• Monster Hunter Wilds Will Reduce VRAM Usage on PC With Title Update 1
    gamingbolt.com
    While Monster Hunter Wilds is a commercial and critical success for Capcom, technical issues continue to nag players, like PC optimization. Director Yuya Tokuda acknowledged this in the latest Directors Letter and confirmed reduced VRAM usage and an upgrade to DirectStorage, which should improve stability.These will go live with Title Update 1 on April 3rd, but the team continues to identify stability issues and make improvements where possible, especially on the Steam version. This will be an ongoing process, where well aim to make continual steps forward in this area and respond to critical issues.Of course, it also wants to improve the overall technical experience for all platforms. Perhaps this could lead to further improvements for Xbox Series X/S and PS5 (which arent the best-looking versions by a long shot). The developer also plans to take a wider look at overall gameplay flow, which encompasses a range of things, such as the in-game economy, balance and other areas.Finally, the May end update will add a list of captured endemic life, thus making it easier to spot any missing entries. Layered weapons are also returning, finally allowing players to change the appearance of the Artian weapons to match their fashion. Theres no time frame aside from future update, so stay tuned for more details.Monster Hunter Wilds is available for Xbox Series X/S, PS5, and PC. It sold over 10 million copies in its first month. Check out our review here.
    0 Commentaires ·0 Parts ·7 Vue
  • This Eight-Pound Miniature Dachshund Survived 16 Months on a Rugged Australian Island. But She's Still Evading Rescuers
    www.smithsonianmag.com
    This Eight-Pound Miniature Dachshund Survived 16 Months on a Rugged Australian Island. But Shes Still Evading RescuersValerie the wiener dog is still on the loose, more than a year after she escaped during her parents vacation on Kangaroo Island Valerie was just a year old when she went missing on Kangaroo Island off the coast of South Australia. Kangala Wildlife Rescue via FacebookA tiny, eight-pound dog has defied the odds by surviving in the wild on a rugged Australian island for more than a year. But even as rescuers have tried to bring her to safety, so far, the scrappy pooch has eluded capture.Volunteers and wildlife experts are trying to lure in a miniature dachshund named Valerie on Kangaroo Island, a 1,700-square-mile outpost off the coast of South Australia.Valerie has been on the run for the last 16 months. In November 2023, the then-1-year-old wiener dog came to the island on a camping vacation with her human parents, Georgia Gardner and Josh Fishlock. On the second day of their trip, the couple decided to go fishing at a nearby beach, so they placed Valerie in a playpen with food and toys.But while her owners were away, Valerie broke out of the pen and hid under a parked car. Vacationers who were camping nearby tried to capture the dog, but she got spooked and darted off into the wilderness.The couple spent the rest of their vacation searching for Valerie, with help from some of the islands 5,000 residents. Eventually, though, they had to return to their jobs in Broken Hill, New South Wales, without their beloved pup.Gardner and Fishlock were heartbroken. They assumed Valerie would not survive in the bushland, which is home to several potentially fatal hazardsincluding at least two venomous snake species and wedge-tailed eagles that are known to hunt wallabies, possums and lambs.Even if she did dodge the islands many threats, Valerie was not accustomed to a life of hardship. She slept in bed with her parents each night, wore sweaters when the weather turned cold and got upset if she was left outside for too long. Valerie, who had been a college graduation gift from Gardners family, also loved accompanying her parents to cafes and shops.She was an absolute princess, Gardner tells the Washington Posts Victoria Craw, adding that Valerie was anxiously attached to her parents.She was not a very outside, rough-and-tough dog, Gardner says to the Guardians Daisy Dumas. To think that she even went one night outside in the rain, oh my gosh.But roughly a year after Valerie went missing, reports started coming in of a small dachshund on Kangaroo Island, wearing a pink collar. The sightings occurred roughly nine miles from Stokes Bay, the area where Valerie had escaped.Now, volunteers with Kangala Wildlife Rescue, a local nonprofit, are doing everything in their power to capture Valerie and reunite her with her parents. Theyre using several trapping and luring methodsincluding aromatic foods like roast chicken and tunato try to catch the dog. Theyre also using video surveillance to keep tabs on her movements.We now know that Valerie is alive, the rescue organization wrote on social media on March 21. She runs at the first sign of humans or vehicles, and despite the best efforts of dedicated island locals, Valerie has been impossible to catch.The island is roughly 75 times the size of Manhattan, so rescuers are hoping they can catch a break in their quest to corral Valerie.This is a tiny dog in a huge area, and we will need help from the public to report any sightings and a lot of luck, according to the social media post.How has a small, domesticated dog managed to survive for so long in the wild? Rescuers believe Valerie is likely subsisting on roadkill and dam water.Its also possible shes receiving help from the islands residents. But, more than likely, shes been making it on her own, because if someone had seen her, they probably would have noticed her collar and reported her.Plus, dogs are extremely resourceful, says Paul McGreevy, a veterinarian at the University of Sydney, to the Guardian. They are the greatest opportunists in the animal kingdom: Thats one of their core skills.Dachshunds, in particular, were bred to be tireless hounds and independent hunter[s] of dangerous prey, according to the American Kennel Club (AKC). These short-legged dogs are known for their bold, vivacious personality, and they can be brave to the point of rashness, per the AKC.Debbie Farnden, a 50-year-old nurse who volunteered to search for Valerie on Kangaroo Island, is not at all surprised the sausage dog has lived this long. Farnden has two dachshunds of her own, so she knows firsthand how quick and agile these dogs can be.Theyre sneaky little buggers and smart enough to stay away from snakes, Farnden tells the London Times James Salmon. They are fast and cunning and will play the waiting game.Get the latest stories in your inbox every weekday.Filed Under: Animals, Australia, Dogs, Mammals, Pets, Wildlife
    0 Commentaires ·0 Parts ·6 Vue
  • Who Drank Wine in Ancient Troy? New Research Suggests Just About Everyone
    www.smithsonianmag.com
    A depas goblet excavated from the ruins of Troy by Heinrich Schliemann in the 1870s University of TbingenIn the first book of the Iliad, the god Hephaestus passes a double goblet around at a banquet on Mount Olympus. He poured the drink, going from right to left, for all the other gods, drawing off sweet nectar from the mixing bowl, the epic poem states.For those who enjoy libations from goblets or glasses, the rowdy evening that ensues should sound familiar. Their laughter broke out irrepressibly, Homer writes. No ones heart went unsatisfied.The Iliadis, of course, a work of mythology. But that doesnt mean all the practices, people and places depicted in the poem are fully fictive.The drinking vessel that Hephaestus passes around, for instance, is often identified as the depas amphikypellon, or depas goblet, a well-known relic among archaeologists that features a slender neck and two large handles. Schliemann's haul from Troy, on view at the Neues Museum in Berlin Public domain via Wikimedia CommonsBut whether the ancient residents of Troy truly sipped wine out of these goblets has long been consigned to the realm of speculation.Now, for the first time, researchers have identified chemical residues associated with wine in goblets unearthed at Hisarlik, the Turkish name for a site believed to be the ancient city immortalized in Homers epic, according to a new study published in the American Journal of Archaeology. Heinrich Schliemann, a German businessman and amateur archaeologist with a penchant for embellishment, discovered and haphazardly excavated the site in the 1870s, wrote Smithsonian magazines Meilan Solly in 2022.Schliemann already conjectured that the depas goblet was passed around at celebrationsjust as described in the Iliad, says Stephan Blum, an archaeologist at Germanys University of Tbingen and a co-author of the study, in a statement. But, characteristic of Schliemanns assertions, there was little hard evidence to back up his sweeping claims.Archaeologists at Troy have unearthed more than 100 depas goblets dated to between 2500 and 2000 B.C.E. They tend to measure between 5 and 15 inches tall and can contain up to a liter of liquid, according to the statement. A depas goblet in situ at Troy University of TbingenFor the study, the researchers drilled two-gram samples out of the inner walls of two vessel fragments excavated by Schliemann. Then, they heated the samples to more than 700 degrees Fahrenheit. Using gas chromatography and mass spectrometry to isolate compounds in the mixture, the researchers identified the presence of succinic and pyruvic acids. Both are associated with alcoholic fermentation.The evidence of succinic and pyruvic acids was conclusive: They only occur when grape juice ferments, says Maxime Rageot, a biomolecular archaeologist at Germanys University of Bonn, in the statement. So now we can state with confidence that wine was actually drunk from the depas goblets and not just grape juice.As Popular Sciences Andrew Paul points out, however, these goblets werent everyday items.Schliemann discovered his astonishing cache of goblets among a cache of hundreds of objects made of gold, silver, copper and electrum, a mixture of precious metals, wrote Joshua Hammer for Smithsonian in 2022. He called the hoard Priams Treasure after the mythical Trojan king Priam. Although the treasure was later dated to about 1,000 years before the Trojan War took place in the 12th or 13th century B.C.E., it offered evidence of the stratification of social classes in Troy, raising questions about who had access to wine in ancient times.Did ancient Troy really exist? - Einav Zamir DembinWatch on To determine if wine was only the drink of Troys elitesand its godsthe researchers conducted similar chemical tests with ordinary cups that were found in the outer settlement of Troy and therefore outside the citadel, Blum explains in the statement.Common vessels, the team discovered, contained the same chemical signatures of wine. It is clear that wine was an everyday drink for the common people, too, Blum adds.These results upend longstanding assumptions that wine was an elite beverage during the third millennium B.C.E. Like grapes on the vine, further research into the practices of wine drinking at other sites across the ancient world promises to be fruitful.Schliemann was right: The depas amphikypellon was certainly used for wine consumption, writes Blum for the Conversation. Whether this was tied to religious practices, rituals and public banqueting, or simply drinking wine as part of everyday life remains uncertain.Get the latest stories in your inbox every weekday.Filed Under: Alcohol, Ancient Civilizations, Ancient Greece, Ancient Rome, Archaeology, Chemistry, Cool Finds, Legend, Myth, Turkey, Wine
    0 Commentaires ·0 Parts ·7 Vue
  • Emergence AIs new system automatically creates AI agents rapidly in realtime based on the work at hand
    venturebeat.com
    The new system is literally a no code, natural language, AI-powered multi-agent builder, and it works in real time.Read More
    0 Commentaires ·0 Parts ·8 Vue
  • London is now the third-largest global hub for game makers after LA and San Francisco
    www.gamesindustry.biz
    London is now the third-largest global hub for game makers after LA and San FranciscoThe UK capital has now overtaken most US cities and Canadian hotspots like MontralImage credit: Games London News by Vikki Blake Contributor Published on April 1, 2025 London is now the third-largest global hub for game makers.That's according to a new analysis by Unscrambled and BOP for London, which discovered the UK capital has "shot up" the leaderboard in terms of size of workforce, securing the runner-up spot after Los Angeles and San Francisco.London was already the "biggest UK cluster" and Europe's largest game development hub, but with 13,700 game developers working across the sector, it has now overtaken most US cities and Canadian hotspots like Montral.Another 9000 London workers work in "games-associated" areas or work in industries that have expanded to include games, such as entertainment and animation.The news comes as the tenth annual London Games Festival kicks off on April 2, with over 30 events across the capital "catering to tens of thousands of professionals and players.""London is now a global gaming capital and Europes leading city with a thriving industry bringing significant investment to our country," said London Mayor Sadiq Khan."I'm proud to support the London Games Festival, which is a great showcase for this dynamic and growing industry. As UKs biggest games event, it generates millions for our economy and helps support our up-and-coming talent as we continue to build a better London for everyone."
    0 Commentaires ·0 Parts ·6 Vue
  • Azra Games appoints PlayStation's former co-head of mobile, Kris Davis, as CBO
    www.gamesindustry.biz
    Azra Games appoints PlayStation's former co-head of mobile, Kris Davis, as CBO"His track record in mobile gaming and business operations is unmatched," says CEO Mark OteroImage credit: Azra Games News by Vikki Blake Contributor Published on April 1, 2025 Azra Games has appointed PlayStation's former co-head of mobile, Kris Davis, as its chief business officer.Davis will be tasked with overseeing and optimizing business operations as Azra ramps up for the release of its upcoming combat RPG, codenamed Project Legends, by "creating the overall business strategy, including short and long-term strategic plans."Davis brings over 20 years of experience, having held prior leadership roles at Beeline Interactive, Kabam, and PlayStation Studios.As CBO, Davis will also "provide leadership, set goals and objectives for product development and business intelligence, and manage sales, licensing, analytics, and business development.""My experience in mobile games business development will help drive Project Legends' success and the studio's growth. I am thrilled to contribute to Azra's mission of delivering unforgettable gaming experiences," Davis said."Kris Davis impressive history of success and his strategic insights will help shape the future of Azra Games and Project Legends. His track record in mobile gaming and business operations is unmatched. He will help us deliver games that captivate players on mobile and all other gaming platforms," added Mark Otero, CEO of Azra Games.
    0 Commentaires ·0 Parts ·6 Vue
  • Thousands of federal health workers are losing their jobs in the US
    www.theverge.com
    Drastic reductions in force are upending agencies within the Department of Health and Human Services (HHS), including the Centers for Disease Control and Prevention (CDC) and the Food and Drug Administration (FDA).Thousands of people who work at the CDC were notified by email today that they were subject to the Trump administrations efforts to cull federal workforce jobs, Wired reports. Top officials were among those either put on administrative leave, laid off, or reassigned to remote roles, the Washington Post reports.HHS announced last week that it would slash its workforce by 20,000 people. Now, the nation is starting to see how that purge is rolling out, affecting programs meant to prevent and treat HIV infection and sexually transmitted disease, respiratory diseases, and foodborne illnesses, to name a few.Were going to have patients die, Jade Pagkas-Bather, an infectious disease doctor at the University of Chicago, tells Wired. Unnecessary, preventable death.Unnecessary, preventable death.HHS, on the other hand, says the changes will save $1.8 billion a year. Over time, bureaucracies like HHS become wasteful and inefficient even when most of their staff are dedicated and competent civil servants, HHS Secretary Robert F. Kennedy, Jr. said in a press release when HHS announced its restructuring last week. Kennedy is a staunch anti-vax crusader who has spread disinformation falsely linking vaccines to autism. The FDAs Center for Biologics Evaluation and Research (CBER) regulates vaccines, and director Peter Marks resigned on Friday, writing that it has become clear that truth and transparency are not desired by the Secretary, but rather he wishes subservient confirmation of his misinformation and lies. The FDA as weve known it is finished, with most of the leaders with institutional knowledge and a deep understanding of product development and safety no longer employed, Robert Califf, FDA commissioner under Joe Biden and Barack Obama, wrote on LinkedIn today. In an email to The Verge, HHS press secretary Vianca Rodriguez Feliciano maintains that ongoing critical public health efforts will remain a top priority and will not be impacted by this administrative realignment.See More:
    0 Commentaires ·0 Parts ·8 Vue
  • All of the updates about OpenAI
    www.theverge.com
    What was once a humble research lab has transformed into one of the biggest consumer technology companies of all time.OpenAI, founded in 2015 to develop artificial general intelligence (AGI) AI systems with human-level intelligence has transformed dramatically since launching ChatGPT, which was once considered to be the fastest-growing consumer application in history. Most cofounders have left either to create a competitor or work for one. The company has secured billions in funding and partnerships with Apple and Microsoft, even announcing a $500 billion datacenter project called Stargate. Meanwhile, it faces copyright lawsuits from authors and news organizations, legal action from cofounder Elon Musk over the companys alleged departure from its original mission, and criticism for burning through cash despite projected billions in revenue. After Altmans brief ouster, OpenAI is now expected to restructure from a nonprofit-led organization to a full for-profit company to stabilize operations and reassure investors.As San Franciscos hottest AI company continues to barrel towards ever growing valuations, its claims become more nebulous. Altman expects we may see the first AI agents join the workforce and materially change the output of companies in 2025, and says his team is now confident we know how to build AGIas we have traditionally understood it. A decade since it was founded, OpenAI has become synonymous with the future of AI, with the tech industry and beyond closely monitoring its next moves.All of the news and updates about OpenAI continue below.OpenAI just raised another $40 billion round led by SoftBankChatGPTs Ghibli filter is political now, but it always wasOpenAI says our GPUs are melting as it limits ChatGPT image generation requestsChatGPT is turning everything into Studio Ghibli art and it got weird fastChatGPTs new image generator is delayed for free usersOpenAI rolls out image generation powered by GPT-4o to ChatGPTOpenAI reshuffles leadership as Sam Altman pivots to technical focusChatGPT accused of saying an innocent man murdered his childrenWhat does OpenAI really want from Trump?The questions ChatGPT shouldnt answerOpenAI announces GPT-4.5, warns its not a frontier AI modelChatGPT is a terrible, fascinating, and thrilling to-do list appMira Murati is launching her OpenAI rival: Thinking Machines LabOpenAI is rethinking how AI models handle controversial topicsOpenAI lays out plans for GPT-5What $200 of ChatGPT is really worthInside OpenAIs $14 million Super Bowl debutI tested ChatGPTs deep research with the most misunderstood law on the internetChatGPT drops its sign-in requirement for searchHeres OpenAIs new logoSoftBanks Masayoshi Son says AGI will arrive much earlier than he thoughtChatGPTs agent can now do deep research for youOpenAI launches new o3-mini reasoning model with a free ChatGPT versionSam Altmans Stargate is science fictionDeepSeek, Stargate, and the new AI arms raceOpenAI and SoftBank are starting a $500 billion AI data center companyLas Vegas police release ChatGPT logs from the suspect in the Cybertruck explosionOpenAIs Sam Altman says we know how to build AGIYou can now call 1-800-CHATGPTChatGPTs AI search engine is rolling out to everyoneSoras AI video revolution is still a ways offInside the launch and future of ChatGPTiOS 18.2 is rolling out now, adding ChatGPT integration and more Apple Intelligence toolsOpenAI has finally released SoraMicrosofts AI boss and Sam Altman disagree on what it takes to get to AGISam Altman lowers the bar for AGIOpenAIs 12 days of shipmas include Sora and new reasoning modelChatGPTs search results for news are unpredictable and frequently inaccurateSiris big ChatGPT upgrade is here for better and worseOpenAI plans to release its next big AI model by DecemberOpenAI was a research lab now its just another tech companyOpenAIs for-profit switch could include equity for Sam AltmanOpenAI says Iran tried to influence US elections with ChatGPTElon Musk is suing OpenAI and Sam Altman againOpenAI is making ChatGPT cheaper for schools and nonprofitsFormer OpenAI board member explains why they fired Sam AltmanOpenAI has a new safety team its run by Sam AltmanChatGPT has a Scarlett Johansson problemChatGPT will be able to talk to you like Scarlett Johansson in HerOpenAI releases GPT-4o, a faster model thats free for all ChatGPT usersSam Altman rejoins OpenAIs board after investigation into sudden firingOpenAI introduces Sora, its text-to-video AI modelOpenAI says theres only a small chance ChatGPT will help create bioweaponsOpenAI CEO Sam Altman is still chasing billions to build AI chipsChatGPT is winning the future but what future is that?Sam Altman on being fired and rehired by OpenAIMicrosoft joins OpenAIs board with Sam Altman officially back as CEOSam Altman to return as CEO of OpenAIWhat happened to Sam Altman?Sam Altman fired as CEO of OpenAIJony Ive is reportedly developing an AI gadget with OpenAIs Sam AltmanSam Altman sells superintelligent sunshine as protestors call for AGI pauseAs conservatives criticize woke AI, here are ChatGPTs rules for answering culture war queriesOpenAI announces ChatGPT Plus at $20 a monthChatGPT provesAIis finally mainstream and things are only going to get weirderOpenAIs new chatbot can explain code and write sitcom scripts but is still easily tricked
    0 Commentaires ·0 Parts ·9 Vue
  • The Complete Beginners Guide to Terminal/Command Prompt
    www.marktechpost.com
    The terminal (on Mac/Linux) or command prompt (on Windows) is a powerful tool that allows you to interact with your computer using text commands instead of clicking through a graphical interface. While it might seem intimidating at first, mastering basic terminal commands can help you:Navigate through files and folders more efficientlyPerform tasks that arent possible through the regular interfaceAutomate repetitive tasksGain a deeper understanding of how your computer worksThis guide will introduce you to the essential commands and concepts to get you started, regardless of which operating system you use.Getting StartedOpening the TerminalOn Windows:Press Win + R, type cmd, and press EnterOr search for Command Prompt in the Start menuOn Mac:Press Command + Space to open Spotlight, type Terminal, and press EnterOr find Terminal in Applications Utilities TerminalOn Linux:Press Ctrl + Alt + T (on most distributions)Or search for Terminal in your applications menuUnderstanding the PromptWhen you first open the terminal, youll see a prompt that looks something like this:Windows: C:\Users\YourUsername>Mac/Linux: username@computer:~$This tells you:Your current location in the file systemWhere to type your commandsOn Mac/Linux, the ~ symbol represents your home directoryBasic Navigation CommandsViewing Your Current LocationWindows: cdMac/Linux: pwd (Print Working Directory)Example:Listing Files and DirectoriesWindows: dirMac/Linux: lsExample:Options:ls -l List with detailed information (file size, date modified, permissions)ls -a Show hidden files (files that start with a dot)ls -la Combine both optionsChanging DirectoriesAll platforms: cd DirectoryNameExamples:Creating DirectoriesAll platforms: mkdir DirectoryNameExample:Creating FilesWindows: type nul > filename.txtMac/Linux: touch filename.txtExample:Working with FilesViewing File ContentsWindows: type filename.txtMac/Linux: cat filename.txtFor larger files:Windows: more filename.txtMac/Linux: less filename.txt (use q to quit)Copying FilesWindows: copy source destinationMac/Linux: cp source destinationExample:Moving/Renaming FilesWindows: move source destinationMac/Linux: mv source destinationExamples:Deleting Files and DirectoriesWindows:Mac/Linux: Warning: Be very careful with delete commands, especially rm -r! There is no Recycle Bin or Trash when using the terminal deletions are permanent.Helpful TipsCommand HistoryPress the up arrow to cycle through previously used commandsOn Mac/Linux, type history to see a list of recent commandsTab CompletionStart typing a file or directory name, then press TabThe terminal will attempt to complete it for youIf there are multiple options, press Tab twice to see all possibilitiesGetting HelpWindows: help command or command /?Mac/Linux: man command (manual pages, press q to exit)Examples:Clearing the ScreenWindows: clsMac/Linux: clear or Ctrl+LPower User CommandsSearching for FilesWindows: dir /s filenameMac/Linux: find . -name filenameSearching Within FilesWindows: findstr text filenameMac/Linux: grep text filenameChaining CommandsAll platforms: Use && to run commands in sequenceExample:Redirecting OutputAll platforms: Use > to send output to a fileExample:Next StepsAs you become more comfortable with these basic commands, you might want to explore:Command line text editors like Nano, Vim, or EmacsWriting simple shell scripts to automate tasksPackage managers like apt (Linux), Homebrew (Mac), or Chocolatey (Windows)Environment variables and how to set themSSH to connect to remote computersCommon Mistakes and TroubleshootingCommand not found: Check spelling or ensure the command is available on your systemPermission denied: You may need administrator/root privilegesWindows: Run Command Prompt as AdministratorMac/Linux: Use sudo before commands that need elevated privilegesNo such file or directory: Double-check path and file namesOperation not permitted: Similar to permission denied, you might need special permissionsTasksWindowsMac/LinuxCurrent locationcdpwdList filesdirlsChange directorycd dircd dirCreate directorymkdir dirmkdir dirCreate filetype nul > filetouch fileCopy filecopy source destinationcp source destinationMove/renamemove source destinationmv source destinationDelete filedel filerm fileDelete directoryrmdir /s dirrm -r dirClear screenclsclearGet helphelp commandman commandConclusionIn this tutorial, we have covered everything beginners need to know about using the terminal. We explored how to open the terminal across different operating systems, navigate file systems, create and manage files and directories, and use essential commands. We also learned helpful shortcuts, power user commands, and troubleshooting tips. With these foundational skills, you can now confidently use the command line as a powerful tool in your computing journey.Remember, the terminal is a powerful tool that rewards practice and experimentation. Dont be afraid to try new commands, but always be careful with commands that modify or delete files. Also,feel free to follow us onTwitterand dont forget to join our85k+ ML SubReddit. [Register Now] miniCON Virtual Conference on OPEN SOURCE AI: FREE REGISTRATION + Certificate of Attendance + 3 Hour Short Event (April 12, 9 am- 12 pm PST) + Hands on Workshop [Sponsored]The post The Complete Beginners Guide to Terminal/Command Prompt appeared first on MarkTechPost.
    0 Commentaires ·0 Parts ·6 Vue
  • This AI Paper from ByteDance Introduces a Hybrid Reward System Combining Reasoning Task Verifiers (RTV) and a Generative Reward Model (GenRM) to Mitigate Reward Hacking
    www.marktechpost.com
    Reinforcement Learning from Human Feedback (RLHF) is crucial for aligning LLMs with human values and preferences. Despite introducing non-RL alternatives like DPO, industry-leading models such as ChatGPT/GPT-4, Claude, and Gemini continue to rely on RL algorithms like PPO for policy optimization. Recent research focuses on algorithmic improvements, including eliminating critic models to reduce computational costs, filtering noisy samples during PPO sampling, and enhancing reward models to mitigate reward hacking problems. However, only a few studies focus on RLHF data construction (i.e., training prompts) and its performance scaling based on these training prompts.The success of RLHF heavily depends on reward model quality, which faces three challenges: mis-specified reward modeling in representing human preferences, incorrect and ambiguous preferences in training datasets, and poor generalization ability. To address these issues, GenRM was introduced to validate model predictions against ground-truth responses, showing good resistance to reward hacking and gaining adoption in advanced LLMs like DeepSeekV3. Methods like principled data selection that filter overly challenging instances during training and strategic selection methods identify key training prompts to achieve comparable performance with reduced data. Performance scale analysis reveals that RLHF shows superior generalization compared to SFT on novel inputs but significantly reduces output diversity.Researchers from ByteDance Seed address a critical gap in RLHF research where the role of prompt-data construction and its scalability has received less attention. They explore data-driven bottlenecks that limit RLHF performance scaling, focusing on reward hacking and decreasing response diversity challenges. A hybrid reward system is introduced by combining reasoning task verifiers (RTV) and a generative reward model (GenRM) that shows stronger resistance to reward hacking and enables a more accurate assessment of responses against ground-truth solutions. Moreover, a novel prompt-selection method called Pre-PPO is introduced to identify inherently challenging training prompts less susceptible to reward hacking.The experimental setup employs two pre-trained language models of different scales: a smaller model with 25B parameters and a larger model with 150B parameters. The training dataset contains one million prompts from diverse domains, including mathematics, coding, instruction-following, creative writing, and logical reasoning. Moreover, the researchers constructed a detailed evaluation framework covering multiple skill areas: logical reasoning, instruction-following, STEM tasks, coding, natural language processing, knowledge, contextual understanding, and out-of-distribution generalization. The evaluation framework includes two versions (V1.0 and V2.0) with overlapping prompts, though V2.0 features more challenging prompts.The experimental results show that the proposed approach combining Pre-PPO with prioritized mathematical and coding tasks consistently outperforms the baseline method across model sizes and evaluation datasets. The approach shows an improvement of +1.1 over the baseline when evaluated at 100-step intervals using TestSet V1.0. When tested on the more challenging TestSet V2.0, the performance improvement increases to +1.4. The most substantial gains appear in mathematics-intensive and coding tasks, with an improvement of +3.9 points in STEM and +3.2 points in coding. These improvements are attributed to the strategic prioritization of mathematical reasoning and coding tasks during early RLHF training phases.In conclusion, this paper addresses critical bottlenecks in RLHF data scaling, specifically identifying reward hacking and reduced response diversity as significant challenges. The researchers proposed a combined approach featuring strategic prompt construction and early-stage training prioritization to solve this issue. The method uses RTV and GenRM to combat reward hacking alongside the novel Pre-PPO prompt selection strategy that identifies and prioritizes challenging training prompts. Analysis reveals that RTV supervision shows the strongest resistance to reward hacking, followed by GenRM with ground-truth labels and then the BT Reward Model. The research establishes a foundation for optimizing RLHF data construction and developing more principle methods to reward hacking and model alignment.Check outthe Paper and GitHub Page.All credit for this research goes to the researchers of this project. Also,feel free to follow us onTwitterand dont forget to join our85k+ ML SubReddit. [Register Now] miniCON Virtual Conference on OPEN SOURCE AI: FREE REGISTRATION + Certificate of Attendance + 3 Hour Short Event (April 12, 9 am- 12 pm PST) + Hands on Workshop [Sponsored] Sajjad AnsariSajjad Ansari is a final year undergraduate from IIT Kharagpur. As a Tech enthusiast, he delves into the practical applications of AI with a focus on understanding the impact of AI technologies and their real-world implications. He aims to articulate complex AI concepts in a clear and accessible manner.Sajjad Ansarihttps://www.marktechpost.com/author/sajjadansari/VideoMind: A Role-Based Agent for Temporal-Grounded Video UnderstandingSajjad Ansarihttps://www.marktechpost.com/author/sajjadansari/PilotANN: A Hybrid CPU-GPU System For Graph-based ANNSSajjad Ansarihttps://www.marktechpost.com/author/sajjadansari/This AI Paper Propose the UI-R1 Framework that Extends Rule-based Reinforcement Learning to GUI Action Prediction TasksSajjad Ansarihttps://www.marktechpost.com/author/sajjadansari/TokenBridge: Bridging The Gap Between Continuous and Discrete Token Representations In Visual Generation
    0 Commentaires ·0 Parts ·7 Vue