Examining what it means to make money by selling the bereaved on an illusion, the film feels like something of a warning.

MacworldAt a glanceExpert's RatingProsEstimated 10-year lifespan versus 6 to 12 months for an AirTag-compatible lithium coin batteryWaterproof against steam cleaning and high-pressure jets (IP69)Low cost relative to standard rugged waterproof AirTag casesUses standard AA batteriesScrews use standard hex-head style (hex wrench included)ConsLarger form factor compared to AirTagOur Verdict The waterproof case design and use of standard AA batteries lets an AirTage in the TimeCapsule 10-Year Battery Case keep tracking stuff for extended periods.Price When ReviewedThis value will show the geolocated pricing text for product undefinedBest Pricing TodayApples AirTag has proved an invaluable tool in tracking down baggage lost or delayed by airlines, misplaced items youve left behind, or stuff stolen from you, whether a bag containing a camera or your car or bike. Its biggest failing? The CR2032 lithium coin battery required to power it lasts about 6 to 12 months. (Appleclaims it can last over a year, but I and many others see a more typical shorter life using brand-name batteries.)Some companies offer Find My network compatible items certified by Apple that have non-replaceable batteries with longer lives, like the 2-year warrantied promise of theChipolo Card Spot(review), after which they offer a discounted replacement. Others, like Pebblebee with itsCard Universal(Apple or Google network support) (review) or KnogsScout Bike Alarm & Finder(review) let you recharge an internal battery via external contacts or a port.Apple AirTagRead our reviewNow, theres something new: not a new Find My device, but a new case that incorporates external batteries. With an estimated 10-year lifespan using two extended-use lithium AA batteries, the Elevation Lab TimeCapsule 10-Year Battery Case may be an inexpensive answer for certain use cases where you want to go huge periods between changing batteries.A clever case adds external batteriesThe AirTag gave rise to a huge number of cases (now numbering in the hundreds) from dozens of companies. But only one has risen to the challenge of long life. Elevation Lab, an early maker of rugged AirTag cases for bikes, pet collars, and other uses, has released the TimeCapsule 10-Year Battery Case. (Yes, its called a TimeCapsule, one word, and has nothing to do with Apples discontinued networked Time Machine/AirPort Base Station product, the Time Capsule.)The TimeCapsule 10-Year Battery Case is shown here disassembled (left, right) and put together (bottom). The AirTag (top) fits securely on top of the CR2032 battery adapter in the case.This case uses a clever internal adapter to power an AirTag with two AA batteries instead of the small coin battery. The case is also highly waterproof, with an IP69 rating versus the AirTags native IP67 rating.The company achieves its life-extending trick by having you remove the plastic backing of an AirTag as if to replace the coin battery. The interior of the TimeCapsule case has a metal raised area the exact dimensions of a CR2032 battery. You place the open side of the AirTag onto it, add two AA batteries, and use the included screwsapparently milled to precision instead of stampedto firmly seal the case.Why might you want an AirTag that lasts ten years between battery changes? Apple prompts you in the Find My app and through notifications as the battery life of an AirTag or other item starts to get low. If you ignore these notifications long enough, they disappear when the battery dies. Elevation Labs founder said in announcing the product that he was inspired by just this when he discovered hed failed to replace the battery in an AirTag inside a camera bag after it was stolen, and the last location was tracked three months previously.You might also want to place an AirTag in a location thats difficult to reach or requires disassembly to get to, and youd ordinarily avoid with the knowledge you would need to get to it every several months to replace the battery. That could be useful if you want to put an AirTag somewhere that a thief couldnt remove it without substantial effort, taking minutes or even hours, but where the AirTags Bluetooth signal will still penetrate to reach nearby devices. An AirTag will keep broadcasting unless its battery is removed orits removed from the paired users account. If the AirTag (or third-party item) is not within Bluetooth range when removed from an account, it has to be reset by hand in addition to the account removal.Pick a battery that is rated for long usage.Pick the right AA batteryOn the battery front, dont use standard alkaline batteries, which the company warns against. Alkalines wont last long enough and can degrade in the voltage they produce or even leak battery chemicals. You may have seem this leakage before when youve spotted a white crust on the terminals inside a device that hasnt had its batteries changed for a while.ElevationLab recommends the Energizer Ultimate Lithium battery, which is highly rated in independent testing, such asthis exhaustive examination at Cinema Sound. Theyre about $2 to $4 each depending on the quantity purchased in a package. The Energizer batteries shouldnt need replacement under conditions measured by ElevationLab for as long as ten years. (Theres no way for us to review that particular aspect, so check back on this review in the year 2035.)You could use rechargeable lithium batteries, such as those fromPale Blue Earth, but they cost two to five times as much each, last for hundreds of recharges, and are better suited for heavy continuous power usage, like with a battery-powered doorbell cameraAssembling and deploying a caseYou receive a TimeCapsule in disassembled form with a set of four hex screws and a small hex wrench. You can also use a standard H2.5 (2.5mm) hex bit or screwdriver in a standard tool set. Assembly takes a minute or less. Once screwed shut, the case has an IP69 rating. The IP67 rating of the AirTag makes it resistant to 30 minutes of submersion at a depth of a meter (3.3 ft). IP69 allows for prolonged immersion at up to a meter plus provides protection against jet sprays.The assembled TimeCapsule 10-Year Battery Case provides a high level of waterproofing.The case is large compared to an AirTag due to housing two AA batteries and requisite circuitry. It measures 4.45 by 1.57 by 0.75 inches (11.3 by 4 by 1.9 cm). Make sure the place you want to use it can accommodate the bulk. The product doesnt come with a mounting bracket, screw holes for mounting, or an adhesive. Youll have to figure that out for locations where it cant float, as in a bag.The only downside I can see of this ten-year battery case is that Apple may improve its AirTag enough in the meantime that youll want to swap in a fresher one with newer features, or a third party will develop a longer-live, rechargeable battery with a bigger form factor. Apple could also change the size and shape of an AirTag, requiring a case redesign to use newer ones.However, you have years to think about that while your TimeCapsule-encased AirTag keeps on pinging.Check out Macworlds guide to the best AirTag key rings, luggage tags, wallets & accessories.

President Donald Trump has signed a sweeping executive order to make the US a global leader in artificial intelligence (AI), revoking regulatory measures enacted by the Biden administration.The executive order, signed Thursday, tasks federal agencies with drafting an AI action plan within 180 days to strengthen US competitiveness and safeguard national security.Trumps directive focuses on fostering innovation by eliminating what he called harmful government overreach that hampered AI development under the previous administration.This Executive Order establishes the commitment of the United States to sustain and enhance Americas dominance in AI to promote human flourishing, economic competitiveness, and national security, Trump said in the order.Repealing Biden-era AI restrictionsThe new order reverses a 2023 executive order by then-President Joe Biden that imposed stringent regulations on AI developers. Bidens policy required companies to conduct safety testing and share results with the federal government before releasing AI systems with potential risks to national security, public health, or the economy.The Biden AI Executive Order established unnecessarily burdensome requirements for companies developing and deploying AI that would stifle private sector innovation and threaten American technological leadership, the executive order stated.Critics argued the measures stifled innovation and burdened developers with excessive compliance requirements. Trumps order directs federal agencies to rescind or revise all policies tied to the Biden-era rules.Trumps executive order has the potential to accelerate the development and deployment of AI models by easing compliance burdens on developers, said Mansi Gupta, senior analyst at Everest Group. However, this shift introduces significant risks for sensitive industries like healthcare, where responsible AI considerations are crucial.Earlier this week, on his first day in office, President Donald Trump revoked a 2023 executive order issued by Joe Biden that imposed stricter oversight on artificial intelligence (AI) development.AI action plan and streamlined oversightThe executive order calls for the creation of a comprehensive AI Action Plan to guide the US toward sustained global leadership. The plan will be led by top federal science and national security officials and will include efforts to streamline AI oversight, prioritize ethical development, and ensure models remain free from ideological bias.The order directs the development of an AI Action Plan to sustain and enhance Americas AI dominance, led by the Assistant to the President for Science & Technology, the White House AI & Crypto Czar, and the National Security Advisor, the order read.It also requires the White House Office of Management and Budget (OMB) to issue revised guidelines for federal AI governance to promote agility in private-sector collaboration in order to ensure that harmful barriers to Americas AI leadership are eliminated.Earlier this week, President Trump unveiled a private sector investment plan of up to $500 billion to bolster artificial intelligence infrastructure, aiming to position the US ahead of global competitors in this critical technology sector.Implications for enterprisesTrumps pro-AI stance signals reduced regulatory friction for US enterprises. Industry leaders anticipate increased investments in AI research and development, particularly in sectors such as defense, healthcare, and autonomous technologies.However, the deregulation has sparked concerns over potential ethical lapses and security risks. Without robust oversight, critics warn that biases and malicious uses of AI could proliferate.The responsibility for ensuring ethical and accurate AI now shifts from developers to enterprises, which must navigate the challenge of balancing rapid innovation with rigorous risk mitigation measures, Gupta noted.Building on a legacy of AI leadershipThe order builds on Trumps 2019 executive order, the first US policy framework on AI, which doubled research funding and established national AI research institutes. Trump also issued federal guidance on AI adoption, making the US an early player in shaping global AI governance.This new directive reaffirms Americas commitment to innovation, free speech, and economic growth, the White House said.With the new AI policy, the Trump administration aims to recalibrate the balance between fostering innovation and ensuring responsible development. Enterprises are likely to welcome reduced compliance hurdles, but the long-term success of the initiative will depend on industry players ability to uphold ethical practices amid the drive for dominance. As the countdown begins for the AI Action Plan, the executive order positions the US at a crossroads where innovation and governance must align to maintain leadership in the AI race.

OpenAI has unveiled Operator, a new AI agent designed to perform web-based tasks, offering potential productivity enhancements for enterprises.The tool enables interaction with on-screen elements, positioning it as a solution for automating routine processes in business workflows amid growing competition in the generative AI space.Powering Operator is Computer-Using Agent (CUA), a model that combines GPT-4os vision capabilities with advanced reasoning through reinforcement learning, OpenAI said in a blog post. CUA is trained to interact with graphical user interfaces (GUIs) the buttons, menus, and text fields people see on a screen just as humans do. This gives it the flexibility to perform digital tasks without using OS- or web-specific APIs.CUA leverages years of research in multimodal understanding and reasoning, combining advanced GUI perception with structured problem-solving, the company added.It can break tasks into multi-step plans and self-correct when encountering challenges, representing a significant step in AI development by enabling models to use tools commonly relied on by humans and unlocking new application possibilities.The increasing role of AI agentsAI agents, designed to handle tasks such as scheduling and online transactions, are gaining interest in corporate AI initiatives. Rumors about OpenAIs agent have been circulating for months.Perplexity introduced an Android-based assistant this week, offering features like booking reservations, ride-hailing, and reminders.Apple expanded Siris capabilities last year by integrating Apple Intelligence and adding ChatGPT support through a partnership with OpenAI.Last year Anthropic also introduced a feature for its AI models called Computer Use, enabling Claude 3.5 Sonnet to perform tasks on computers autonomously.However, analysts suggest that Operator may have some advantages over the competition.Operator is quite advanced than Perplexity and other AI agents out there and more customizable and configurable, said Neil Shah, partner and co-founder at Counterpoint Research. The capability of the agent to let user takeover when needed or confirm actions with users or filtering out sensitive info or keeping watch is unique and gives users more control while being autonomous.Another differentiator is that most current agents are designed to take direct action based on user prompts and requests but not to maneuver through websites.Open AIs Operator is designed to be a web agent that can autonomously go through websites and conduct multi-step tasks, said Hyoun Park, CEO and chief analyst at Amalgam Insights. This is designed to be helpful both for accessing data within each website as well as conducting complex and time-consuming tasks that currently require repetitive clicking and typing.Enterprise applications and accessibility potentialAI agents open doors for various industries, particularly those seeking to enhance efficiency and streamline workflows. Their ability to automate tasks such as data collection and interaction with web-based platforms offers significant value for businesses.AI agents like Operator, still in their nascent stages, have the long-term potential to revolutionize industries such as customer service, healthcare, retail, and logistics by automating repetitive tasks, personalizing interactions, and enhancing workflow efficiency, said Prabhu Ram, VP of the industry research group at Cybermedia Research.Additionally, the tool redefines the concept of accessibility, making it easier for individuals who struggle to navigate or interact with the web to access online resources.This agent could also be useful in helping employees to quickly gather information or access all of the accessible data and content on a website that would be appropriate for that user to be able to see, Park said. Web agents will likely be an important tool for collecting long tail information on websites that may be hidden behind multi-step workflows that are time-consuming or difficult for humans to negotiate or maneuver.With features like custom API integration and configurability, the Operator tool could also benefit enterprises by enabling them to deploy these agents for internal purposes, such as extracting and organizing data from their own websites or intranets, Shah added.Safety concerns to overcomeAI agents introduce a new wave of safety challenges, with potential risks including misuse for bypassing system safeguards.These risks can range from automating form submissions on public sector websites to launching traffic attacks that disrupt website performance or evading CAPTCHA protections, among other violations.OpenAI, in its blog post, said that a layered safety approach with safeguards for the model, system, and post-deployment processes is essential.OpenAI also needs to ensure how this will protect the privacy of sensitive information when used to fill forms, whether it remains on the device, and that data used for reinforcement is not eventually misused for ads or sponsored listings as a lucrative business model, Shah said. The tools capabilities could also pose challenges for Google and other search engines, which rely on collecting user data and processing cookies to target ads. By giving users and OpenAI more control over data, the technology may disrupt traditional advertising models.

SvetaZi/Getty Images IN BRIEF: AI-First Leadership: Leaders must reimagine how humans and AI collaborate to harness AIs potential and bridge the gap between technological capabilities and strategic goals. Role of Midlevel Leaders: Midlevel leaders are crucial in driving strategy execution and enabling transformational change by embedding AI into personal practices, team workflows, and cross-functional processes. Developing AI-First Capabilities: Organizations need to invest in a structured developmental journey for leaders, focusing on building foundational AI knowledge, cultivating an AI-first mindset, honing AI-specific skills, and leading with confidence. In the era of generative artificial intelligence (Gen AI), leaders across industries are realizing []The post AI-First Leadership: Embracing the Future of Work appeared first on Harvard Business Publishing.

html PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN" "http://www.w3.org/TR/REC-html40/loose.dtd"Originally posted on 12 June 2024. Scroll down for details of the Dash 1.8.5 update.Polygonflow has updated Dash, its world building toolset for Unreal Engine 5.Dash 1.7 adds a new vines tool for growing vegetation procedurally across any 3D model, a new Blend material and an instant color grading system.Dash 1.8 adds Dash Board, a new PureRef-style reference image organizer, and a Chat-GPT-based AI assistant.Dash 1.8.5 adds Data Tables, a new system for layering effects, makes it possible to create terrain using height maps or guide curves, and cuts the price of the software.A new name for Polygonflows GraphN Library, minus GraphN itselfReleased in 2023, Dash is a rebranded version of the library of Unreal Engine 5 tools that used to accompany GraphN, Polygonflows now-discontinued node-based programming environment for Maya and Unreal Engine.It is intended to let artists to build complex environments without having to navigate the Unreal Editors interface, working primarily in fullscreen mode in the viewport.It provides readymade behaviours accessible by typing natural-language search terms into a floating prompt bar for common scene-building tasks like create terrain and apply water, plus scattering tools for dressing environments.As well as game development, Dash can be used for offline work: for example to create animations, visual effects projects or architectural visualizations in Unreal Engine.https://www.cgchannel.com/wp-content/uploads/2024/06/240611_Dash17_WorldbuildingToolUnrealEngine5_tw.mp4Dash 1.7: new vines tool generates procedural vegetation across any 3D modelDash 1.7 adds a lot of new features to the software, but the most eye-catching is Create Vines on Selection, for growing 3D vegetation procedurally across any surface.Despite the name, it isnt confined to growing vines, since you can use texture atlases from the Megascans asset library free for use in UE5 to customize the leaf type.The basic workflow will be familiar to anyone who has used an ivy generator such as the ones available for 3ds Max and Blender, with users able to control the form of the vegetation by adjusting parameters for growth, surface adhesion and gravity.New Blend material and automatic color grading systemOther new features include the Blend material, which makes it possible to create custom ground surfaces by blending from three source surfaces.Polygonflow describes is as being a lite version of Quixel Mixer, blending the surface color, height map and noise settings from the sources.There is also a new Image to Grading system, which makes it possible to color grade the viewport display to match a reference image: either a local file, or a URL.In addition, Dash now includes a library of IES profiles for recreating real-world lights in architectural visualizations, based on free online libraries like IES Library.https://www.cgchannel.com/wp-content/uploads/2024/06/240611_Dash17_VolumeScatter.mp4Updates to the scattering tools, including support for 3D scatteringThe scattering toolset has also been expanded, with the new Volume Scatter tool making it possible to scatter objects in 3D, as well as over surfaces.It is possible to use any mesh as the scatter volume, as shown in the video above, where a blobby mesh is used as a volume within which to distribute a flock of birds.The 2D scattering tools get the option to use any landscape paint layer as a mask, and the randomize the colors of the scattered objects. Automatic AI-based object taggingIn addition, Dash 1.7 makes it possible to tag assets within a users Asset Library automatically using Open AIs GPT-4 AI model.AI tagging was originally introduced in Dash 1.5, but Polygonflow describes the original open-source AI model it was using as fairly limited in quality and speed.The new GPT-based system uses a ChatGPT Enterprise plan, so OpenAI cannot use Dash users data for training its AI models.When tagging assets manually, it is now possible to multi-select objects.Other workflow improvementsWorkflow improvements include a new Reference system, which supersedes the old Variables system for controlling shared properties used by Dash tools.It is also now possible to export assets and all of their dependencies from an Unreal Engine project as .zip archives.Other key changes since we last wrote about Dash include the option to use assets from multiple UE5 projects in an environment, which was added in Dash 1.6.Updated 23 October 2024: Polygonflow has released Dash 1.8.The update reworks Dashs user interface, splitting features into distinct menus, and unifying many floating panels into a single Tools panel.The softwares path-based tools Road Tool, Quick Pipe, Decal Scatter and Path Scatter have also been revamped from the ground up to make them easier to use.New PureRef-style image organizer and new AI assistantNew features in Dash 1.8 include the new Dash Board tool, an experimental new reference image organization tool, described as similar to Miro and PureRef.The update also adds a new Chat-GPT-based AI assistant, which makes it possible to type in plain-language questions and receive answers from the online documentation.Updates to the Surface Scatter tool and Blend materialThe Surface Scatter tool gets several new features, including the option to use the top, bottom and center of objects being scattered as pivot points, and to modify the size of scattered objects according to distance from a target object.The Blend Material added in Dash 1.7 has been updated to support rain and snow.Updated 23 January 2025: Polygonflow has released Dash 1.8.5.New features include Data Tables, an experimental system for layering effects.In the initial release, it is limited to layering masks to control object scattering, and curves to deform terrain, but Polygonflow aims to extend it to other systems in the next release.Create terrain surfaces using guide curves or heightmapsUsing curves to deform terrain is itself a new feature in Dash 1.8.5: in previous releases, it was only possible to use noise patterns to deform a landscape mesh.The update also introduces basic heightmap support.Other updated featuresOther changes include new options to hide visible texture repetitions in the Material Edit and Blend Material tools, and new options to control object rotation in the scattering tools. The update also adds support for Fab, Epic Games new online marketplace, with Unreal Engine assets downloaded using the Fab plugin now added to the Dash Content Browser.Price cuts to subscriptions, and perpetual licenses for studios as well as freelancersPolygonflow has also cut the price of Dash. For Freelance users those with funding under $250,000/year the cost of monthly subscriptions falls from $20/month to $14/month, while annual subscriptions fall from $190/year to $133/year.For larger studios, the cost of monthly Enterprise subscriptions falls even further, from $89/month to $14/month, although annual subscriptions remain unchanged at $850/year.Studio can also buy perpetual licenses of Dash, which were previously only available to Freelance users.Price and system requirementsDash is compatible with Unreal Engine 5+ on Windows 10+ only.For artists earning under $250,000/year, perpetual Freelance licenses cost $230, while Freelance subscriptions cost $14/month or $133/year.For larger studios, perpetual Enterprise licenses cost $999, while Enterprise subscriptions cost $14/month or $850/year.Read a full list of new features in Dash in the online release notesRead more about Dash on Polygonflows websiteHave your say on this story by following CG Channel on Facebook, Instagram and X (formerly Twitter). As well as being able to comment on stories, followers of our social media accounts can see videos we dont post on the site itself, including making-ofs for the latest VFX movies, animations, games cinematics and motion graphics projects.

Pocketpair launches publishing arm, partners with Surgent Studios for first titleSurgent Studios to release standalone horror game in late 2025, unrelated to Tales of Kenzera universeImage credit: Pocketpair News by Sophie McEvoy Staff Writer Published on Jan. 24, 2025 Palworld developer Pocketpair has launched a publishing division, and has partnered with Surgent Studios to produce its upcoming horror title.The untitled game is set for release in late 2025, and is not set in the Tales of Kenzera universe. Surgent Studios clarified that it is continuing conversations with potential partners centered on games set in that world."Both Surgent and Pocketpair are well-versed in taking risks," said Surgent Studios CEO Abubakar Salim. "We noticed a pattern in the entertainment industry, and Pocketpair has given us the opportunity to make a horror game about it.""We're still in an earnest conversation about further projects set in the Tales of Kenzera universe, but this will be a standalone piece: a mile marker between where we've come from and where we're going."Head of Pocketpair Publishing John Buckley added: "As the games industry continues to grow, more and more games find themselves struggling to get funded and greenlit.""The ultimate goal of Pocketpair Publishing is to find creators with a genuine passion for games and enable them to create fun, new experiences all over the world."Last October, Surgent Studios put its entire games division on hiatus as it struggled to find funding for future projects. An unknown number of staff were put at risk of being made redundant.In July 2024, over a dozen workers were affected by layoffs three months after the release of Tales of Kenzera: Zau.

Its been three long years since Sony launched its flagship WH-1000XM5 headphones, our pick for the best wireless headphones for most people, so were due a follow-up and a regulatory filing indicates that the XM6 cans might nearly be ready to launch. The Federal Communications Commission (FCC) filing hints at a slight redesign, including a new hinge and what appear to be detachable earpads.The Walkman Blog first spotted the filing, which includes a product diagram showing off some of the details of the right earcup. Most interestingly, the diagram shows the earpad lifting off to reveal the location of the user-visible name plate containing the model and serial numbers. Since the FCC requires this name plate be accessible to users, it seems likely that the pad is attached without screws, though we dont know if thats with clips, magnets, or something else. The XM5s hid their FCC details under the inner edge of the pads instead, which were not removable.Sonys FCC filing suggests that its upcoming headphones might have a removable earpad. Image: SonyOtherwise theres not too much to glean from the filing. The headphones hinge looks slightly different from the previous generations, though theres no reason yet to believe that means a return to the folding design of the XM4s. The Bluetooth version is reported to be 5.3, and there are some details on the antenna and power specs, but nothing to suggest how the headphones will look or sound.The FCC filing doesnt name these as the XM6 headphones, but the model numbers seem to give it away. These headphones have the model number YY2984, remarkably similar to the XM5s YY2954 designation. The only note of caution is that the tested model is marked as a prototype, rather than a pre-production or mass production unit, so some details may change before release.The FCC filing includes a short-term confidentiality request dated to end on July 22nd 2025, which certainly suggests that well get an official reveal before that date. The XM5s launched in May 2022, so we may be due another spring release.

Large language models (LLMs) have introduced impressive capabilities, particularly in reasoning tasks. Models like OpenAIs O1 utilize long-thought reasoning, where complex problems are broken into manageable steps and solutions are refined iteratively. While this approach enhances problem-solving, it comes at a cost: extended output sequences lead to increased computational time and energy use. These inefficiencies raise concerns about scalability and the practical usability of such models in real-world applications. Addressing this issue is essential for making LLMs more efficient and broadly applicable.Researchers from Sun Yat-sen University, China Agriculture University, Tsinghua University, the University of Oxford, Didichuxing, and NTU propose Length-Harmonizing Fine-Tuning (O1-Pruner). This technique seeks to reduce the inefficiencies in reasoning models while maintaining accuracy. The primary focus is on optimizing token usage, which is a significant bottleneck in current models. O1-Pruner uses reinforcement learning (RL) techniques to encourage the generation of shorter reasoning paths without sacrificing precision.The process begins with evaluating baseline performance through pre-sampling. A customized RL-style loss function then fine-tunes the models reasoning length, ensuring that the generated solutions are proportional to the complexity of the problem. By aligning reasoning length with task difficulty, O1-Pruner reduces computational costs without compromising on quality.Technical Details and Benefits of O1-PrunerAt the heart of O1-Pruner is the Length-Harmonizing Fine-Tuning approach, which balances reasoning length and accuracy. The key steps include:Reference Model Sampling: A reference model evaluates reasoning quality and length by generating multiple solutions for each problem, creating a performance benchmark.Reward Function Design: This involves two components:Length Reward: Shorter solutions relative to the reference model are encouraged.Accuracy Reward: Ensures that shorter reasoning paths do not compromise correctness.Reinforcement Learning Framework: Proximal Policy Optimization (PPO) is used to train the model efficiently. Off-policy training further simplifies the workflow and reduces training complexity.The benefits of O1-Pruner include:Improved Efficiency: Reduces redundant computations, leading to faster inference.Accuracy Preservation: Ensures that shorter solutions maintain or even enhance accuracy.Task Adaptability: Dynamically adjusts reasoning depth based on problem complexity, making it applicable to a variety of tasks.Results and InsightsExperiments on mathematical reasoning benchmarks such as MATH, GSM8K, and GaoKao showcase O1-Pruners effectiveness. For example:The Marco-o1-7B model, fine-tuned with O1-Pruner, achieved a 40.5% reduction in solution length while improving accuracy to 76.8%.The QwQ-32B-Preview model demonstrated a 34.7% reduction in solution length alongside a slight accuracy increase to 89.3%.Inference time also improved significantly. On the MATH dataset:Marco-o1-7B reduced its inference time from 2 minutes to just over 1 minute.QwQ-32B-Preview decreased from 6 minutes to approximately 4 minutes.These results highlight O1-Pruners ability to balance accuracy and efficiency. Its superior performance, as measured by the Accuracy-Efficiency Score (AES), establishes it as a better alternative to other methods like Supervised Fine-Tuning (SFT) and Direct Preference Optimization (DPO).ConclusionO1-Pruner demonstrates that efficient reasoning in LLMs is achievable without compromising accuracy. By harmonizing reasoning length with problem complexity, it addresses the computational inefficiencies inherent in long-thought reasoning. This work lays the groundwork for further advancements in optimizing reasoning models, enabling their application in diverse, real-world scenarios where efficiency and accuracy are equally critical.Check out the Paper and GitHub Page. All credit for this research goes to the researchers of this project. Also,dont forget to follow us onTwitter and join ourTelegram Channel andLinkedIn Group. Dont Forget to join our70k+ ML SubReddit. Aswin Ak+ postsAswin AK is a consulting intern at MarkTechPost. He is pursuing his Dual Degree at the Indian Institute of Technology, Kharagpur. He is passionate about data science and machine learning, bringing a strong academic background and hands-on experience in solving real-life cross-domain challenges. Meet 'Height':The only autonomous project management tool (Sponsored)

Mythic Quest's two-part season 4 premiere streams on Apple TV+ beginning Wednesday, January 29, with new episodes debuting weekly through March 26.After lengthy hiatuses, Severance and Mythic Quest return to the office this month. But thats not the only thing the sci-fi thriller and the workplace comedy have in common: Both Apple TV+ shows are equally invested in exploring personal and professional boundaries (or lack thereof). In its fourth season, the MQ team still struggles to separate home and office, but Poppy Liu (Charlotte Nicdao) embraces a work work life balance. That phrase, with its purposeful repetition, hits on the relatable and absurd qualities of this workaholic environment even after all this time away, Mythic Quest remains sharp and incisive in depicting the world of video-game development. No cheat code is available to prevent Poppy from getting too entangled with creative partner Ian Grimm (Rob McElhenney), and the duo remains the series' driving force. Introducing another man into Poppys life (an artist called Storm, played by Chase Yi) upsets the equilibrium, giving the season a different energy from previous years. Disrupting the status quo ensures Mythic Quest isnt repeating itself, even if Poppy and Ian have returned to the MQ fold after their stint running GrimPop Studios.While Poppy and Ians rollercoaster partnership brilliantly anchors season 4, creators McElhenney, Megan Ganz, and Charlie Day (the last of whom guest stars this season) have always understood that the strength of Mythic Quest lies in its ensemble. Everyone is back in the same place, yet their relationships are more fractured than ever. Danas (Imani Hakim) game within Playpen is driving huge numbers, and still, MQ gets the lions share of the revenue it generates. The goalposts of success continue to move, and she tries to hit the elusive sweet spot between artistic fulfillment and earning her worth. Her girlfriend, Rachel (Ashly Burch), is the head of monetization for MQ, and theres no way to predict how the pair will navigate that conflict of interest. Jo (Jessie Ennis) and Brad (Danny Pudi) are part of Danas team, and they immediately add a level of intensity when anyone tries to enter Danas office. She might not have financial power, but Danas support system terrifies David (David Hornsby). This trio isnt the only winning dynamic in the mix. Season 4 soars when it pairs Rachel and David for a congressional hearing; the same goes for when most of the crew comes together for an Ethics Committee meeting. Theres no weak link here, but with a cast this big and so many ongoing threads to manage, some moments from earlier in the season fizzle too quickly. The gang gets a new hot potato to toss around in the form of generative AI, but through the first nine episodes of season 4, that topic with its profound implications for the characters onscreen, the actors playing them, the writers room, heck everyone involved with the production of Mythic Quest provides some brief laughs but not much else. Witnessing a pro-AI Ian come up against a flawed system only left me wanting more.Other ebbs and flows in the video game industry still provide a lot of humor and tension. Giving the users power turns Playpens kid-friendly platform into a haven of animated smut, delivering a bounty of visual gags. Though the platform still generates interest, an overall dip in sales in a post-lockdown world leads to a recurring gag in which David pines for the financially plum heyday of the pandemic. His people-pleasing attempts often backfire, yet his lack of spine, tone-deaf reasoning, and constant siding with corporate ensure you dont feel too bad for him. The more exasperated he becomes, the funnier Hornsby gets.Mythic Quest Season 4 GalleryPlaying with the formula keeps MQs dysfunctions fresh. (Another feather of invention and innovation in the series cap: The forthcoming four-part spinoff Side Quest, which debuts the same day as the fourth-season finale.) Season 4 doesnt skip the Mythic Quest tradition of a one-off detour from the main story, and though it doesnt hit the heights of season 1s A Dark Quiet Death, the episodes focus creatively fits the setting boundaries theme. This episode taps into a corner of the industry that continues to grow, and the sitcom maintains its funny and silly streak while delving into (but not getting bogged down by) real life issues.Another standout episode takes place outside the offices, with the entire team (plus Storm) receiving an invitation to a secluded location for a murder mystery. Playing games is central to the MQ business model, but in this case, some good old-fashioned sleuthing taps into the personal stakes that make Mythic Quest so appealing. In any creative field, you have to move forward and reinvent; this episode both demonstrates and achieves this in the way storylines advance and characters evolve. A whodunnit in the vein of Mafia (or even Traitors) provides a familiar enough backdrop to convince each of the participants that they a) know the game inside and out, and b) can win. Costume designer Sabrina Rosen nails the Agatha Christie-leaning wardrobe required for the episode, and the entire ensemble flourishes in this setting. The episode, directed by Ganz (who co-wrote it with Humphrey Ker), gets to the heart of the nuanced relationships that have been building since season 1, while also throwing thorny and sweet surprises into the mix. Mythic Quest remains sharp and incisive.Poppy and Ian have survived plenty of storms in the past, but a new one is brewing thanks to her love life. (Yes, there are plenty of puns on Poppys new boyfriends name, and no, I couldnt resist getting in the mix.) The platonic creative partners exist outside of a will-they/wont-they situation, yet Storm introduces all the ingredients of a classic love triangle. Ians jealousy isnt informed by romantic attraction, but his reaction to Poppy finding something fulfilling beyond her career makes the duos journey fascinating. If work is no longer all Poppy thinks about, then where does that leave Ian? They have weathered myriad conflicts (including the failure of GrimPop last season), and creating the next thing is the glue that sticks them together. However, Poppy finding a new source of happiness and even inspiration has the potential to break them apart. The previous victories and ruptures fueled by ego, expectation, and talent make this all work. Nicdao gets to tap into new and old insecurities and is equally convincing across a spectrum of emotions that make her the MVP of the season. And her comedic timing here is next level. Poppy is still a hot mess, which further adds to the comedy and this arc is a brilliant showcase for Nicdao (who also directs this season). PlayAt the helm, McElhenney is dependable in showing Ians endless arrogance and vulnerabilities. Change is hard, and while Ian resorts to old tactics, Mythic Quests depiction of evolving priorities ensures this season feels as fresh as ever, immediately pulling us back into this world after a prolonged absence and proving it is far from game over.