention for the show, one that gives the episode Game of Thrones vibes, or calls to mind a battle like the siege of Helm’s Deep in Lord of the Rings: The Two Towers. It’s epic in scale, with the overmatched defenders showing their skill and bravery against overwhelming odds; there is loss and pain but the good guys eventually triumph. That mass-scale battle is paired with the most intimate and brutal violence we’ve seen in the entire series so far, as Joel’s actions finally catch up with him. His death at the hands of Abby and her crew of ex-Fireflies from Seattle is truly upsetting, even if you know it is coming — and that in the twisted world that Joel and Abby inhabit, it was a form of “justice.” Of course, it was pretty obvious after the first five minutes of episode one that this is where Abby’s journey would take her, but it sure happened fast. You could say the same thing about when Joel dies in the video game, but ever since the HBO series was announced I’ve been wondering about just how Craig Mazin would choose to tackle the events of the second game and Joel’s death. Part of me wondered if we wouldn’t know for sure Joel was dead until near the end of the season, or if we might not find out Abby’s motivations until much later. But after two episodes, we know that Abby’s dad was the doctor Joel killed at the end of season one to save Ellie, a fact that makes her disturbing lust for vengeance more understandable to the television audience. In the game, on the other hand, players didn’t find that out until much later — a move that cast Abby in the role of "villain" that made Ellie’s own desire for revenge more relatable for the player Photograph by Liane Hentscher/HBO But what works in a game doesn't necessarily work in a more passive adaptation. I do think there was a way for the writers to save the full reveal for Abby’s motivations for later, but I didn’t really bump on how things played out here. Similarly, I wasn’t really bothered by the introduction to Abby’s crew and her desire to find Joel that happened at the beginning of the season premiere. But I do know a number of people who played the game feel like this has robbed the show of some mystery and changed the momentum in a way that didn’t work for them, and I can appreciate that argument. For anyone who hasn’t played the games, though, I’m sure that seeing the co-lead character killed off in the second episode of the season was a bit of a shock, no matter how much it was hinted at — particularly after Joel went out of his way to save Abby from certain death. In another world, you could imagine that blunting her need for revenge, or at least having her forgo the extended torture of a helpless Joel. But she’s clearly captivated by the opportunity in a way that even her friends are a bit horrified by. Ellie being there to find Dina unconscious and then forced to watch Abby deliver the final blow is the cherry on the disaster sundae that is this episode. It’s even harder to swallow when you consider their last interaction was Ellie’s extreme rebuff of Joel at the dance the night before. Photograph by Liane Hentscher/HBO It’s truly a wild choice for such a high-profile show, even though there was nowhere else for it to go. Pedro Pascal has been everywhere the last few years, and The Last of Us could have been his signature role. Joel’s death doesn’t fully diminish that, but obviously even if we see him in flashbacks in this season and the next, it won’t be the same as him leading an episode. To let such a marketable and popular star go must have been a tough pill for HBO to swallow, though chances are that the network knew from the beginning of negotiations that would have to happen. Regardless, it’s a move up there with the death of Sean Bean’s Ned Stark in terms of subverting audience expectations about who is going to lead the show going forward. That’ll fall to Bella Ramsey for the rest of the season. I will say that Mazin and director Mark Mylod did a great job of balancing the many stories and cast members in this episode. Kaitlyn Dever gets her first extended amount of screen time and captures Abby’s barely-contained ferocity in a way that feels like a great tribute to the game source material, despite the fact that she is not nearly as physically ripped as Abby is in the game. Her scenes with Joel and Dina are absolutely chilling, even before she has Joel on the ropes; big credit to Pascal and Isabela Merced, again, for showing a major range of emotions as they help Abby escape from the Infected only to be betrayed. Kaitlyn Dever in HBO's The Last of UsPhotograph by Liane Hentscher/HBO Ramsey, meanwhile, shares a lot of time with Young Mazino’s Jesse as they go on patrol and frantically search for Joel; back in Jackson, Gabriella Luna and Rutina Wesley as Tommy and Maria keep Jackson from falling to the Infected. It’s a lot of threads to tie together, and the fact that this episode includes the biggest battle we’ve ever seen as well as Joel’s capture and death without feeling overstuffed is a testament to the writers, directors and cast. Speaking of that Infected battle, I wasn’t fully sold on it when I saw teases in the trailers for this season, but it was a thrilling delight. I loved seeing the preparations Jackson put into place, like the oil barrels coming down from the walls and the flamethrower brigade surrounded by snipers up on the buildings. Of course, all the best-planned defenses fail once the Infected get through the wall and all hell breaks loose, and the show successfully had me questioning whether the town would make it even though the idea of it fully falling seemed like a deviation too far from the game. It definitely felt like a response to the criticisms of a relative lack of Infected action in season one, but it worked, and the juxtaposition of the disaster befalling Jackson at large as well as the smaller-scale loss of one of its most important citizens added another layer to the grief. Infected Score: 11/10 The showrunners say season two will have more Infected than season one — let’s see if they’re sticking to their word. Tommy taking down a Bloater with a flamethrower in front of a Mailboxes Etc. store? Barrels of oil lighting a field of Infected on fire? The extremely creepy reveal of the “live” Infected buried and laying in wait under a field of frozen ones that emerged and chased Abby right into the path of Joel? Jackson’s guard dogs coming to the rescue to keep the horde from overtaking the town? This was the most epic action we’ve seen out of the show yet, on a scale that we didn’t get in the games. It’ll be hard to top this battle.This article originally appeared on Engadget at https://www.engadget.com/entertainment/tv-movies/the-last-of-us-season-two-through-the-valley-recap-well-that-happened-020017596.html?src=rss

Looking for Quordle clues? We can help. Plus get the answers to Quordle today and past solutions.

First impressions matter—they shape how we’re judged in mere seconds, research shows. People are quick to evaluate others’ competence, likability and honesty, often relying on superficial cues such as appearance or handshake strength. While these snap judgments can be flawed, they often have a lasting impact. In employment, first impressions not only affect hiring choices but also decisions about promotion years later. As a researcher in cognitive science, I’ve seen firsthand how first impressions can pose a challenge for individuals with autism spectrum disorder, or ASD. People with ASD often display social behaviors—such as facial expressions, eye contact, gestures, and sense of personal space—that can differ from those of neurotypical individuals. These differences are often misunderstood, leading people with ASD to be perceived as awkward, odd, or even deceptive. People form these negative impressions in just seconds and report being reluctant to talk to, hang out with, or even live near people on the spectrum. It’s not surprising, then, that unfavorable first impressions create barriers for people with ASD in the workplace. The interview trap It starts with the job interview. Whether you’re seeking a position as a computer programmer at a tech firm or a dog groomer at a vet clinic, the job interview is a critical gateway. Success depends on your ability to think on your feet, communicate your qualifications, and present yourself as likable, agreeable, and collegial. My research demonstrates that job seekers with ASD often perform poorly in interviews due to the social demands of the situation. This is true even when the candidate is highly qualified for the job they are seeking. In one study, my colleagues and I videotaped mock job interviews with 30 young adults (half with ASD, half neurotypical) who were all college students without an intellectual disability. We asked them to discuss their dream jobs and qualifications for five minutes. Afterward, evaluators rated them on social traits, such as likability, enthusiasm, and competence, and indicated how likely they were to hire each interviewee. As in most professional interviews, the evaluators weren’t aware that some candidates were on the autism spectrum. Candidates with autism spectrum disorder were consistently rated less favorably on all social dimensions compared with people without the condition, and those unfavorable social ratings weighed heavily on hiring decisions. Even though candidates with ASD were rated as equally qualified as neurotypical candidates, they were significantly less likely to be hired. Interestingly, when evaluators only read the candidates’ interview transcripts without watching the interviews, ratings for ASD candidates were the same as, or even better than, those for neurotypical candidates. This suggests that it’s not just what candidates say in an interview but how they present themselves socially that affects hiring decisions. This is especially problematic for jobs that require minimal social interaction (think data analyst or landscaper), where a candidate’s qualifications should be the main consideration. By relying on interviews as a primary screening tool, employers may miss out on competent, qualified applicants with unique strengths. Rethinking what makes a good candidate Scientists have explored whether it’s possible to teach adults with ASD how to improve their interview skills, for example by maintaining more eye contact or standing at a socially acceptable distance from an interviewer. While such training can help, it addresses only a small part of the problem, and I think this approach may not significantly improve employment outcomes for autistic adults. For one, it reduces the challenges faced by adults with ASD to a limited set of behaviors. ASD is a complex condition, and research shows that the negative evaluations of individuals with ASD are not driven by a single difference or a collection of specific differences, but rather by the individual’s overall presentation. In addition, this type of training often encourages individuals to mask their autistic traits, which could make a stressful interview even more difficult. Finally, if ASD candidates successfully mask their autism during the interview but can’t maintain that mask once they are hired, their longevity in the position could be at risk. A more effective approach may be to change how interviews are conducted and how candidates are perceived. This includes giving employers meaningful education about autism and giving job applicants a way to disclose their diagnosis without penalty. Research shows that when people know more about autism spectrum disorder, they have more positive views of people with ASD. In addition, ratings of people with ASD are often more favorable when evaluators know about their diagnosis. Combining these two approaches—that is, pairing ASD education for employers with diagnostic disclosure for candidates—may lead to better outcomes. My colleagues and I explored this possibility in a series of studies. Again, we showed raters the mock job interviews of candidates with and without ASD. This time, however, some evaluators watched a brief educational video about autism, learning about characteristics and strengths often associated with ASD before evaluating the mock interviews. In addition, these raters knew which candidates had an ASD diagnosis. Even though raters still perceived the candidates with ASD as more awkward and less likable, they rated those candidates as equally qualified as neurotypical candidates and were just as likely to hire them. This boost in hiring ratings persisted even when the educational video about autism was viewed months before candidates were evaluated. Notably, neither of these interventions was effective on its own. In different conditions, some evaluators simply got the training but didn’t receive diagnostic information about candidates; others received no education about autism but were aware of which candidates had ASD. Both groups continued to select against candidates with ASD in hiring decisions, even though the candidates with ASD were rated as highly qualified. It appears that both knowing a person has autism and understanding more about autism are important for overcoming negative first impressions. We believe that our training fostered a greater understanding of the atypical interactive style and behaviors that can be common among adults with ASD. This understanding, when coupled with the knowledge of a candidate’s diagnosis, may have helped evaluators contextualize those behaviors and, in turn, place more emphasis on qualifications when making hiring decisions. When hiring decisions are based on merit, both employees and employers benefit. First impressions, though impactful, can be deceptive and often bias decisions, particularly for individuals with ASD. Our findings highlight an important truth: Understanding autism enables employers to focus on qualifications, giving candidates with ASD a fair opportunity to succeed based on their true potential. Cindi May is a professor of psychology at the College of Charleston. This article is republished from The Conversation under a Creative Commons license. Read the original article.

Back in the day, handheld gaming meant slipping AA batteries into a Game Boy and hoping your save file didn’t vanish with a sneeze. Fast forward past the PSP era, through the Switch’s heyday, and here we are – surrounded by sleek, cloud-connected consoles that look like they could file your taxes if you asked nicely. Into this already-packed arena walks the Abxylute One Pro, strutting like it knows something the others don’t. Spoiler: it kind of does. The One Pro isn’t trying to outgun a Steam Deck or push frame rates like a ROG Ally. That’s not its lane. What it offers instead is something far more focused – and frankly, way more fun. This is a machine that understands handheld gaming isn’t always about performance brute force. Sometimes it’s about slipping into retro bliss, emulating PS2 titles flawlessly, and doing it all without melting your palms or draining your wallet. Designer: abxylute Click Here to Buy Now: $209 $249 ($40 off) Hurry! Only 10 days left Powered by the MediaTek Genio 510, the One Pro gets a solid boost over the OG Abxylute. This chip isn’t exactly a household name, but under the hood it’s no slouch. We’re talking dual Cortex-A78 cores backed by four Cortex-A55s – enough to push 60fps on PS2 emulation, which is kind of the sweet spot for most old-school titles. Paired with 4GB of LPDDR4 RAM and 64GB of storage (expandable to 2TB), it’s well equipped for managing retro libraries, mobile games, and light Android multitasking. And yes, it’s running Android 13, which means Google Play Store access is baked in. Emulators? Check. Remote play for Steam, PS5, and Xbox? You bet. Mobile games and streaming? Abxylutely (geddit?). But what makes the One Pro feel unusually tuned-in is how it’s tailored to the hands of gamers who actually care about control fidelity. The joysticks are a big deal here. These aren’t the mini thumb-nubs you find on cheaper handhelds or Joy-Con imitators. They’re full-sized capacitive digital joysticks, which should mean better accuracy, less drift, and a bit of that Xbox-controller muscle memory. For something this compact, that’s a surprisingly premium touch – and one that’ll matter when you’re trying to stick a landing in Tony Hawk or line up a sniper shot in Metal Gear Solid 3. Display-wise, you’re getting a 7-inch 1080p LTPS screen with a 60Hz refresh rate. It’s not OLED, but for the price point it’s respectable – colors are punchy, motion feels smooth, and the screen size hits that ergonomic sweet spot. The design goes semi-transparent, evoking that late ’90s aesthetic like it just walked out of a Toys“R”Us catalog – but the vibe works. It’s retro without being cosplay. You’ll also find touchscreen mapping tools, controller customization, and even 4K video output – ideal if you’re hooking this up to a TV or some wild AR glasses setup. Connectivity includes Wi-Fi 5 with MU-MIMO (so your game stream won’t choke), Bluetooth, and USB-C for charging or docking. The 5,200mAh battery promises up to 8 hours of use, which feels just right – not marathon-level, but good enough to get you through a long flight or a Sunday couch session. Size-wise, it clocks in at 250mm x 115mm x 30mm and weighs 430 grams. That’s chunky enough to feel sturdy without veering into wrist-cramp territory. Luckily, it’s still hyper-compact, leaning into Switch Lite or Legion Go S territory, which makes it perfect for on-the-go gaming… and although the abxylute One Pro doesn’t really ship with a protective case, you can grab a nifty soft-shell carry case on abxylute’s website for pretty cheap, complete with mesh pockets for keeping things like charging cables, MicroSD cards, and whatnot. The Kickstarter launched on March 27, with early bird pricing at $209, and the full retail price landing at $249. Shipping kicks off mid-to-late May, so it won’t be long before you get the unit in your hands – and the sooner the better, judging by the whole tariff-measuring competition the US and China are having… Click Here to Buy Now: $209 $249 ($40 off) Hurry! Only 10 days leftThe post The Abxylute One Pro is what the Steam Deck should have been – with a $209 price tag first appeared on Yanko Design.

Large language models (LLMs) have gained prominence for their ability to handle complex reasoning tasks, transforming applications from chatbots to code-generation tools. These models are known to benefit significantly from scaling their computation during inference, often producing higher accuracy by dedicating more resources to hard problems. However, this approach brings along considerable drawbacks. Longer processing times and higher computing costs make it challenging to scale such solutions in real-world settings, where responsiveness and affordability are crucial. As technology advances toward more intelligent systems, there is a growing need to explore how LLMs can become not only smarter but also more efficient, especially when operating within repetitive or familiar contexts. One of the biggest inefficiencies in current LLM deployment occurs during query resolution. Typically, when a user poses a question, the model processes it simultaneously with the necessary background context. This test-time compute assumes that the context and question always arrive together. But in real scenarios, such as document Q&A or debugging code, context is usually persistent and can be accessed well before a specific question is asked. Yet, the model processes everything from scratch for each query, even if it has seen the context before. This redundancy results in increased computational costs and response delays, particularly in scenarios involving multiple queries within a single context. To deal with this inefficiency, various methods have been developed. Sequential and parallel test-time computation are two major strategies. Sequential approaches extend the model’s reasoning path, allowing it to consider more possibilities, while parallel approaches involve sampling multiple outputs simultaneously, known as pass@k. Techniques like speculative decoding aim to cut latency by making early guesses, but their usefulness is limited when the model still has to think from scratch. While helpful, these methods don’t eliminate the need to process context alongside every new question repeatedly. They also typically require test-time conditions that aren’t always feasible, such as access to an oracle or an ideal verifier. Researchers from Letta and the University of California, Berkeley, introduced a novel solution they call sleep-time compute. The method involves utilizing idle time between user interactions to increase productivity. Instead of waiting for a user question, the model begins analyzing the context beforehand. It anticipates possible future queries and builds a new version of the context enriched with relevant inferences. When a user finally asks a question, the model can simply refer to this pre-processed context. Since much of the thinking is already done, it requires less computational effort to produce accurate answers. This approach becomes even more effective when multiple questions relate to the same context, allowing for shared inferences and distributed computational cost. The implementation of sleep-time compute relies on decomposing the traditional prompt into two parts: a static context and a dynamic query. During the sleep-time window, only the context is used to generate a pre-processed version. This enhanced context, called c′, is built using test-time compute techniques like reasoning chains or summarization. Once this enriched version is stored, it replaces the raw context during real-time queries. The final answers are then generated using much fewer resources. This system not only minimizes redundant reasoning but also paves the way for more proactive LLMs that can think ahead and be better prepared. To evaluate the effectiveness of sleep-time compute, the research team tested it using two specially designed benchmarks: Stateful GSM-Symbolic and Stateful AIME. Both datasets are derived by splitting existing problem sets into separate contexts and questions. In experiments using models like GPT-4o and GPT-4o-mini, researchers observed a 5× reduction in test-time compute for similar accuracy levels. Notably, accuracy improved by up to 13% for the GSM-Symbolic P2 dataset and by 18% on Stateful AIME when sleep-time compute was scaled. Multi-Query GSM-Symbolic, a new dataset introduced for this evaluation, helped demonstrate that the cost per query could be reduced by 2.5× when 10 queries shared the same context. When pitted against popular strategies like pass@k, sleep-time compute consistently outperformed them. Unlike pass@k, which assumes access to a perfect evaluator, sleep-time compute works under more realistic conditions. Results show that even at low test-time compute budgets, sleep-time compute produced comparable or better accuracy while consuming fewer tokens. For instance, the GPT-4o-mini model achieved higher accuracy with fewer than 200 test-time tokens using sleep-time compute compared to over 500 tokens needed in the baseline. Even when models like Claude Sonnet 3.7 and DeepSeek R1 were evaluated, similar improvements were observed. Scaling the amount of compute dedicated to sleep-time further improved outcomes. By running five parallel generations during sleep-time on complex tasks, researchers pushed the pareto curve further. However, they noted diminishing returns beyond this point. Importantly, results showed that stronger models handling more difficult tasks benefited more from additional sleep-time compute. Also, amortizing sleep-time computation became highly cost-effective when contexts served multiple related queries. By weighting test-time tokens as ten times more expensive than sleep-time tokens, aligned with industry latency-cost ratios, the researchers confirmed a reduction of up to 2.5 times in the average cost per query. Another interesting finding was that sleep-time compute worked best when user queries were predictable. Using Llama2-70B, researchers scored the predictability of each query given its context and found a strong correlation: the more predictable the query, the greater the benefit. In examples where the question logically followed from the given context, sleep-time computation yielded higher gains. Conversely, less predictable or abstract queries experienced reduced effectiveness, although they still showed benefits compared to traditional test-time-only methods. Altogether, this research presents a smart and scalable technique to enhance the efficiency of LLMs without compromising accuracy. By leveraging otherwise idle time, sleep-time computing reduces the burden on real-time systems, lowers operational costs, and improves response time. The clear quantitative improvements, such as a 5× reduction in compute, 13–18% accuracy gains, and a drop of up to 2.5× in cost per query, demonstrate that forward-thinking approaches like this could shape the next generation of intelligent, context-aware assistants. Several Key Takeaways from the Research are as follows: Sleep-time compute allows models to anticipate queries by reasoning on context before the query arrives. Accuracy improved by 13% on GSM-Symbolic and 18% on AIME datasets when sleep-time computation was scaled. Test-time compute requirements decreased by approximately 5 times for similar performance levels. When sharing context across 10 related queries, the average query cost decreased by a factor of 2.5. Outperformed the pass@k strategy in parallel compute settings at equivalent budgets. More effective on predictable queries, identified via log-probability scoring. Diminishing returns noted beyond five parallel generations for sleep-time computation. Check out the Paper. Also, don’t forget to follow us on Twitter and join our Telegram Channel and LinkedIn Group. Don’t Forget to join our 90k+ ML SubReddit. [Register Now] miniCON Virtual Conference on AGENTIC AI: FREE REGISTRATION + Certificate of Attendance + 4 Hour Short Event (May 21, 9 am- 1 pm PST) + Hands on Workshop The post LLMs Can Think While Idle: Researchers from Letta and UC Berkeley Introduce ‘Sleep-Time Compute’ to Slash Inference Costs and Boost Accuracy Without Sacrificing Latency appeared first on MarkTechPost.

The following contains full spoilers for The Last of Us Season 2 Episode 2.Let’s face it, this week’s The Last of Us is all about one thing: the death of Joel Miller. As such, it creates certain expectations: Abby killing Joel will leave as big an impact as possible – a crater in Ellie’s life that she can only fill with vengeance – and everything else in the episode will be geared toward making sure that happens. It’s the inciting moment for the rest of the story. The most crucial moment in the series (perhaps, outside of Joel’s fateful choice at the end of season 1). It’s therefore disappointing that for most parts of season 2 of The Last of Us’ second episode, it feels like an afterthought – a sideshow to an explosive, albeit exciting, battle sequence. Neither halves of this story are executed poorly – far from it, with both thrilling visual style and character-led substance – they just don’t fuse together into a satisfying whole.Mark Mylod of Succession fame (and seemingly HBO’s go-to pair of hands when it comes to an episode ticking down to a father’s death) has a lot to juggle when directing this chapter. It’s partially an ode to The Lord of the Rings’ Helms Deep sequence (plus some Hardhome from Game of Thrones), and partially a cold, brutal examination of murder as a means of closure. The latter is the most crucial piece of the jigsaw, and for the first half of the episode, the tension is ratcheted up beautifully, culminating in a wonderfully claustrophobic sequence that sees Abby struggling through the snow under a wire fence buckling under stacks of infected. The pressure generated is suffocating, and a true display of the rabid relentlessness of the infected.PlayIt’s preceded by a truly haunting image of her pursuers emerging from the ground and defrosting themselves like some incredibly angry frozen peas, as the camera pulls up to reveal the sheer scale of their numbers. It’s a clever representation of the layers of evil being exposed to us in this second season. The avalanche of human fungus makes us fear for Abby, a character we should show little sympathy towards, considering what brought her to Jackson – the irony of Joel rescuing her is not lost on anyone.It’s therefore a shame that such tension in that scenario is then completely let out by a lengthy detour back into the town, the new target of the infected’s ire. It takes all of the emotional stakes out of the episode for a good chunk of time as the characters we’ve grown most attached to (Joel and Ellie included) find themselves outside the walls and away from the onslaught.The battle, as a result, almost serves as a distraction from Joel's death, which is overshadowed by the bombast.“Episode 2 is a budget-flexer, and whereas the first season was restrained in its infected approach, it’s enjoyable to see mayhem take hold and fire meet ice in a way only HBO knows how to pull off. There’s no doubting the skill at hand when it comes to the production design and performances that breathe life into it. The invasion is stunningly shot at times, with vast mountain faces beautifully photographed as characters dot them like ants and snarling Clickers rise from the snow in incredible detail. There’s no denying the spectacle that the battle itself delivers – the Bloater breach and subsequent flamethrower retaliation is a real standout piece of staging. But Tommy and his family aside, our ties to these characters are limited.The battle, as a result, almost serves as a distraction from Joel's death, which is overshadowed by the bombast. The fateful scene itself is well constructed, but just feels at odds with much of the episode. I should make it clear that I’m a huge fan of The Last of Us Part 2 and the bold decisions the game makes; when I first played the game, this turn packed a hell of a gut punch, shocking me with its stark violence and sudden nature. Maybe I’ve grown desensitized to its brutality after experiencing it several times, but I think my muted reaction to episode 2 is also due to a lack of calm before the storm. With some of our attention directed to what’s going on a few miles from the lodge that’s been commandeered by Abby and her WLF comrades, we’re ushered into the torture suddenly before the re-creation of the game’s swift, devastating shotgun blast to Joel’s knee.PlayOnce we’re fully in the lodge, however, the scene mostly achieves its goals. The unsteady camera takes on the palpable unease as it floats around the room and Owen, Mel, Manny, and Nora share uncomfortable glances with each other. They’re in stark contrast to Abby, who is cold and menacing, and taken hold by an anger that instills fear in her closest allies. It’s a great showcase for Kaitlyn Dever’s range, which is fully on display as she slips into an almost sadistic mindset – the slow focus pull from her determined face to a nearby set of golf clubs almost met with a smile.I am torn about the actual method of Joel’s departure, though. On the one hand, I like the finality of the mangled club shaft nestling into his neck and the way Abby leaves sitting there almost like a monument to her victory. It also offers some nice symmetry to when Joel finds himself impaled in season 1, but this time, there is nothing Ellie can do about it. And though it’s undoubtedly violent, some of the shock is dulled, falling short of the blunt punctuation delivered by Abby’s final swing in the game.Of course, Pedro Pascal is no stranger to having his head pulped on HBO (the GoT parallels abound in this episode), and in depicting the death of the show’s antihero (or villain, depending on your viewpoint), he delivers a note-perfect performance. There’s an acceptance in his eyes that his time has come and an acknowledgment that he must pay for his decisions. It may be barely believable that Pascal is portraying a 60-year-old Joel here, but I completely bought into his softer portrayal of a character Troy Baker originally brought a gnarlier edge to. A word should also go out to Bella Ramsey who excels at conveying Ellie’s devastation. Her cries are heartfelt and genuine, but laced with an anger towards Abby that seemingly won’t soon dissipate. PlaySo there is some shock in this sequence – and within the 15 minutes or so where it’s the sole focus of the episode – it’s just a shame then that so much of the runtime is dedicated to a story that arguably could have been told as a separate chapter. It’s breathtakingly rendered at times, but watching people we barely know fall prey to waves of infected takes us away from the episode’s emotional core all too often. (Give or take Tommy’s alleyway blowout with a Bloater.) This is a rare miss from a creative team that has historically weaved together its big moments with much less strain.