RoboCop: Rogue City – Unfinished Business preview: ‘We created something bigger than we expected’ Adam Starkey Published June 3, 2025 9:00am Stay out of trouble (Nacon) GameCentral goes hands-on with the standalone expansion of RoboCop: Rogue City, which dials up the action and gory splatter of 2023’s surprise hit. For a franchise that has arguably done nothing of worth since the early 90s, the future of RoboCop is looking surprisingly bright. Following Amazon’s acquisition of MGM, a new TV show is currently in the works, with rumbles of a new film as well. Whether this leads to a major rejuvenation for everyone’s favourite cyborg law enforcer remains to be seen, but the original source of any RoboCop redemption arc has to start with 2023’s RoboCop: Rogue City.  Developed by Polish studio Teyon, RoboCop: Rogue City was the kind of unexpected surprise you rarely get from licensed games. It recaptured the original’s wit and 80s aesthetic, but also found a way to deliver the fantasy of playing as the half-human cyborg without streamlining any of the character’s personality. The bloody action was built around his hulking, slow movement, dry one-liners were in abundance, and missions weren’t always reduced to mowing down thugs in corridors – you also handed people parking tickets, settled trivial civilian disputes, and, in one wonderfully mundane side mission, did the rounds in the office for a get well card.  The game became publisher Nacon’s ‘best ever launch’ with 435,000 players within two weeks. Now, a year and a half later, developer Teyon is back with a standalone expansion. Marketing around Unfinished Business has purposefully dodged the term *DLC*, but as explained by the studio’s communications manager, Dawid Biegun, it started out as exactly that.  ‘When we released RoboCop: Rogue City, we were thinking about, this story has many things [we can] do in the future,’ says Biegun. ‘We had many paths we could choose. So we basically started slowly developing some new storyline. The game was planned to be DLC but it grew out of control. It was a really rare situation where we created something bigger than we expected, so it became a standalone expansion from then.’ Unlike Rogue City, this expansion, which we’re told spans around eight hours on average, is centred around one location in the OmniTower. Like most things in the RoboCop realm created by OCP, this promised idyllic housing complex quickly goes south when a band of mercenaries assume control. To restore order, and after a creepy opening where an attack on the Detroit police station leaves several officers frozen solid, RoboCop is assigned to the case.  RoboCop has new moves at his disposal (Nacon) Unfinished Business wastes little time in throwing you into the action, and quickly amps up the chaos. For anyone who has played Rogue City, all the original tenets of the combat are here, albeit with a slight increase in difficulty. You’ll be looking for explosive cans to blast, illuminated panels to ricochet bullets off walls, and all the while trying not to expose yourself to too much gunfire. The combat purposefully doesn’t have the slick speed of Call Of Duty, but it is still aggressively punchy, with headshots resulting in satisfyingly bloody splatters and RoboCop’s famed Auto-9 machine pistol still having the kickback of a pocket pneumatic drill.  From the get-go, Unfinished Business pushes back in a way Rogue City never did. New enemies equipped with riot shields are a real nuisance if you don’t utilise the ricochet panels, while the ability to slow down time is a much bigger crutch to chip down the enemy numbers from a distance. Health pick-ups felt in shorter supply too, even on the normal difficulty, to the point where we barely scraped through several encounters.  While it’s unclear if this applies to the whole game, Unfinished Business feels like a gnarlier experience, when compared to the original. RoboCop has some new context sensitive finishing moves, like throwing enemy heads into concrete walls or vending machines, which is a satisfying addition to the melee arsenal. There’s greater enemy variety too, between fierce minigun heavyweights and flying drones, along with some neat action set pieces. In one standout, we had to operate a walkway bridge to deactivate a giant turret at the end of a room, dashing between cover as it rains down bullets and destroys the surrounding environment. Anyone who has played action games before will recognise all the mechanics at play in this scenario, but it was still well executed and effective. Another had a whiff of Star Wars, as you rush around shooting electrical panels to stop a trash compactor from crushing you via the descending ceiling. The action shift in Unfinished Business is best defined by a later sequence we got to play, where you take control of the franchise’s signature mech, ED-209. If the power fantasy of playing as RoboCop is tested in this expansion, ED-209’s section was pure mental catharsis, where you blast away enemy hordes with miniguns and rockets, and clean up any stragglers with a rigid, robotic stomp. The rush of piloting ED-209, with its cacophony of explosions and bullets, felt like a throwback to vehicle sections in a long lost Xbox 360 game – but in a good way.  While there’s a definite lean towards combat, rather than gift card signing, when compared to Rogue City, it hasn’t entirely abandoned the detective side. According to the developers, if Rogue City had a 60/40 percent split between guns and detective work, Unfinished Business ‘would be like 70/30, or 80/20’ in comparison. More Trending We saw some of this , with one memorable encounter seeing you quizzed by a RoboCop superfan who is unconvinced you’re the actual RoboCop, leading to a series of questions based on the history of the franchise. There is optional side missions too, although the time we had with our preview limited our chance to fully delve into them. The sales and positive reviews for RoboCop: Rogue City emboldened Teyon’s vision and scope for Unfinished Business – and that confidence shines through in what we played. Some might be disappointed by the steer towards action, and we were heading into this preview, but by the end, this felt like a welcome extension with its own unique flavour. This is RoboCop: Rogue City with its pedal to the floor, confined and concentrated into a lean, tightly focused machine.  As for the studio’s next steps, the success of RoboCop has only reaffirmed Teyon’s strengths and identity as a team. Between its three studios across Poland and Japan, with over 140 employees in total, Teyon wants to maintain its grip within the AA space. ‘We feel strong here in such games,’ Biegun said. ‘We wouldn’t want to grow like 200, 300, 400 people, because we’re going to lose our soul this way. We want to stay as we are right now.’ ED-209 needs be wary of stairs (Nacon) Email gamecentral@metro.co.uk, leave a comment below, follow us on Twitter, and sign-up to our newsletter. To submit Inbox letters and Reader’s Features more easily, without the need to send an email, just use our Submit Stuff page here. For more stories like this, check our Gaming page. Arrow MORE: How to get a Nintendo Switch 2 this week in the UK GameCentral Sign up for exclusive analysis, latest releases, and bonus community content. This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply. Your information will be used in line with our Privacy Policy

We may earn a commission from links on this page.Purchasing a full set of individual dumbbell pairs between five and 60 pounds can cost well over $1,500, not to mention how much space you'd have to dedicate to storing them. That's why I'm a huge fan of adjustable dumbbells like the REP QuickDraws or Nüobells, which I reviewed here and here, respectively. If you know you want to be able to up to 60 pounds at five-pound increments, effectively combining 30 weights into one compact set, then adjustables like the REP QuickDraws are how you can save significant money and space in your home gym equipment setup. But let's say your strength progresses beyond your adjustable dumbbells' capacity. First of all: Nice work. Secondly: What do you do now? I've seen users on Reddit directly asking "what's stopping me from adding more plates to keep going heavier?" Adding extra weight plates seems like a logical solution, as opposed to spending all that money on an entirely new set for just a few pounds beyond the recommended capacity. So, let's take a look at what is stopping you—and what's not. Here's what you should know about using add-on weights to your adjustable dumbbells, from what is officially considered unsafe to why it gets impractical even aside from that. How to add weight to your adjustable dumbbellsFirst off, not all adjustable dumbbells allow for this type of modification. Selector pin systems (like PowerBlocks) generally don't allow for unofficial weight additions due to their enclosed design. Dial systems (like my favorite, the Nüobells) have security features specifically preventing modifications outside the cradle. Then you have spin-lock dumbbells, like the QuickDraws, which allow for customization and additional weight outside the cradle. For this reason, I'll be speaking to the REP QuickDraws specifically, as opposed to Nüobells or Powerblocks.Still, adding weight beyond the manufacturer's specifications will almost certainly void your warranty. REP Fitness designed these dumbbells for specific weight ranges and cannot guarantee safety beyond those limits. I spoke with a REP representative (REP-resentative?) who informed me, "internal testing was not conducted beyond the 60 pound maximum, so REP cannot guarantee the safety of additional weight beyond that."Outside of your warranty, there are safety concerns. The handles, locking mechanisms, and materials are engineered for specific maximum loads. Exceeding these limits risks equipment failure and potential injury. Even if the dumbbells can technically handle the additional plates, the result might be bulky and awkward, affecting your form and increasing injury risk.All that being said, it still is entirely feasible to bend the rules and take your 60-pound dumbbells up to 70 pounds. Assuming you play it safe and only add a few plates to each side, here's what you should know:Only add compatible plates: The QuickDraw system uses standard plate dimensions that can be supplemented with additional compatible plates.Manual addition required: You'll need to add these plates manually outside the cradle system, as the selection mechanism only works within the advertised weight range. This pretty much negates the whole "quick draw" appeal.Secure them properly: Ensure the additional plates are secured with the locking mechanism to prevent dangerous slippage during exercises.Manual tracking required: The dial indicator becomes inaccurate once you exceed the designed range. You'll need to manually calculate and track the added weight.Can you (and should you) add more weight to REP QuickDraw Dumbbells?The short answer: Yes, technically you can add additional weight plates to REP QuickDraw dumbbells beyond their advertised maximum. The long answer: Just because something is possible doesn't mean you should make a habit of it. Sure, it's fully above-board to order add-on weights in five-pound increments (which, side-note, I love so much more than a 10-pound jump). However, going beyond 60 pounds isn't sanctioned by REP. For this reason, I naturally had my boyfriend test my DIY super-QuickDraws creation. In the image below, you can see how I manually added weights outside of the cradle. With the spin-locks, you can keep adding weights outside the cradle without any resistance. Just slide the weights on and flip the lock switch.I snapped a picture of it in its lopsided form, so you can see just how far outside the cradle the five-pound add-ons will go—and how precarious it starts to look right off the bat. Behold, my DIY super-QuickDraws creation. Credit: Meredith Dietz As we steadily increased the weight from 65, to 70, to 85 pounds, he noted growing concerns with how much the dumbbells could handle while still feeling safe. The all-steel design gives these guys a premium feel during standard use. But when you go beyond the sanctioned weight capacity, there's a sense that the dumbbells wouldn't bend or snag, but simply snap. And again, you'll need to manually keep track of how much weight you're adding, since the individual plates are not labeled, and you're exceeding the labels of the cradle. That mental math might not be a deterrent for you, but it's something to note.The bottom lineIn most cases, the hassle and safety concerns outweigh the benefits. The primary advantage of the QuickDraws—quick, convenient weight changes—is totally undermined when manually adding plates.Investing in a higher-capacity adjustable dumbbell set (like 80- to 90-pound models) or supplementing with specific heavier fixed dumbbells for certain exercises might be safer and more practical. For serious lifters approaching the upper limits of their adjustable dumbbells, I'd recommend supplementing with individual heavier dumbbells.

DeepSeek's R1 update, plus the rest of the AI news this week. Credit: Thomas Fuller / SOPA Images / LightRocket / Getty Images This week, DeepSeek released an updated version of its R1 model on HuggingFace, reigniting the open-source versus closed-source competition. The updated version, called DeekSeek-R1-0528, has 685 billion parameters, an upgrade from January's version, which had 671 billion. Unlike OpenAI and Google's models, which are famously closed-source, DeepSeek's model weights are publicly available. According to the benchmarks, the R1-0528 update has improved reasoning and inference capabilities and is closing the gap with OpenAI's o3 and Google's Gemini 2.5 Pro. DeepSeek also introduced a distilled version of R1-0528 using Alibaba's Qwen3 8B model. This is an example of a lightweight model that is less capable but also requires less computing power. DeepSeek-R1-0528-Qwen3-8B outperforms both Google's latest lightweight model Gemini-2.5-Flash-Thinking-0520 and OpenAI's o3-mini in certain benchmarks. But the bigger deal is that DeekSeek's distilled model can reportedly run on a single GPU, according to TechCrunch. You May Also Like To… distill all this information, the Chinese rival is catching up to its U.S. competitors with an open-weight approach that's cheaper and more accessible. Plus, DeepSeek continues to prove that AI models may not require as much computing power as OpenAI, Google, and other AI heavyweights currently use. Suffice to say, watch this space.That said, DeepSeek's models also have their drawbacks. According to one AI developer (via TechCrunch), the new DeepSeek update is even more censored than its previous version when it comes to criticism of the Chinese government. Of course, a lot more happened in the AI world over the past few days. After last week's parade of AI events from Google, Anthropic, and Microsoft, this week was lighter on product and feature news. That's one reason DeepSeek's R1 update captured the AI world's attention this week. In other AI news, Anthropic finally gets voice mode, AI influencers go viral, Anthropic's CEO warns of mass layoffs, and an AI-generated kangaroo. Google's Veo 3 takes the internet by stormOn virtually every social media platform, users are freaking out about the new Veo 3, Google's new AI video model. The results are impressive, and we're already seeing short films made entirely with Veo 3. Not bad for a product that came out 11 days ago. Not to be outdone by AI video artists, a reporter from The Wall Street Journal made a short film about herself and a robot using Veo 3.Mashable's Tech Editor Timothy Werth recapped Veo's big week and had a simple conclusion: We're so cooked.More AI product news: Claude's new voice mode and the beginning of the agentic browser eraAfter last week's barrage, this week was lighter on the volume of AI news. But what was announced this week is no less significant.  Mashable Light Speed Want more out-of-this world tech, space and science stories? Sign up for Mashable's weekly Light Speed newsletter. By clicking Sign Me Up, you confirm you are 16+ and agree to our Terms of Use and Privacy Policy. Thanks for signing up! Anthropic finally introduced its own voice mode for Claude to compete with ChatGPT, Grok, and Gemini. The feature is currently in beta on mobile for the Claude app and will even be available to free plans with a limit of 20 to 30 voice conversations per day. Anthropic says you can ask Claude to summarize your calendar or read documents out loud. Paying subscribers can connect to Google Workspace for Calendar, Gmail, and Docs access. OpenAI is exploring the ability to sign into third-party apps with ChatGPT. We don't know much yet, but the company posted an interest form on its site for developers using Codex, its engineering agent, to add this capability to their own apps. It may not sound like a big deal, but it basically means users could easily link their personalized ChatGPT memories and settings to third-party apps, much like the way it works when you sign into a new app with your Google account.Opera announced a new agentic AI browser called Neon. "Much more than a place to view web pages, Neon can browse with you or for you, take action, and help you get things done," the announcement read. That includes a chatbot interface within the browser and the ability to fill in web forms for tasks like booking trips and shopping. The announcement, which included a promo video of a humanoid robot browsing the robot, which is scant on details but says Neon will be a "premium subscription product" and has a waitlist to sign up.The browser has suddenly become a new frontier for agentic AI, now that it's capable of automating web search tasks. Perplexity is working on a similar tool called Comet, and The Browser Company pivoted from its Arc browser to a more AI-centric browser called Dia. All of this is happening while Google might be forced to sell off Chrome, which OpenAI has kindly offered to take off its hands. Dario Amodei's prediction about AI replacing entry-level jobs is already starting to happenAnthropic CEO Dario Amodei warned in an interview with Axios that AI could "wipe out half of all entry-level white-collar jobs." Amodei's predictions might be spot on because a new study from VC firm SignalFire found that hiring for entry-level jobs is down to 7 percent from 25 percent in the previous year. Some of that is due to changes in the economic climate, but AI is definitely a factor since firms are opting to automate the less-technical aspects of work that would've been taken on by new hires.  Related Stories The latest in AI culture: That AI-generated kangaroo, Judge Judy, and everything elseGoogle wants you to know its AI overviews reach 1.5 billion people a month. They probably don't want you to know AI Overviews still struggles to count, spell, and know what year it is. As Mashable's Tim Marcin put it, would AI Overviews pass concussion protocol?The proposal of a 10-year ban on states regulating AI is pretty unpopular, according to a poll from Common Sense Media. The survey found that 57 percent of respondents opposed the moratorium, including half of the Republican respondents. As Mashable's Rebecca Ruiz reported, "the vast majority of respondents, regardless of their political affiliation, agreed that Congress shouldn't ban states from enacting or enforcing their own youth online safety and privacy laws."In the private sector, The New York Times signed a licensing deal with Amazon to allow their editorial content to be used for Amazon's AI models. The details are unclear, but from the outside, this seems like a change of tune from the Times, which is currently suing OpenAI for copyright infringement for allegedly using its content to train its models. That viral video of an emotional support kangaroo holding a plane ticket and being denied boarding? It's AI-generated, of course. Slightly more obvious, but no less creepy is another viral trend of using AI to turn public figures like Emmanuel Macron and Judge Judy into babies. These are strange AI-slop-infested times we're living in. AI has some positive uses too. This week, we learned about a new humanoid robot from HuggingFace called HopeJr (with engineering by The Robot Studio), which could be available for sale later this year for just $3,000.And to end this recap on a high note, the nonprofit Colossal Foundation has developed an AI algorithm to detect the bird calls of the near-extinct tooth-billed pigeon. Also known as the "little dodo," the tooth-billed pigeon is Samoa's national bird, and scientists are using the bioacoustic algorithm to locate and protect them. Want to get the latest AI news, from new product features to viral trends? Check back next week for another AI news recap, and in the meantime, follow @cecily_mauran and @mashable for more news.Disclosure: Ziff Davis, Mashable’s parent company, in April filed a lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems. Topics OpenAI DeepSeek Cecily Mauran Tech Reporter Cecily is a tech reporter at Mashable who covers AI, Apple, and emerging tech trends. Before getting her master's degree at Columbia Journalism School, she spent several years working with startups and social impact businesses for Unreasonable Group and B Lab. Before that, she co-founded a startup consulting business for emerging entrepreneurial hubs in South America, Europe, and Asia. You can find her on X at @cecily_mauran.

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

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

We may earn a commission from links on this page.Adjustable dumbbells are incredible space-savers compared to full dumbbell sets, but still, they're heavy and awkward to reposition. And look, I get it—the point of weights is to lift them, right? When testing and reviewing dumbbells like the Nüobells or REP Quickdraws, I quickly discovered that the genius of adjustable dumbbells is exactly what makes them so hard to lug around: In order to adjust from 5 to 80 pounds, they end up weighing, you know, at least 80 pounds. And not every dumbbell user has built up to that strength yet. You can buy a specialized dumbbells storage cart, but that'll run you anywhere from $112 to $350 to even $595. Enter this hack from the home gym enthusiasts of Reddit (specifically, Redditor stalechocmuffin): Use a miniature furniture dolly to roll your weights around.How to move and store your adjustable dumbbellsSure, you can drag and shove your adjustable dumbbells around in their cradles, but that sort of rough handling isn't ideal. A mini furniture dolly provides a quick, easy transition between exercises or storage positions. Plus, you'll be protecting both your dumbbells and your flooring. And if you're like me, it's stressful to drag 80 pounds around the tiny apartment you rent for way too much money. Even if I'm not worried about scratching my floors, lifting too heavy increases risk of injury.To make the most of this dolly hack, here are some tips to keep in mind:Place one dolly under each dumbbell stand for independent movement.Consider securing the dumbbells to the dolly with removable hook-and-loop straps if you have children or pets.Add rubber matting between the dolly and dumbbell base for extra protection and stability.Store against a wall when not in use to maximize floor space.The key to this hack working effectively is using the right-sized dolly. Standard furniture dollies are too large, but here are some options that should fit adjustable dumbbell bases perfectly:Ironton Carpeted Mover's Dolly ($22.99)18" x 12" platform1,000 lb capacity (far more than needed, but ensures stability)Carpeted surface provides cushioningHeavy Duty Furniture Dolly ($24.99)16.3" x 11" platform500 lb capacity360-degree rotating caster wheelsGlobal Industrial Hardwood Dolly ($65.06)24" x 16" platform 1,000 lb capacitySolid deck for extra stabilityCompared to a storage unit that costs hundreds of dollars, a mini dolly is the obvious choice to move your dumbbells around your home. You can make your workout flow more smoothly while protecting both your investment and your home.

I never considered myself a walking girl. I never engaged in the "hot girl walk" trends on social media or went on "mental health walks" during the pandemic lockdown. In fact, I long thought walking—the milestone most of us reach as babies, the activity the majority of us do each day to accomplish all the other basic tasks of living—had a little too much PR hype, especially after learning that the much-ballyhooed "10,000 steps" we're supposed to take every day relied on an arbitrary, made-up figure for marketing pedometers. If I am going to do cardio, I reasoned, I'm going to do cardio: cycling, running, swimming, or playing sports with my friends. If I'm not sweating, what's the use? After trying out Peloton's guided walks, available in the at-home fitness giant's incredibly versatile app, I have learned the use. I am now, finally, a walking girl. Is walking good cardio?The reductive view I formerly held of cardio—that I have to be sweaty and tired for it to matter—is and was always false, which I knew, intellectually. As Lifehacker senior health editor Beth Skwarecki has explained before, walking is cardio—and it's actually a pretty good form of it, too. How fast you walk can even be used to measure your health and capacity. Different intensities of cardio do different things for your body, but at its most basic level, walking still burns calories. It's also a great, easy way to work a little extra movement into your life, especially if you're a fitness beginner or have an injury. The catalyst for me checking out Peloton's walking offerings was actually my mom being "prescribed" walking as a treatment for an issue she's been having with her back. The issue prevented her from walking long periods of time or walking fast, so after addressing it medically with doctors and physical therapists, her at-home assignment was to walk longer and longer durations on a walking pad in the living room.As an able-bodied person living in a walkable city (by which I mean a city where I am basically forced to hit my arbitrary 10,000 steps per day whether I want to or not), I have definitely taken the ability to walk for granted. I decided to check out Peloton's walking workouts to see if they'd be useful for my mom—but they ended up being useful for me. What are Peloton's walking workouts all about?To find walking workouts on the Peloton app, select Walking from the top of your home screen or type "walking" into the search bar. Peloton's walking workouts are designed for use on their Tread treadmills (or any treadmill, really)—but I've found that I enjoy them just as much if I go outside, although I obviously can't control the incline if I do that. The guided walks available in the app are like any class Peloton offers: They come in a variety of lengths and formats, are led by a certified instructor who encourages you and reminds you of safety cues, and feature playlists of music that keep the energy going. I start off nearly every weekday morning by walking to Dunkin' Donuts and then to the post office to drop off whatever I've sold on resale apps, so I queue up a Peloton walk for my journey. While I don't necessarily need to have an instructor in my ears reminding me to, well, walk, it encourages me to keep my pace up; I just ignore whatever they're saying about messing with incline and resistance buttons, as I'm not on a treadmill. This morning, I walked along with a five-minute warmup walk routine from instructor Logan Aldridge, who shared encouraging reminders that walking, even if it feels easy, is "massively worth it" for a person's health. He also gave speed cues using practical, real-world examples instead of just relying on cues built around treadmill functions. At one point, he described the pace goal as "not Manhattan walking, not New York City walking," which was funny because I was, in fact, Manhattan-walking my way to a Dunkin', so I slowed down a bit. You can enable location sharing for more accurate measurements and, of course, I have my Apple Watch paired with my Peloton app to give me better data on my heart rate, output, and speed, too. I forgot to enable my location tracking at the beginning of the walk (I don't have it set to automatically track that, though you can), so at the end, it prompted me to enter in my distance walked for better measurements. I glanced at my watch, which told me how far I'd walked, entered in that number, and was taken to a screen where I could review my output. You can absolutely do this on a treadmill, too, and the workouts are more or less designed for you to. There are live classes available, which enter the on-demand archive when they're finished, and you can choose from cool-down walks, power walks, hikes, walks set to certain kinds of playlists (like '90s music or EDM), or even "walk & talk" walks that have two instructors if you like that chatty, podcast kind of feel. Some classes feature walking and running and their titles tell you that upfront. As you're scrolling the options, you'll mostly see title cards with instructors on Treads in the Peloton studio, but you'll also see a few where the instructors are outside. These guided walks are designed more for outdoor walks and the instructors will call out the half-way point so you always know when to turn around and head home (or back to the office or whatever). The workouts come in all kinds of lengths, from five minutes up to 75, with the longer ones often incorporating both walking and running. Why I like Peloton's walking workoutsThese workouts are an easy way to slot some extra intentional movement into my day. I'm already walking around a lot, but I'm not always doing it with purpose. Having an instructor reminding me to connect with my steps and a playlist designed to keep me on a certain pace turns a standard coffee run into a mindful exercise. Walking is also low-impact and accessible, so even on a day you're tired or even if other forms of cardio are beyond your reach, this opens up a whole world of fitness opportunities. Perhaps most importantly, this is the most accessible kind of workout on the app because you really don't need anything special. You don't need a floor mat, yoga blocks, or weights, let alone a fancy treadmill. As long as you have some good shoes, you can walk around all you want while still getting the company's signature encouragement and guidance from trained pros.

PROS: Discreet, elegant, and unobtrusive design that doesn't scream "tech" Lightweight and comfortable premium frame Focuses on essential experiences without the unnecessary cruft Impressive transcription and teleprompter features Long battery life and effortless charging case design CONS: No speakers for calls or audio feedback (especially during navigation) Temple tips touch controls can be a bit cumbersome A bit expensive RATINGS: AESTHETICSERGONOMICSPERFORMANCESUSTAINABILITY / REPAIRABILITYVALUE FOR MONEYEDITOR'S QUOTE:With a simple design and useful features, the Even Realities G1 smart glasses prove that you don't need all the bells and whistles to provide an experience. Every day, we’re flooded with more information than our already overworked minds can handle. Our smartphones and computers put all this information at our fingertips, connecting us to the rest of the world while ironically disconnecting us from the people around us. Smart glasses and XR headsets promise to bring all this information right in front of us, bridging the gap that divides physical and virtual realities. And yet at the same time, they erect a wall that separates us from the here and now. It’s against this backdrop that Even Realities chose to take a bold step in the opposite direction. In both form and function, the Even Realities G1 smart glasses cut down on the cruft and promise a distilled experience that focuses only on what you really need to get through a busy day. More importantly, it delivers it in a minimalist design that doesn’t get in your way. Or at least that’s the spiel. Just in time for the upcoming Father’s Day celebration, we got to test what the Even Realities G1 has to offer, especially to some of the busiest people in our families: the dads juggling work responsibilities while trying to stay present for their loved ones. Designer: Even Realities Click Here to Buy Now: $599. Exclusive Father’s Day Special – Get 50% Off the G1 Clip + Clip Pouch! Hurry, offer ends June 15, 2025. Aesthetics You probably wouldn’t even be able to tell the Even Realities G1 is wearable tech if you meet someone on the street wearing a pair. Sure, they might look like slightly retro Pantos, but they’re a far cry from even the slimmest XR glasses from the likes of Xreal or Viture. You can clearly see the eyes of the person wearing them, and the tech is practically invisible, which is exactly the point. The design of the Even Realities G1 is on the plain and minimal side, a stark contrast to the majority of smart glasses and XR/AR headsets currently in the market, even those claiming to be fashionable and stylish. Sure, it’s not going to compete with high-end luxury spectacles, but they’re not entirely off the mark either. Unless you look really closely, you might simply presume them to be a pair of thick-framed glasses. The form of the glasses might be simple, but their construction is anything but. The frame is made from magnesium alloy with a coating that’s fused with sandstone, while the temples use a titanium alloy on the outer sides and soft silicone on the inner surfaces. The mixture of quality materials not only gives the Even Realities G1 a premium character but also a lightweight form that’s only ever so slightly heavier than your run-of-the-mill prescription eyeglasses. While the G1 most looks like normal eyewear, the temple tips are dead giveaways that things are not what they seem. The blocky, paddle-shaped tips that house batteries and electronics are definitely larger than what you’d find on most glasses. They’re not obnoxiously big, but they do tend to stick out a bit, and they’re hard to “unsee” once you’ve noticed their presence. Despite looking quite aesthetic, the Even Realities G1 isn’t pretending to be some posh fashion accessory. After all, the circular G1A and rectangular G1B options hardly cover all possible eyewear designs, and the limited color selection won’t suit everyone’s tastes. Rather than something you flaunt or call attention to, these smart glasses are designed to be an “everyday wear” and disappear into the background, making tech invisible without making it unusable, perfect for the dad who wants to stay connected without looking like he’s wearing a gadget at the family barbecue. Ergonomics If you’ve ever tried any of those hi-tech wearables promising the next wave of computing, then you’d probably know that you’d never wear any of those glasses or visors for more than just an hour or two every day. They may have impressive technologies and apps, but they become practically useless once you take them off, especially when you have to step out into the real world. In contrast, the Even Realities G1 is something you’d be able to wear for hours on end, indoors or outdoors. Made from lightweight materials with a construction that even throws away screws to reduce the heft, it’s almost mind-blowing to think that the glasses houses any electronics at all. This level of comfort is honestly the G1’s most important asset, because it allows people to experience its smart features far longer than any Quest or Viture. When it comes to eyewear, however, prescription lenses have always been a sore point for many consumers, and this is no exception. Because it integrates waveguide optics into the lens, you’ll have to pay extra to have customized prescription lenses when you buy an Even Realities G1. It can be a bit nerve-wracking to ensure you get all the measurements and figures right, especially since you can’t return or exchange glasses with customized lenses. While the G1 eyeglasses are definitely comfortable to wear, the same can’t exactly be said when it comes to manually interacting with them. While most smart glasses and headsets have controls near your temples, the G1’s touch-sensitive areas are at the temple tips, which would be sitting behind your ears when you’re wearing the glasses. They might feel awkward to reach, and those with long hairstyles might find it difficult to use. Fortunately, you will rarely touch those tips except to activate some functions, but it can still be an unsatisfactory experience when you do. Performance The Even Realities G1 takes a brilliantly focused approach to smart eyewear, prioritizing elegant design and practical functionality over unnecessary tech bloat. The 640×200 green monochrome display may seem modest, but it’s deliberate choice that enables the G1 to maintain a sleek, stylish profile. The absence of cameras and speakers isn’t a limitation but a thoughtful design decision that enhances both wearability and privacy, allowing users to seamlessly integrate this technology into their daily lives without social awkwardness. The magic of the G1 lies in its delivery of information directly to your field of vision in a way that not only delights but also transforms how you interact with digital content. The core Even Realities G1 experience revolves around bringing only critical information to your attention and keeping distractions away, all without disconnecting you from reality and the people around you. Its text-centric interface, displayed by two micro-LED displays, one on each lens, ensures that information is distilled down to its most essential. And there’s no denying the retro charm of a green dot-matrix screen in front of your eyes, even if the color won’t work well against light or bright objects. The Even Realities G1 experience starts with the dashboard, which you can summon just by tilting your head up a bit, an angle that you can set on the companion mobile app. One side shows the date and time, temperature, number of notifications, and your next appointment. The other side can be configured to show one of your saved quick notes, news, stocks, or even your current location. None of these items are interactive, and you’ll have to dive into the mobile app to actually get any further information. With Father’s Day approaching, it’s worth noting how the G1’s floating heads-up display, visible only to the wearer, helps dads stay effortlessly connected, organized, and present. The QuickNote and Calendar features are particularly valuable for fathers juggling work and family responsibilities, allowing them to process their to-do lists perfectly on schedule without missing a beat of family time. Spending quality time with your child then suddenly remembering you need to buy batteries on your next errand run? No more frantically scampering for pen and paper or even your phone; just tap and speak. Of course, the smart glasses really shine when it comes to the, well, smart functionality, most of which unsurprisingly revolve around words, both spoken and displayed. Transcription, which is used when making Quick Notes, records your voice and saves it alongside the transcribed text. Fathers who find themselves in never-ending meetings no longer need to worry about missing a beat. Not only do they get to keep notes, but they also receive a summary and recap thanks to the G1’s AI capabilities, a game-changer for busy dads who need to process information efficiently. Translation can make international trips quite fun, at least for some interactions, as you’ll be able to see actual translated captions floating in the air like subtitles on a video. Dads who give a lot of talks, business presentations, interviews, or broadcast videos will definitely love the Teleprompter feature, which can advance the script just based on the words you’re speaking. No more worrying about missing important points during that big presentation, leaving more mental bandwidth for what really matters. It’s also perfect for a captivating Career Day show that will do your kid proud. The accuracy of Even Realities’ speech recognition and AI is fairly good, though there are times when it will require a bit of patience and understanding. There’s a noticeable delay when translating what people say in real time, for example, and it might miss words if the person is speaking too quickly. Navigation can be a hit or miss, depending on your location, and the visual direction prompts are not always reliable. The latter is also one of the cases where the absence of built-in speakers feels a bit more pronounced. There’s no audio feedback, which could be useful for guided turn-by-turn navigation. Even AI can hear you, but it can’t talk back to you. Everything will be delivered only through text you have to read, which might not always be possible in some cases. Admittedly, the addition of such hardware, no matter how small, will also add weight to the glasses, so Even Realities chose their battles wisely. The Even Realities G1 is advertised to last for 1.5 days, and it indeed lasts at least more than a day. The stylish wireless charging case, which has a built-in 2,000mAh battery, extends that uptime to five days. Charging the glasses is as simple as putting them inside the case, no need to align any contact points, as long as you remember to fold the left arm first before the right arm. Oddly enough, there’s no battery level indicator on the glasses, even in the dashboard HUD. Even Realities focused on making the G1 simple, both in design and in operation. Sometimes even to the point of oversimplification. To reduce complexity, for example, each side of the glasses connects to a smartphone separately via Bluetooth, which unfortunately increases the risk of the two sides being out of sync if one or the other connection drops. Turning the glasses into shades is a simple case of slapping on clip-on shades that are not only an additional expense but also something you could lose somewhere. Sustainability By cutting down on the volume of the product, Even Realities also helps cut down waste material, especially the use of plastics. The G1 utilizes more metals than plastic, not only delivering a premium design but also preferring more renewable materials. The company is particularly proud of its packaging as well, which uses 100% recyclable, eco-friendly cardboard. While magnesium and titanium alloys contribute to the durability of the product, the Even Realities G1 is not exactly what you might consider to be a weather-proof piece of wearable tech. It has no formal IP rating, and the glasses are only said to be resistant to splashes and light rain. It can accompany you on your runs, sure, but you’ll have to treat it with much care. Not that it will have much practical use during your workouts in the first place. Value Discreet, useful, and simple, the Even Realities G1 smart glasses proudly stand in opposition to the literal heavyweights of the smart eyewear market that are practically strapping a computer on your face. It offers an experience that focuses on the most important functions and information you’d want to have in front of your eyes and pushes unnecessary distractions out of your sight. Most importantly, however, it keeps the whole world clearly in view, allowing you to connect to your digital life without disconnecting you from the people around you. The Even Realities G1 would almost be perfect for this hyper-focused use case if not for its price tag. At $599, it’s easily one of the more expensive pairs of smart spectacles you’ll see on the market, and that’s only for the glasses themselves. For custom prescription lenses, you need to add another $150 on top, not to mention the $50 (normally $100) clip-on shades for those extra bright days. Given its limited functionality, the G1 definitely feels a bit overpriced. But when you consider how lightweight, distraction-free, and useful it can be, it comes off more as an investment for the future. For family and friends looking for a meaningful tech gift this Father’s Day, the G1 offers something truly unique: a way to stay on top of work responsibilities while remaining fully present for family moments. Whether capturing quick thoughts during a child’s soccer game or discreetly checking calendar reminders during family dinner, these glasses help dads maintain that delicate balance between connectivity and presence. Verdict It’s hard to escape the overabundance of information that we deal with every day, both from the world around us, as well as our own stash of notes and to-do lists. Unfortunately, the tools that we always have with us, our smartphones, computers, and smartwatches, are poor guardians against this flood. And now smart glasses are coming, promising access to all of that and threatening to further drown us with information we don’t really need. The Even Realities G1 is both a breath of fresh air and a bold statement against that trend. Not only is it lightweight and comfortable, but it even looks like normal glasses! Rather than throw everything and the kitchen sink into it, its design and functionality are completely intentional, focusing only on essential experiences and features to keep you productive. It’s not trying to turn you into Tony Stark, but it will help make you feel like a superhero as you breeze through your tasks while still being present to the people who really matter the most in your life. For the dad who wants to stay connected without being distracted, who needs to manage information without being overwhelmed by it, the Even Realities G1 might just be the perfect Father’s Day gift: a tool that helps him be both the professional he needs to be and the father he wants to be, all without missing a moment of what truly matters. Click Here to Buy Now: $599. Exclusive Father’s Day Special – Get 50% Off the G1 Clip + Clip Pouch! Hurry, offer ends June 15, 2025.The post Even Realities G1 Glasses Review: Smart, Subtle, and Perfect for Father’s Day first appeared on Yanko Design.

Artificial intelligence is a deep and convoluted world. The scientists who work in this field often rely on jargon and lingo to explain what they’re working on. As a result, we frequently have to use those technical terms in our coverage of the artificial intelligence industry. That’s why we thought it would be helpful to put together a glossary with definitions of some of the most important words and phrases that we use in our articles. We will regularly update this glossary to add new entries as researchers continually uncover novel methods to push the frontier of artificial intelligence while identifying emerging safety risks. AGI Artificial general intelligence, or AGI, is a nebulous term. But it generally refers to AI that’s more capable than the average human at many, if not most, tasks. OpenAI CEO Sam Altman recently described AGI as the “equivalent of a median human that you could hire as a co-worker.” Meanwhile, OpenAI’s charter defines AGI as “highly autonomous systems that outperform humans at most economically valuable work.” Google DeepMind’s understanding differs slightly from these two definitions; the lab views AGI as “AI that’s at least as capable as humans at most cognitive tasks.” Confused? Not to worry — so are experts at the forefront of AI research. AI agent An AI agent refers to a tool that uses AI technologies to perform a series of tasks on your behalf — beyond what a more basic AI chatbot could do — such as filing expenses, booking tickets or a table at a restaurant, or even writing and maintaining code. However, as we’ve explained before, there are lots of moving pieces in this emergent space, so “AI agent” might mean different things to different people. Infrastructure is also still being built out to deliver on its envisaged capabilities. But the basic concept implies an autonomous system that may draw on multiple AI systems to carry out multistep tasks. Chain of thought Given a simple question, a human brain can answer without even thinking too much about it — things like “which animal is taller, a giraffe or a cat?” But in many cases, you often need a pen and paper to come up with the right answer because there are intermediary steps. For instance, if a farmer has chickens and cows, and together they have 40 heads and 120 legs, you might need to write down a simple equation to come up with the answer (20 chickens and 20 cows). In an AI context, chain-of-thought reasoning for large language models means breaking down a problem into smaller, intermediate steps to improve the quality of the end result. It usually takes longer to get an answer, but the answer is more likely to be correct, especially in a logic or coding context. Reasoning models are developed from traditional large language models and optimized for chain-of-thought thinking thanks to reinforcement learning. (See: Large language model) Techcrunch event Join us at TechCrunch Sessions: AI Secure your spot for our leading AI industry event with speakers from OpenAI, Anthropic, and Cohere. For a limited time, tickets are just $292 for an entire day of expert talks, workshops, and potent networking. Exhibit at TechCrunch Sessions: AI Secure your spot at TC Sessions: AI and show 1,200+ decision-makers what you’ve built — without the big spend. Available through May 9 or while tables last. Berkeley, CA | June 5 REGISTER NOW Deep learning A subset of self-improving machine learning in which AI algorithms are designed with a multi-layered, artificial neural network (ANN) structure. This allows them to make more complex correlations compared to simpler machine learning-based systems, such as linear models or decision trees. The structure of deep learning algorithms draws inspiration from the interconnected pathways of neurons in the human brain. Deep learning AI models are able to identify important characteristics in data themselves, rather than requiring human engineers to define these features. The structure also supports algorithms that can learn from errors and, through a process of repetition and adjustment, improve their own outputs. However, deep learning systems require a lot of data points to yield good results (millions or more). They also typically take longer to train compared to simpler machine learning algorithms — so development costs tend to be higher. (See: Neural network) Diffusion Diffusion is the tech at the heart of many art-, music-, and text-generating AI models. Inspired by physics, diffusion systems slowly “destroy” the structure of data — e.g. photos, songs, and so on — by adding noise until there’s nothing left. In physics, diffusion is spontaneous and irreversible — sugar diffused in coffee can’t be restored to cube form. But diffusion systems in AI aim to learn a sort of “reverse diffusion” process to restore the destroyed data, gaining the ability to recover the data from noise. Distillation Distillation is a technique used to extract knowledge from a large AI model with a ‘teacher-student’ model. Developers send requests to a teacher model and record the outputs. Answers are sometimes compared with a dataset to see how accurate they are. These outputs are then used to train the student model, which is trained to approximate the teacher’s behavior. Distillation can be used to create a smaller, more efficient model based on a larger model with a minimal distillation loss. This is likely how OpenAI developed GPT-4 Turbo, a faster version of GPT-4. While all AI companies use distillation internally, it may have also been used by some AI companies to catch up with frontier models. Distillation from a competitor usually violates the terms of service of AI API and chat assistants. Fine-tuning This refers to the further training of an AI model to optimize performance for a more specific task or area than was previously a focal point of its training — typically by feeding in new, specialized (i.e., task-oriented) data.  Many AI startups are taking large language models as a starting point to build a commercial product but are vying to amp up utility for a target sector or task by supplementing earlier training cycles with fine-tuning based on their own domain-specific knowledge and expertise. (See: Large language model [LLM]) GAN A GAN, or Generative Adversarial Network, is a type of machine learning framework that underpins some important developments in generative AI when it comes to producing realistic data – including (but not only) deepfake tools. GANs involve the use of a pair of neural networks, one of which draws on its training data to generate an output that is passed to the other model to evaluate. This second, discriminator model thus plays the role of a classifier on the generator’s output – enabling it to improve over time.  The GAN structure is set up as a competition (hence “adversarial”) – with the two models essentially programmed to try to outdo each other: the generator is trying to get its output past the discriminator, while the discriminator is working to spot artificially generated data. This structured contest can optimize AI outputs to be more realistic without the need for additional human intervention. Though GANs work best for narrower applications (such as producing realistic photos or videos), rather than general purpose AI. Hallucination Hallucination is the AI industry’s preferred term for AI models making stuff up – literally generating information that is incorrect. Obviously, it’s a huge problem for AI quality.  Hallucinations produce GenAI outputs that can be misleading and could even lead to real-life risks — with potentially dangerous consequences (think of a health query that returns harmful medical advice). This is why most GenAI tools’ small print now warns users to verify AI-generated answers, even though such disclaimers are usually far less prominent than the information the tools dispense at the touch of a button. The problem of AIs fabricating information is thought to arise as a consequence of gaps in training data. For general purpose GenAI especially — also sometimes known as foundation models — this looks difficult to resolve. There is simply not enough data in existence to train AI models to comprehensively resolve all the questions we could possibly ask. TL;DR: we haven’t invented God (yet).  Hallucinations are contributing to a push towards increasingly specialized and/or vertical AI models — i.e. domain-specific AIs that require narrower expertise – as a way to reduce the likelihood of knowledge gaps and shrink disinformation risks. Inference Inference is the process of running an AI model. It’s setting a model loose to make predictions or draw conclusions from previously-seen data. To be clear, inference can’t happen without training; a model must learn patterns in a set of data before it can effectively extrapolate from this training data. Many types of hardware can perform inference, ranging from smartphone processors to beefy GPUs to custom-designed AI accelerators. But not all of them can run models equally well. Very large models would take ages to make predictions on, say, a laptop versus a cloud server with high-end AI chips. [See: Training] Large language model (LLM) Large language models, or LLMs, are the AI models used by popular AI assistants, such as ChatGPT, Claude, Google’s Gemini, Meta’s AI Llama, Microsoft Copilot, or Mistral’s Le Chat. When you chat with an AI assistant, you interact with a large language model that processes your request directly or with the help of different available tools, such as web browsing or code interpreters. AI assistants and LLMs can have different names. For instance, GPT is OpenAI’s large language model and ChatGPT is the AI assistant product. LLMs are deep neural networks made of billions of numerical parameters (or weights, see below) that learn the relationships between words and phrases and create a representation of language, a sort of multidimensional map of words. These models are created from encoding the patterns they find in billions of books, articles, and transcripts. When you prompt an LLM, the model generates the most likely pattern that fits the prompt. It then evaluates the most probable next word after the last one based on what was said before. Repeat, repeat, and repeat. (See: Neural network) Neural network A neural network refers to the multi-layered algorithmic structure that underpins deep learning — and, more broadly, the whole boom in generative AI tools following the emergence of large language models.  Although the idea of taking inspiration from the densely interconnected pathways of the human brain as a design structure for data processing algorithms dates all the way back to the 1940s, it was the much more recent rise of graphical processing hardware (GPUs) — via the video game industry — that really unlocked the power of this theory. These chips proved well suited to training algorithms with many more layers than was possible in earlier epochs — enabling neural network-based AI systems to achieve far better performance across many domains, including voice recognition, autonomous navigation, and drug discovery. (See: Large language model [LLM]) Training Developing machine learning AIs involves a process known as training. In simple terms, this refers to data being fed in in order that the model can learn from patterns and generate useful outputs. Things can get a bit philosophical at this point in the AI stack — since, pre-training, the mathematical structure that’s used as the starting point for developing a learning system is just a bunch of layers and random numbers. It’s only through training that the AI model really takes shape. Essentially, it’s the process of the system responding to characteristics in the data that enables it to adapt outputs towards a sought-for goal — whether that’s identifying images of cats or producing a haiku on demand. It’s important to note that not all AI requires training. Rules-based AIs that are programmed to follow manually predefined instructions — for example, such as linear chatbots — don’t need to undergo training. However, such AI systems are likely to be more constrained than (well-trained) self-learning systems. Still, training can be expensive because it requires lots of inputs — and, typically, the volumes of inputs required for such models have been trending upwards. Hybrid approaches can sometimes be used to shortcut model development and help manage costs. Such as doing data-driven fine-tuning of a rules-based AI — meaning development requires less data, compute, energy, and algorithmic complexity than if the developer had started building from scratch. [See: Inference] Transfer learning A technique where a previously trained AI model is used as the starting point for developing a new model for a different but typically related task – allowing knowledge gained in previous training cycles to be reapplied.  Transfer learning can drive efficiency savings by shortcutting model development. It can also be useful when data for the task that the model is being developed for is somewhat limited. But it’s important to note that the approach has limitations. Models that rely on transfer learning to gain generalized capabilities will likely require training on additional data in order to perform well in their domain of focus (See: Fine tuning) Weights Weights are core to AI training, as they determine how much importance (or weight) is given to different features (or input variables) in the data used for training the system — thereby shaping the AI model’s output.  Put another way, weights are numerical parameters that define what’s most salient in a dataset for the given training task. They achieve their function by applying multiplication to inputs. Model training typically begins with weights that are randomly assigned, but as the process unfolds, the weights adjust as the model seeks to arrive at an output that more closely matches the target. For example, an AI model for predicting housing prices that’s trained on historical real estate data for a target location could include weights for features such as the number of bedrooms and bathrooms, whether a property is detached or semi-detached, whether it has parking, a garage, and so on.  Ultimately, the weights the model attaches to each of these inputs reflect how much they influence the value of a property, based on the given dataset. Topics

NVIDIA has released Llama Nemotron Nano 4B, an open-source reasoning model designed to deliver strong performance and efficiency across scientific tasks, programming, symbolic math, function calling, and instruction following—while being compact enough for edge deployment. With just 4 billion parameters, it achieves higher accuracy and up to 50% greater throughput than comparable open models with up to 8 billion parameters, according to internal benchmarks. The model is positioned as a practical foundation for deploying language-based AI agents in resource-constrained environments. By focusing on inference efficiency, Llama Nemotron Nano 4B addresses a growing demand for compact models capable of supporting hybrid reasoning and instruction-following tasks outside traditional cloud settings. Model Architecture and Training Stack Nemotron Nano 4B builds upon the Llama 3.1 architecture and shares lineage with NVIDIA’s earlier “Minitron” family. The architecture follows a dense, decoder-only transformer design. The model has been optimized for performance in reasoning-intensive workloads while maintaining a lightweight parameter count. The post-training stack for the model includes multi-stage supervised fine-tuning on curated datasets for mathematics, coding, reasoning tasks, and function calling. In addition to traditional supervised learning, Nemotron Nano 4B has undergone reinforcement learning optimization using Reward-aware Preference Optimization (RPO), a method intended to enhance the model’s utility in chat-based and instruction-following environments. This combination of instruction tuning and reward modeling helps align the model’s outputs more closely with user intent, particularly in multi-turn reasoning scenarios. The training approach reflects NVIDIA’s emphasis on aligning smaller models to practical usage tasks that traditionally require significantly larger parameter sizes. Performance Benchmarks Despite its compact footprint, Nemotron Nano 4B exhibits robust performance in both single-turn and multi-turn reasoning tasks. According to NVIDIA, it provides 50% higher inference throughput compared to similar open-weight models within the 8B parameter range. The model supports a context window of up to 128,000 tokens, which is particularly useful for tasks involving long documents, nested function calls, or multi-hop reasoning chains. While NVIDIA has not disclosed full benchmark tables in the Hugging Face documentation, the model reportedly outperforms other open alternatives in benchmarks across math, code generation, and function calling precision. Its throughput advantage suggests it can serve as a viable default for developers targeting efficient inference pipelines with moderately complex workloads. Edge-Ready Deployment One of the core differentiators of Nemotron Nano 4B is its focus on edge deployment. The model has been explicitly tested and optimized to run efficiently on NVIDIA Jetson platforms and NVIDIA RTX GPUs. This enables real-time reasoning capabilities on low-power embedded devices, including robotics systems, autonomous edge agents, or local developer workstations. For enterprises and research teams concerned with privacy and deployment control, the ability to run advanced reasoning models locally—without relying on cloud inference APIs—can provide both cost savings and greater flexibility. Licensing and Access The model is released under the NVIDIA Open Model License, which permits commercial usage. It is available through Hugging Face at huggingface.co/nvidia/Llama-3.1-Nemotron-Nano-4B-v1.1, with all relevant model weights, configuration files, and tokenizer artifacts openly accessible. The license structure aligns with NVIDIA’s broader strategy of supporting developer ecosystems around its open models. Conclusion Nemotron Nano 4B represents NVIDIA’s continued investment in bringing scalable, practical AI models to a broader development audience—especially those targeting edge or cost-sensitive deployment scenarios. While the field continues to see rapid progress in ultra-large models, compact and efficient models like Nemotron Nano 4B provide a counterbalance, enabling deployment flexibility without compromising too heavily on performance. Check out the Model on Hugging Face. All credit for this research goes to the researchers of this project. Also, feel free to follow us on Twitter and don’t forget to join our 95k+ ML SubReddit and Subscribe to our Newsletter. Asif RazzaqWebsite |  + postsBioAsif Razzaq is the CEO of Marktechpost Media Inc.. As a visionary entrepreneur and engineer, Asif is committed to harnessing the potential of Artificial Intelligence for social good. His most recent endeavor is the launch of an Artificial Intelligence Media Platform, Marktechpost, which stands out for its in-depth coverage of machine learning and deep learning news that is both technically sound and easily understandable by a wide audience. The platform boasts of over 2 million monthly views, illustrating its popularity among audiences.Asif Razzaqhttps://www.marktechpost.com/author/6flvq/A Coding Implementation to Build an AI Agent with Live Python Execution and Automated ValidationAsif Razzaqhttps://www.marktechpost.com/author/6flvq/Step-by-Step Guide to Build a Customizable Multi-Tool AI Agent with LangGraph and Claude for Dynamic Agent CreationAsif Razzaqhttps://www.marktechpost.com/author/6flvq/A Comprehensive Coding Guide to Crafting Advanced Round-Robin Multi-Agent Workflows with Microsoft AutoGenAsif Razzaqhttps://www.marktechpost.com/author/6flvq/Microsoft AI Introduces Magentic-UI: An Open-Source Agent Prototype that Works with People to Complete Complex Tasks that Require Multi-Step Planning and Browser Use