Apple finally admits next-gen CarPlay isnt coming in 2024Apple finally admits next-gen CarPlay isnt coming in 2024 / An updated launch timeframe hasnt been announced, but Apple says its definitely coming.By Jess Weatherbed, a news writer focused on creative industries, computing, and internet culture. Jess started her career at TechRadar, covering news and hardware reviews. Jan 24, 2025, 10:21 AM UTCShare this storyBetter late than never. Image: AppleNow that weve almost hit the second month of 2025, Apple has finally updated its CarPlay webpage to say it wont be arriving in 2024. As noted by MacRumors, the 2024 timeframe for the first vehicles to support CarPlay 2 has been removed from the website without adding a new launch date.Apple first announced the next generation of CarPlay back in 2022, but updates about its arrival have been sporadic. Porsche and Aston Martin havent provided any launch dates despite saying their cars would be the first to get the new CarPlay. Some automakers like Ford and Mercedes were slow to confirm support, while others like General Motors and Rivian have snubbed CarPlay entirely in favor of having more control over their vehicles.Despite not posting a revised date, there are indications that Apple will launch it eventually. There are references to next-generation CarPlay in the iOS 18.3 beta released last month, for example, and Apple has recently filed new images of it in an EU database. Apple also told 9to5Mac that its working closely with several automakers that will implement the new CarPlay experience.Each car brand will share more details as they near the announcements of their models that will support the next generation of CarPlay, Apple told the outlet.Most PopularMost Popular

Inside DeepSeek-R1: The Amazing Model that Matches GPT-o1 on Reasoning at a Fraction of the Cost 0 like January 23, 2025Share this postLast Updated on January 24, 2025 by Editorial TeamAuthor(s): Jesus Rodriguez Originally published on Towards AI. Created Using MidjourneyI recently started an AI-focused educational newsletter, that already has over 175,000 subscribers. TheSequence is a no-BS (meaning no hype, no news, etc) ML-oriented newsletter that takes 5 minutes to read. The goal is to keep you up to date with machine learning projects, research papers, and concepts. Please give it a try by subscribing below:TheSequence | Jesus Rodriguez | SubstackThe best source to stay up-to-date with the developments in the machine learning, artificial intelligence, and datathesequence.substack.comQuite often we see releases in generative AI that truly challenges peoples imagination. This is the DeepSeek-R1, the newest model by the famous Chinese eval lab that dabbles into reasoning. One of the dominat reasoning thesis in the market is that its an emerging property of the scaling laws. In other words, you need big models to get reasoning. DeepSeek-R1 challenges that thesis achieving reasoning by leveraging a very clever post-training process. The model is able to match the performance of GPT-o1 at a fraction of the compute cost. Quite amazing.Image Credit: DeepSeekLets dive in:Introduction to DeepSeek-R1 and its MotivationThe field of Large Language Models (LLMs) has seen remarkable progress, yet achieving robust reasoning capabilities remains a significant challenge. Many models rely on extensive supervised fine-tuning (SFT), which can be computationally expensive and may not fully unlock a models potential for self-improvement. DeepSeek-R1 and its precursor, DeepSeek-R1-Zero, represent a departure from this paradigm, exploring the power of reinforcement learning (RL) to develop and enhance reasoning capabilities in LLMs. This essay will delve into the technical details of the DeepSeek-R1 architecture and training process, highlighting key innovations and contributions.DeepSeek-R1s development was driven by the goal of exploring the potential of LLMs to develop reasoning skills without relying on a foundation of supervised data. The research began with the idea of pure RL to allow the model to self-evolve. This approach resulted in DeepSeek-R1-Zero, a model that demonstrated the possibility of incentivizing reasoning capabilities purely through RL. DeepSeek-R1 was created to address the issues of poor readability and language mixing observed in DeepSeek-R1-Zero, while further improving reasoning performance. DeepSeek-R1 incorporates multi-stage training and a cold-start data approach before RL. The goal of the DeepSeek project is to create better models and share them with the research community.DeepSeek-R1-Zero: A Pure Reinforcement Learning ApproachDeepSeek-R1-Zero stands out as a model trained via large-scale reinforcement learning (RL) without any prior supervised fine-tuning (SFT). This approach aimed to explore the models capacity for self-evolution in reasoning.Reinforcement Learning Algorithm: DeepSeek-R1-Zero utilizes Group Relative Policy Optimization (GRPO). GRPO is a cost-effective RL method that omits the use of a critic model and instead estimates the baseline from group scores. Given a question q, GRPO samples a group of outputs from the old policy and optimizes the policy by maximizing a defined objective function. The objective function includes an advantage term, calculated using the rewards of the groups output and a Kullback-Leibler (KL) divergence term which ensures that the policy updates are not too large.Reward Modeling: The reward system for DeepSeek-R1-Zero focuses on accuracy and format.Accuracy rewards evaluate the correctness of responses. For example, math problems require a precise answer in a specific format, which allows for rule-based verification.Format rewards enforce the inclusion of the models thinking process within <think> and </think> tags.Notably, DeepSeek-R1-Zero does not use process-based or outcome neural reward models.Training Template: A simple template guides the model to produce a reasoning process followed by the final answer. The template is designed to be free of content-specific biases to observe the models natural progression during RL.Image Credit: Hugging FaceDeepSeek-R1-Zero showed significant improvements on the AIME 2024 benchmark during training, going from 15.6% to 71.0% pass@1, which is comparable to OpenAIs o10912. With majority voting, its score further improved to 86.7%. The model also demonstrated self-evolution by increasing its thinking time (response length) as training progressed, enabling more sophisticated problem-solving strategies such as reflection and exploration of alternative approaches. The model also exhibited an aha moment, where it learned to rethink its initial approach by allocating more thinking time.Image Credit: DeepSeekDeepSeek-R1: Incorporating Cold Start Data and Multi-Stage TrainingWhile DeepSeek-R1-Zero demonstrated the potential of pure RL, it suffered from issues such as poor readability and language mixing. DeepSeek-R1 was developed to address these issues and to further enhance performance through a multi-stage training pipeline that incorporates a small amount of cold-start data.Cold Start Data: DeepSeek-R1 is fine-tuned on thousands of long Chain-of-Thought (CoT) examples before RL training, which acts as the cold start. These examples are collected using methods such as few-shot prompting with long CoTs, directly prompting models for detailed answers with reflection and verification, refining DeepSeek-R1-Zeros outputs and post-processing by human annotators. This cold start data helps address readability by using a readable output format that includes a summary at the end of each response and filters out responses that are not user-friendly.The output format is defined as |special_token|<reasoning_process>|special_token|, with the reasoning process being the CoT for the query and the summary summarizing the reasoning results.Reasoning-Oriented Reinforcement Learning: After fine-tuning on cold-start data, DeepSeek-R1 undergoes the same large-scale RL training as DeepSeek-R1-Zero. This phase focuses on enhancing reasoning capabilities for coding, math, science, and logic reasoning tasks. A language consistency reward was introduced to mitigate language mixing during RL training, though ablation experiments show that the reward results in a small performance degradation.Rejection Sampling and Supervised Fine-Tuning: Upon reaching convergence in the reasoning-oriented RL process, SFT data is generated via rejection sampling using the RL checkpoint, combined with supervised data from DeepSeek-V3 in areas such as writing and factual QA. Data is expanded beyond rule-based reward evaluation by incorporating a generative reward model using DeepSeek-V3 to judge ground-truth and model predictions. Non-reasoning data was also included from DeepSeek-V3 to enhance the models general capabilities.Reinforcement Learning for All Scenarios: A second RL stage aligns the model with human preferences, focusing on helpfulness and harmlessness. Rule-based rewards are used for reasoning data, while reward models capture preferences in general data.Distillation and EvaluationDeepSeek-R1s reasoning capabilities were also transferred to smaller, more efficient models through distillation.Distillation Process: Open-source models like Qwen and Llama were directly fine-tuned using the 800k samples from DeepSeek-R1. This approach is effective in improving the reasoning abilities of smaller models. The base models used include Qwen2.5-Math-1.5B, Qwen2.5-Math-7B, Qwen2.514B, Qwen2.532B, Llama-3.18B, and Llama-3.370B-Instruct. Only SFT is applied to the distilled models, with no RL stage.Evaluation Metrics and Benchmarks: Models are evaluated on a range of benchmarks including MMLU, MMLU-Redux, MMLU-Pro, C-Eval, CMMLU, IFEval, FRAMES, GPQA Diamond, SimpleQA, SWE-Bench Verified, Aider, LiveCodeBench, Codeforces, Chinese National High School Mathematics Olympiad (CNMO 2024) and American Invitational Mathematics Examination 2024 (AIME 2024). Additionally, open-ended generation tasks are judged using LLMs, specifically AlpacaEval 2.0 and Arena-Hard. Evaluation prompts follow the setup in DeepSeek-V3, using the simple-evals framework, or their original protocols.Key Findings: DeepSeek-R1 achieves performance comparable to OpenAI-o11217 on a range of tasks. It shows superior performance in STEM-related questions compared to DeepSeek-V3, demonstrating the effectiveness of large-scale reinforcement learning. DeepSeek-R1 also shows strong document analysis capabilities as well as fact-based query abilities. The model also excels in writing tasks and open-domain question answering. On math tasks, DeepSeek-R1 is comparable to OpenAI-o11217. The distilled models show significant improvements, with DeepSeek-R17B outperforming GPT-4o-0513. Furthermore, DeepSeek-R114B surpassed QwQ-32B-Preview on all metrics. The distilled 32B and 70B models significantly outperformed o1-mini on most benchmarks, highlighting the effectiveness of distillation.Key Contributions, Discussion, and Future DirectionsDeepSeek-R1s development highlights several key contributions:Pure RL for Reasoning: It validates that reasoning capabilities in LLMs can be incentivized purely through RL, without the need for SFT.Effective Multi-Stage RL Training Pipeline: The approach combines two RL and two SFT stages to improve reasoning patterns and align with human preferences.Distillation of Reasoning: DeepSeek-R1 demonstrates that reasoning patterns from larger models can be distilled into smaller ones, yielding improved performance.The R1 paper also discusses some unsuccessful attempts including Process Reward Model (PRM) and Monte Carlo Tree Search (MCTS).Process Reward Model (PRM) was found to have limitations, such as difficulty in defining fine-grained steps in general reasoning, challenging evaluation of intermediate steps, and reward hacking.Monte Carlo Tree Search (MCTS) encountered difficulties due to the exponentially large search space and the challenge of training a fine-grained value model for token generation.Future research directions include:General Capability Enhancement: Expanding DeepSeek-R1s abilities in function calling, multi-turn interactions, complex role-playing, and JSON output.Language Mixing Mitigation: Addressing language mixing issues when handling queries in languages other than English and Chinese.Prompt Engineering: Improving the models robustness to variations in prompts, moving beyond its sensitivity to few-shot prompting.Software Engineering Tasks: Expanding RL to software engineering tasks by implementing rejection sampling or asynchronous evaluations to improve efficiency.ConclusionDeepSeek-R1 represents a significant advancement in the development of LLMs with enhanced reasoning capabilities. By employing innovative reinforcement learning techniques, a multi-stage training pipeline, and effective distillation methods, DeepSeek-R1 not only achieves impressive performance but also offers valuable insights into the potential for self-evolution and knowledge transfer in AI. The open-sourcing of DeepSeek-R1 and its distilled models will significantly contribute to the research community, enabling further advancements in this field.Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming asponsor. Published via Towards AITowards AI - Medium Share this post

10 FAQs on AI Agents: Decoding Googles Whitepaper in Simple Terms 0 like January 23, 2025Share this postLast Updated on January 24, 2025 by Editorial TeamAuthor(s): Kshitij Darwhekar Originally published on Towards AI. This member-only story is on us. Upgrade to access all of Medium.This article is part of a new series Im launching called 10 FAQs. In this series, I aim to break down complex concepts by answering the ten most common questions youre likely to have on the topic. My goal is to use simple language and relatable analogies to make these ideas easy to grasp.Kshitij DarwhekarView list1 storyPhoto by Solen Feyissa on UnsplashDont have a paid Medium membership (yet)? You can read the entire article for free by clicking here with my friends link.Youtube VideoGoogle, in September 2024, published a paper titled Agents by Julia Wiesinger, Patrick Marlow, and Vladimir Vuskovic. Recently, this paper went viral on Twitter. I read through the entire paper (so you dont have to) and answered ten key questions to help you understand AI agents in depth. This single article is all you need to get started and excited about AI agents.Generative AI agents can be defined as applications that attempt to achieve a goal by observing the world & acting upon it using the tools they have at their disposal. Agents are autonomous & can act independently of human intervention. You can understand it Read the full blog for free on Medium.Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming asponsor. Published via Towards AITowards AI - Medium Share this post

From his very first movie The Witch, Robert Eggers taught us a simple but important equation: Robert Eggers + Monster + Archaic Spelling Conventions = Good Movie. So while weve loved The Lighthouse, The Northman, and Nosferatu, were even more excited about his announced next project: Werwulf.More than just a stylistic flourish, the title of Eggers latest, rooted in the Old English spelling of werewolf, indicates the approach hell likely take. As when he borrowed from Puritan sermons, court documents, and journals for The Witch, and the ur-Hamlet poem with Norse roots in The Northman, Eggers will probably look at early myths about lycanthropy for his film, finding something fundamental in the monster that predates 1935s The Werewolf of London and 1941s The Wolf Man. At present, we know that Eggers has re-teamed with Sjn, co-writer of The Northman, and that the movie will apparently be set in the 13th century. For those who want a better idea of the duos inspiration, here are some prominent stories, myths, and alleged real incidents from the medieval and early modern periods.BisclavretWhile Eggers certainly loves horror and the macabre, he rarely goes for straightforward scares. For that reason, he may find special inspiration in Bisclavret, one of the Twelve Lais (short poetic narratives) by Marie de France, a French poet from the late 12th century.Bisclavret begins with elements familiar to anyone who has heard an ancient wolfman story before. Every week, a beloved Baron named Bisclavret disappears for three days, only to return without explanation. At his wifes urging, Bisclavret (whose name literally means werewolf, which should have been a big giveaway) admits that he transforms into a wolf on those absent evenings.Unsurprisingly, the confession disgusts his wife and she looks for a way to get rid of him. Her solution comes in the form of a knight who has long held a flame for her and from a caveat that Bisclavret revealed: he must return to his clothes to revert into human form. The wife orders the knight to steal Bisclavrets clothes, forcing him to stay in wolf form. With the husband gone, the wife marries the knight.One year later, the king goes on a hunting trip when he encounters a wolf of remarkable gentleness, who nuzzles up to the ruler and expresses its admiration. The king does not recognize the wolf as Bisclavret but admire the beasts inherent nobility all the same. He brings the creature to his castle. But soon that gentleness turns to savagery when the knight who aided his wife comes to a celebration. The wolf attacks him and does worse to the wife, ripping off her nose.Although the king initially thinks that hes misjudged the wolf, an advisor points out that the beast only showed such hostility to the knight and his wife. The advisor determines that the wolf must be Bisclavret (he understands names, apparently) and after some medieval torture, the wife confesses. Once the wolf has been given clothes, and a private space to transform, it reverts to Bisclavret once more. Human again, Bisclavret gets his land and standing restored while the wife and the knight are banished, cursed to sire children all born without noses.Its easy to see how Eggers could render the scenes of shocking violence, and especially the noseless children of the wife and the knight. But more compelling is the storys play with concepts of loyalty, from the duty that Bisclavret shows to his king to the infidelity of the wife and her faithless knight.The Account of Gerald of WalesDo not be afraid! Do not fear! Do not worry! There is nothing to fear! Those arent words we often hear in werewolf stories, especially coming from the mouth (snout?) of the wolf itself. But thats exactly what happens in an account of Giraldus Cambrensis, or Gerald of Wales in its anglicized form. Gerald was a 12th century priest and historian who described the geography and animals of Ireland.In TopographiaHibernica, an 1187 English study of Ireland, Gerald passed down to us the story of a priest and his servant who made camp during a journey through the Irish wilderness. Recounted as a real historical incident, the duo were startled on one fateful evening by the appearance of a wolf at their campfire. They were even more disturbed when the wolf began speaking to them, assuring them that they have no reason for fear.The wolf explained that his people in the region of Ossory suffer under an ancient curse, which causes one man and one woman from the community to transform into a wolf every seven years and live out their days in furry exile. The wolf approached the priest because his companion was dying and needed last rites performed. The priest agrees to come with the wolf, expecting to find another human being. Instead he finds a dying female wolf who also seems to speak with a human tongue.Join our mailing listGet the best of Den of Geek delivered right to your inbox!Despite the human sounds coming from the she-wolf, the priest could not deliver the rites until the male wolf pulled back the fur on her paw. There, the priest saw a human hand, which was enough to compel him to complete the rites and allow the female wolf to die in peace. To show his gratitude, the male wolf revealed the best route through the forest.On one hand, the wolf in Geralds account goes out of its way to avoid scaring the priest, which makes it an unlikely monster for a horror movie and maybe not the best choice for a Robert Eggers film. On the other, the priest is scared enough for everyone, in part because of his religious superstition. This spiritual dread, combined with the opportunity to depict rituals, might appeal to Eggers enough to overcome a potentially inert central monster.Otia Imperialia of Gervase of TilburyWritten by 13th century lawyer and cleric Gervase of Tilbury, the Otia Imperialia was an index of incredible places, people, and events throughout Europe. Those people include a couple of werewolves, which Gervases pen hastens to assure his reader (Holy Roman Emperor Otto IV, to be specific) truly exist.By way of evidence, Gervasea Norman son of an English knight and a one-time member of the court of King Henry II in Englanddescribes the plight of disgraced knight named Raimbaud de Pouget. After being rejected by his lord, de Pouget goes into the woods and loses his mind, leading to his transformation into a wolf. He stays in this condition until a woodsman cuts off his paw, which changes him back into a human, so he can go into town and testify about the experience.Gervase also describes Chaucevaire, a man who regularly leaves his friends and families to go into the woods. In an echo of Bisclavret, Chaucevaire ditches his clothes and takes wolf form, traipsing about the forest until he can return to his human shape. For Gervase, the fact that Chaucevaire can explain secrets about a wolfs behavior is reason enough to believe his claims. If it all sounds incredulous, know that Gervase wrote down for the Holy Roman emperor that in England we have often seen men change into wolves.Eggers may very well draw from these accounts, or better the pseudo-scientific way that Gervaase of Tilbury talks about them, to inform Werwulf. But the vignettes themselves are too short to offer much inspiration for adaptation beyond a single scene or momentsuch as a severed wolfs hand reverting to human form before horrified eyes.Peter Stubbe, the Werewolf of BedburgOkay, the pamphlet telling the story of Peter Stubbe was published in London in 1590, long after the 13th century setting that Eggers has in mind for Werwulf. Furthermore, it is derived from the Germanic werewolf panic that swept through central Europe between the 15th and 17th centuries. However, it contains so many of the themes and fixations found in Eggers work that it is difficult to believe he wont at least crib a few ideas from this relatively more recent era of werewolf lore.Peter Stubbe (also spelled Stumpp in other less detailed accounts) was a man who lived near the 16th century village of Bedburg in what is modern day Germany. He apparently got so intoxicated with evil that he made a deal with the Devil to commit more. The Devil, always accommodating in these types of matters, gave Stubbe a magical belt which when worn would transform him to have the likeness of a greedy, devouring wolf, strong and mighty, with eyes great and large, which in the night sparkled like fire, a mouth great and wide, with most sharp and cruel teeth, a huge body, and mighty paws.The pamphlet operated like a 16th century version of a True Crime podcast, pinning responsibility for numerous murders and deaths on Stubbe over a 25-year period. It leaned into gory details about Stubbes evil, too, including his tendency to eat his victims. After he was accused of werewolfery and tortured, Stubbe apparently confessed his wicked deeds, his taste for human flesh, and even to committing incest with his daughter. He also claimed to have lain with a succubus sent from Hell for his pleasure. He also confessed to eating goats, neighbors livestock, women, and more than a dozen children, including his own son whom he claimed to devour after turning into a wolf before the boy. He was executed on Oct. 31, 1589, alongside his daughter and his mistress. All were tortured in ghastly fashion before execution (think the flayed man in Game of Thrones), although Stubbe bore the worst of it before finally being executed and having his severed head placed atop the body of a wolf and hung from a pole in a town square. It was erected to warn against future werewolves (modern historians believe the real Stubbe was probably a serial killer).Again, Stubbes story occurs several centuries after Werwulfs setting. But given Eggers interest in social dynamics, fear of the unknown, and horror, its hard to believe the director wont draw at least a little from the gruesome confession and the monstrous execution.Werwulf is slated for release on Dec. 25, 2026.

The post Check Out The First Berserker: Khazans Demo Trailer appeared first on Xbox Wire.

Apple commentator John Gruber yesterday described Siris current performance as an unfunny joke, giving its inability to correctly name the winner of Super Bowl 13 an example, noting that this is a basic query that any US chatbot ought to be able to answer.It turns out that wasnt an entirely random example: it was prompted by his friend Paul Kafasis, who decided to test Siri on Super Bowl 1 to 60 inclusive and the results were not good Kafasis shared the results in a blog post.So, how did Siri do? With the absolute most charitable interpretation, Siri correctly provided the winner of just 20 of the 58 Super Bowls that have been played. Thats an absolutely abysmal 34% completion percentage. If Siri were a quarterback, it would be drummed out of the NFL.Siri did once manage to get four years in a row correct (Super Bowls IX through XII), but only if we give it credit for providing the right answer for the wrong reason. More realistically, it thrice correctly answered three in a row (Super Bowls V through VII, XXXV through XXVII, and LVII through LIX). At its worst, it got an amazing 15 in a row wrong (Super Bowls XVII through XXXII).Siris a big Eagles fan, it seems.Most amusingly, it credited the Philadelphia Eagles with an astonishing33 Super Bowl winsthey havent earned, to go with the one 1 they have.The right answer for the wrong reason part refers to Siri being asked to name the winner of Super Bowl X. For unknown reasons, Siri decided to respond with a lengthy reply about Super Bowl IX, and coincidentally the winner was the same both times.Sometimes Siri went completely off-piste and completely ignored the question, quoting unrelated Wikipedia entries.Who won Super Bowl 23?Bill Belichick owns the record for the most Super Bowl wins (eight) and appearances (twelve: nine times as head coach, once as assistant head coach, and twice as defensive coordinator) by an individual.But maybe the Roman numerals cause confusion, and other AI systems struggle just as much? Gruber decided to carry out a few spot checks.I havent run a comprehensive test from Super Bowls 1 through 60 because Im lazy, but a spot-check of a few random numbers in that range indicates that every other ask-a-question-get-an-answer agent I personally use gets them all correct.I tried ChatGPT, Kagi, DuckDuckGo, and Google. Those four all even fare well on the arguably trick questions regarding the winners of Super Bowls 59 and 60, which havent yet been played. E.g., asked the winner of Super Bowl 59, Kagis Quick Answerstarts: Super Bowl 59 is scheduled to take place on February 9, 2025. As of now, the game has not yet occurred, so there is no winner to report.Super Bowl winners arent some obscure topic, like, say, asking Who won the 2004 North Dakota high school boys state basketball championship?a question I just completely pulled out of my ass, but which, amazingly,Kagi answered correctlyfor Class A, andChatGPT answered correctlyforboth Class A and Class B, and provided a link tothis video of the Class A championship game on YouTube.Thats amazing! I picked an obscure state (no offense to Dakotans, North or South), a year pretty far in the past, and the high school sport that I personally played best and care most about. And both Kagi and ChatGPT got it right. (Id give Kagi an A, and ChatGPT an A+ for naming the champions of both classes, and extra credit atop the A+ for the YouTube links.)Gruber notes that the old Siri on macOS 15.1.1 actually does better. Sure, it seems less capable, as it gave its classic Heres what I found on the web response, but at least that gives links to the correct answer. New Siri doesnt.New Siripowered by Apple Intelligence with ChatGPT integration enabledgets the answer completely but plausibly wrong, which is theworstway to get it wrong. Its alsoinconsistentlywrongI tried the same question four times, and got a different answer, all of them wrong, each time. Its a complete failure.Photo byCaleb WoodsonUnsplashAdd 9to5Mac to your Google News feed. FTC: We use income earning auto affiliate links. More.Youre reading 9to5Mac experts who break news about Apple and its surrounding ecosystem, day after day. Be sure to check out our homepage for all the latest news, and follow 9to5Mac on Twitter, Facebook, and LinkedIn to stay in the loop. Dont know where to start? Check out our exclusive stories, reviews, how-tos, and subscribe to our YouTube channel

The modern workplace has undergone a seismic transformation over recent years, with hybrid work becoming the norm and businesses rapidly adopting cloud-based Software-as-a-Service (SaaS) applications to facilitate it. SaaS applications like Microsoft 365 and Google Workspace have now become the backbone of business operations, enabling seamless collaboration and productivity. However, this dependence on SaaS solutions has also attracted a surge in cyberthreats, exposing critical business data to risks like ransomware and phishing attacks. Amid these challenges, the 2025 State of SaaS Backup and Recovery Report offers a timely analysis of the SaaS data protection landscape. By surveying over 3,700 IT professionals from diverse industries and company sizes, this report highlights trends, challenges and solutions shaping the future of SaaS data resilience. For organizations seeking clarity in navigating these turbulent waters, the findings provide invaluable guidance.Key revelations from the report are both eye-opening and concerning. For instance, 87% of IT professionals reported experiencing SaaS data loss in 2024, with malicious deletions as the leading cause. Moreover, while 61% of applications and workloads are expected to run on public cloud platforms in the next two years, only 14% of IT leaders feel confident they can recover critical SaaS data within minutes following an incident. These findings underscore an urgent call to action for businesses to strengthen their data resilience strategies.In this blog, we'll uncover the survey's key findings, revealing how organizations are adapting to meet the ongoing challenges. From revealing the biggest threats to understanding the strategies IT leaders are leveraging to stay ahead, you'll gain a clear picture of the trends shaping SaaS data protection.The shift toward cloud: Major trends reshaping modern IT operationsOrganizations worldwide are increasingly relying on hybrid cloud environments, with 54% of workloads and applications already cloud-hosted. This trend shows no signs of slowing down, as businesses anticipate growing this figure to 61% by 2026. The acceleration of cloud adoption reflects the critical role cloud solutions play in driving digital transformation and enabling organizations to scale effectively.What's driving cloud adoption?The benefits propelling rapid cloud migration are clear: leveraging cloud platforms allows organizations to enhance operational efficiency, optimize strategic analytics and support real-time decision-making. However, this enthusiasm comes with a caveat companies are carefully navigating concerns about data sensitivity, security and compliance as they transition.Data types moving to the cloudCertain types of data dominate cloud migration trends, illustrating where organizations feel confident leveraging the cloud's potential. The top candidates for data migration to the cloud include:Non-sensitive analytics data (39%): Leading the pack, these datasets highlight the cloud's role in powering strategic insights.The Internet of Things (IoT) and edge data (34%): The inclusion of high-velocity datasets reflects growing trust in the cloud's ability to handle real-time analytics at scale.Sales and orders data (34%): Organizations are increasingly turning to cloud-hosted solutions to drive operational efficiency in sales processes.Data that's staying on-premisesDespite the allure of the cloud, businesses are cautious about entrusting their most sensitive information to third-party environments. The top candidates for data to remain on-premises include:Personally identifiable information (PII) and protected health information (PHI) (42%): Strict regulations and concerns about breaches keep this data largely on-premises.Corporate financial data (42%): Companies remain wary of potential risks in the cloud for critical financial records.Sensitive intellectual property (40%): Maintaining tight control over proprietary assets remains a top priority.Top use cases for public cloud adoptionCloud adoption is driven by practical needs that align with modern business challenges:Collaboration (39%): The flexibility and scalability of cloud solutions are essential for supporting hybrid and remote work environments.Disaster recovery (37%): Ensuring fast recovery and protecting data from loss or downtime remain critical priorities.Data warehousing and Database-as-a-Service (32%): Organizations value the cloud's ability to manage large datasets and enable seamless data-driven operations.SaaS application trends in 2024The adoption of SaaS applications continues to transform how businesses collaborate, manage operations and scale their services. Collaboration platforms remain at the forefront of SaaS adoption, reflecting their critical role in today's hybrid and remote work environments.Microsoft 365 retains leadership while Google Workspace gains momentumMicrosoft 365 maintains its position as the leader in SaaS collaboration solutions, with a 53% adoption rate among survey respondents. However, this represents a decline from its 71% adoption rate in 2022, indicating a potential shift in preferences or increasing competition in the SaaS space. Meanwhile, Google Workspace has seen a consistent rise in adoption, climbing to 35% in 2024 compared to 25% in 2022. This growth is particularly pronounced among SMBs, with 38% adoption compared to 32% among enterprises. Diverging SaaS adoption trends between SMBs and enterprisesThe survey also highlighted distinct preferences between SMBs and enterprise organizations when it comes to SaaS tools:SMBs favor applications that simplify everyday operations and finance management, such as:Google Workspace: 38% of SMBs vs. 32% of enterprises.Dropbox: 26% of SMBs vs. 20% of enterprises.Intuit QuickBooks: 20% of SMBs vs. 17% of enterprises.Enterprises lean towards tools that support large-scale operations and customer engagement, such as:Microsoft Dynamics: 32% of enterprises vs. 28% of SMBs.Salesforce: 28% of enterprises vs. 22% of SMBs.HubSpot: 24% of enterprises vs. 20% of SMBs.What are the major barriers to cloud adoption?Despite the widespread adoption of cloud solutions, many organizations face significant hurdles in migrating workloads and data to the cloud.Cloud cost optimization (24%): Managing expenses during and after migration remains the top concern for organizations.Compatibility and performance issues (20%): Ensuring workloads function seamlessly post-migration is another significant hurdle that causes delays and disruptions.Some organizations also face difficulties with provider selection and feasibility (15%), post-migration management (14%), right-sizing cloud instances (8%), licensing complexities (7%) and managing application dependencies (5%).The gaps IT pros can't ignore: Why confidence in recovery is lowBackup strategies are essential to securing critical SaaS data, yet IT professionals' confidence in these systems remains alarmingly low.Backup strategy adoption across SaaS platformsOrganizations leveraging SaaS applications report varying levels of backup strategy implementation:Microsoft 365: 70% have a backup strategy in place, the highest among SaaS platforms.Google Workspace: 66% of users report having a backup plan, reflecting strong adoption.Salesforce: Only 53% have a dedicated backup strategy, signaling a critical vulnerability.Low confidence in backup system effectivenessDespite the presence of backup strategies, only 40% of IT professionals expressed confidence in their systems' ability to protect critical data during a crisis. This hesitancy is fueled by:Outdated backup solutions: Over 28% of respondents indicated their backup systems haven't evolved in five years, leaving them ill-equipped to handle modern-day threats.Dissatisfaction with existing solutions: Nearly 30% of IT pros believe their backup and recovery tools fall short of what their organization needs.Fewer than 10% of respondents feel their organization's approach to backup and disaster recovery is sufficient without requiring any changes. What are the challenges in backup management?Managing backups for SaaS applications presents unique challenges, varying by platform and user cohort. The survey reveals distinct pain points for users of Microsoft 365, Google Workspace and Salesforce:Data recovery issues: Google Workspace (23%) and Salesforce (23%) users report the highest rates of difficulty with data recovery, compared to 20% of Microsoft 365 users.Alerting and reporting: Google Workspace users (11%) are most challenged in setting up and managing alerts, surpassing Microsoft 365 (8%) and Salesforce (8%) users.Compliance maintenance: Salesforce users (24%) struggle most with maintaining compliance, followed by Google Workspace (23%) and Microsoft 365 (21%).The growing time burden of backup managementBackup management has become increasingly time-intensive for IT teams:Over 50% of respondents spend more than two hours daily equating to over 10 hours per week on monitoring, managing and troubleshooting backups.The cohort spending less than one hour daily has dropped sharply, from 39% in 2022 to 23% in 2024, while those dedicating three or more hours daily have grown from 5% in 2022 to 14% in 2024.How organizations are securing backup infrastructureThe majority of organizations report having policies and controls in place to secure access to their backups across key areas, including public cloud (77%), servers or virtual machines (76%), SaaS applications (74%) and endpoints/PCs (73%).While these numbers reflect a proactive approach, around 25% of organizations still lack policies and controls for backup security, presenting vulnerabilities in an increasingly hybrid and multicloud environment.What are the major causes of SaaS data loss?SaaS data loss remains a significant challenge, with only 13% of organizations reporting no data loss incidents in the past year. Key causes include:Malicious deletion: More than 50% of organizations suffered data loss from malicious deletion, 29% were impacted by external threats and 27% by insider actions.Accidental deletion: 34% of respondents experienced data loss due to human error.Misconfigurations: 30% of organizations faced data loss due to improper setup or maintenance.Integration issues: Conflicts with third-party applications affected 30% of respondents.Technical errors: 18% experienced scripting errors, while 14% faced sync issues.The race to recovery: How quickly can organizations restore lost SaaS data?The ability to swiftly recover lost SaaS data is critical for minimizing downtime and avoiding costly compliance breaches. Yet, recovery times vary widely:Only 14% of organizations surveyed can recover within minutes, ensuring minimal disruption.Just over 40% achieve recovery within hours, meeting operational and regulatory demands.About 35% require days or weeks, risking prolonged downtime and potential non-compliance.Alarmingly, 8% are unsure of recovery times, and 2% cannot recover lost data at all.Recovery of SaaS data objectsThe frequency of recovering SaaS data objects is as follows:Most frequently recovered:Email (20%) and mail contacts (17%) are restored daily, emphasizing their critical role in communication and business continuity.Least frequently recovered:Calendar objects (15%) and messaging app data (16%) see fewer recovery incidents, indicating lower frequency of loss or reduced immediate operational impact.Key takeaways and recommendations from the 2025 State of SaaS Backup and Recovery ReportThe 2025 State of SaaS Backup and Recovery Report paints a vivid picture of the evolving SaaS data protection landscape. Cloud adoption continues to surge among organizations, driven by the operational efficiency and scalability of SaaS platforms. Yet, alongside these advancements lie significant challenges. From tackling data threats to addressing the complexities of backup management, the findings highlight an urgent need for businesses to rethink and modernize their data protection strategies. On that front, businesses must adopt a comprehensive, scalable backup strategy that aligns with the complexities of today's hybrid and multicloud environments. Such a strategy should include robust data protection and rapid recovery capabilities to address SaaS data threats like accidental deletions and ransomware attacks. By investing in the right tools and practices, organizations can enhance their data resilience, minimize downtime and confidently meet the demands of an increasingly cloud-driven future.For deeper insights into the SaaS world and actionable recommendations, download the full 2025 State of SaaS Backup and Recovery Report.Found this article interesting? This article is a contributed piece from one of our valued partners. Follow us on Twitter and LinkedIn to read more exclusive content we post.

With the installation of new leadership at the FCC, heres what network owners and operators are likely to see unfold. And it may not be what you expect.

About FilestageFilestage frees people from chaotic approval processes, making work more joyful and productive. From large enterprises to independent agencies, our review and approval platform helps teams share, discuss, and approve all their files, all in one place including documents, designs, images, videos, and audio files.We're a fully remote team with people working in home offices, co-working spaces, and coffee shops all over the world. Together, we're on a mission to create a seamless approval process that helps people deliver their best work.We've raised our Series A and have over half a million users across 500+ companies, including AB InBev, LG, Havas, GroupM, and Emirates. So if you're looking for a fast-growing startup in a booming market, you've found it!What youll be working on as our Customer Success Manager (DACH)This is your chance to help shape the future of our remote company. We believe that a strong Customer Success Team is the backbone of healthy business growth.You build a good relationship with your customers. You work closely with your customers, build relationships and make them happy users who will gladly recommend Filestage to others.You turn customers into experienced users. You help customers get the most out of Filestage software in training sessions via chat, e-mail, phone, webinars and video conferencing. Don't worry, our customers are friendly and don't bite ;-)You proactively look for ways to make our customers even more successful. With your contribution to customer success, you improve the lives of creative people and project managers around the globe. You are committed to your clients and help them find pragmatic solutions and improve their workflow.You drive the growth of Filestage. You always have your feelers out and find opportunities to grow and expand in your clients' companies and to further distribute Filestage software.Life at FilestageWe believe people are more productive when they can choose their own schedule. So were proud to offer fully-remote roles that give you the perfect balance between work and life.Here are some of the benefits you can look forward to at Filestage:Work from where youre happiest and enjoy a flexible schedule. Weve been fully remote from the start, giving you the opportunity to meet people all over the world and broaden your horizons.For this role, were looking for someone based in Europeto make sure theres a high time overlap with our team members, customers, and partners.Meet up in real life. We all travel together at least once a year (in normal times) at our team retreat to have fun and get to know each other.Enjoy a strong team culture. Were a group of knowledge seekers, reflective thinkers, clear communicators, goal owners, problem solvers, and team players. These are the values we strive for to help us achieve our mission.Join a happy team. Weve been rated five stars on Glassdoor by our lovely team. You can take a look at our reviews here.Create a workspace that suits you. Youll get a 1500 budget for hardware, as well 500 for home office to buy whatever you need to do your best work including a computer, webcam, or standing desk.Get 38 days of holiday. Plenty of time for city breaks, summer escapes, and everything in between. Youll also get a half day on your birthday to give you a chance to celebrate! Volunteer/Charity Day. Enjoy a Volunteer/Charity Day to support a cause close to your heart.Continue to grow and develop your career. We care about your development and want you to be able to learn new things! After six months in the company, youll get a budget to be able to use for personal development.Make your voice heard. We trust our team members to make the best decisions to achieve their goals, so you wont have to put up with micromanagers here.Say goodbye to pointless meetings. We practice what we preach when it comes to productivity, so you can expect flat hierarchies, fast iterations, and no bullshit meetings.What youll bring to the role You enjoy connecting with people. You know the definition of empathy without checking Wikipedia and put it into practice daily. Youre passionate about customer success management. You have experience in customer success management or account management. Sales experience is highly valued. You have business acumen and enjoy advising customers and consultative selling. You take ownership. You are filled with passion to take responsibility and to make things happen. You are an energetic self-starter who wants to make a tangible impact on a rapidly growing product and company. You are a team player. You understand that we can only be successful as a team. Like everyone else in our great team, you are fair, honest and respectful and you are ready to have a lot of fun. You are fluent in English & German. Speaking Spanish or French is a plus.

This article first appeared in The Checkup,MIT Technology Reviewsweekly biotech newsletter. To receive it in your inbox every Thursday, and read articles like this first,sign up here. On January 20, his first day in office, US president Donald Trump signed an executive order to withdraw the US from the World Health Organization. Ooh, thats a big one, he said as he was handed the document. The US is the biggest donor to the WHO, and the loss of this income is likely to have a significant impact on the organization, which develops international health guidelines, investigates disease outbreaks, and acts as an information-sharing hub for member states. But the US will also lose out. Its a very tragic and sad event that could only hurt the United States in the long run, says William Moss, an epidemiologist at Johns Hopkins Bloomberg School of Public Health in Baltimore. Trump appears to take issue with the amount the US donates to the WHO. He points out that it makes a much bigger contribution than China, a country with a population four times that of the US. It seems a little unfair to me, he said as he prepared to sign the executive order. It is true that the US is far and away the biggest financial supporter of the WHO. The US contributed $1.28 billion over the two-year period covering 2022 and 2023. By comparison, the second-largest donor, Germany, contributed $856 million in the same period. The US currently contributes 14.5% of the WHOs total budget. But its not as though the WHO sends a billion-dollar bill to the US. All member states are required to pay membership dues, which are calculated as a percentage of a countrys gross domestic product. For the US, this figure comes to $130 million. China pays $87.6 million. But the vast majority of the USs contributions to the WHO are made on a voluntary basisin recent years, the donations have been part of multibillion-dollar spending on global health by the US government. (Separately, the Bill and Melinda Gates Foundation contributed $830 million over 2022 and 2023.) Theres a possibility that other member nations will increase their donations to help cover the shortfall left by the USs withdrawal. But it is not clear who will step upor what implications it will have to change the structure of donations. Martin McKee, a professor of European public health at the London School of Hygiene and Tropical Medicine, thinks it is unlikely that European members will increase their contributions by much. China, India, Brazil, South Africa, and the Gulf states, on the other hand, may be more likely to pay more. But again, it isnt clear how this will pan out, or whether any of these countries will expect greater influence over global health policy decisions as a result of increasing their donations. WHO funds are spent on a range of global health projectsprograms to eradicate polio, rapidly respond to health emergencies, improve access to vaccines and medicines, develop pandemic prevention strategies, and more. The loss of US funding is likely to have a significant impact on at least some of these programs. Diseases dont stick to national boundaries, hence this decision is not only concerning for the US, but in fact for every country in the world, says Pauline Scheelbeek at the London School of Hygiene and Tropical Medicine.With the US no longer reporting to the WHO nor funding part of this process, the evidence on which public health interventions and solutions should be based is incomplete. Its going to hurt global health, adds Moss. Its going to come back to bite us. Theres more on how the withdrawal could affect health programs, vaccine coverage, and pandemic preparedness in this weeks coverage. Now read the rest of The Checkup Read more from MIT Technology Review's archive This isnt the first time Donald Trump has signaled his desire for the US to leave the WHO. He proposed a withdrawal during his last term, in 2020. While the WHO is not perfect, it needs more power and funding, not less, Charles Kenny, director of technology and development at the Center for Global Development, argued at the time. The move drew condemnation from those working in public health then, too. The editor in chief of the medical journal The Lancet called it a crime against humanity, as Charlotte Jee reported. In 1974, the WHO launched an ambitious program to get lifesaving vaccines to all children around the world. Fifty years on, vaccines are thought to have averted 154 million deathsincluding 146 million in children under the age of five. The WHO has also seen huge success in its efforts to eradicate polio. Today, wild forms of the virus have been eradicated in all but two countries. But vaccine-derived forms of the virus can still crop up around the world. At the end of a round of discussions in September among WHO member states working on a pandemic agreement, director-general Tedros Adhanom Ghebreyesus remarked, The next pandemic will not wait for us, whether from a flu virus like H5N1, another coronavirus, or another family of viruses we dont yet know about. The H5N1 virus has been circulating on US dairy farms for months now, and the US is preparing for potential human outbreaks. From around the web People with cancer paid $45,000 for an experimental blood-filtering treatment, delivered at a clinic in Antigua, after being misled about its effectiveness. Six of them have died since their treatments. (The New York Times) The Trump administration has instructed federal health agencies to pause all external communications, such as health advisories, weekly scientific reports, updates to websites, and social media posts. (The Washington Post) A new virtual retina, modeled on human retinas, has been developed to study the impact of retinal implants. The three-dimensional model simulates over 10,000 neurons. (Brain Stimulation) Trump has signed an executive order stating that it is the policy of the United States to recognize two sexes, male and female. The document defies decades of research into how human bodies grow and develop, STAT reports, and represents a dramatic failure to understand biology, according to a neuroscientist who studies the development of sex. (STAT) Attention, summer holiday planners: Biting sandflies in the Mediterranean region are transmitting Toscana virus at an increasing rate. The virus is a major cause of central nervous system disorders in the region. Italy saw a 2.6-fold increase in the number of reported infections between the 201621 period and 202223. (Eurosurveillance)