Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More Coding with the help of AI models continues to gain popularity, but many have highlighted issues that arise when developers rely on coding assistants.  However, researchers from MIT, McGill University, ETH Zurich, Johns Hopkins University, Yale and the Mila-Quebec Artificial Intelligence Institute have developed a new method for ensuring that AI-generated codes are more accurate and useful. This method spans various programming languages and instructs the large language model (LLM) to adhere to the rules of each language. The group found that by adapting new sampling methods, AI models can be guided to follow programming language rules and even enhance the performance of small language models (SLMs), which are typically used for code generation, surpassing that of large language models. In the paper, the researchers used Sequential Monte Carlo (SMC) to “tackle a number of challenging semantic parsing problems, guiding generation with incremental static and dynamic analysis.” Sequential Monte Carlo refers to a family of algorithms that help figure out solutions to filtering problems.  João Loula, co-lead writer of the paper, said in an interview with MIT’s campus paper that the method “could improve programming assistants, AI-powered data analysis and scientific discovery tools.” It can also cut compute costs and be more efficient than reranking methods.  The researchers noted that AI-generated code can be powerful, but it can also often lead to code that disregards the semantic rules of programming languages. Other methods to prevent this can distort models or are too time-consuming.  Their method makes the LLM adhere to programming language rules by discarding code outputs that may not work early in the process and “allocate efforts towards outputs that more most likely to be valid and accurate.” Adapting SMC to code generation The researchers developed an architecture that brings SMC to code generation “under diverse syntactic and semantic constraints.”  “Unlike many previous frameworks for constrained decoding, our algorithm can integrate constraints that cannot be incrementally evaluated over the entire token vocabulary, as well as constraints that can only be evaluated at irregular intervals during generation,” the researchers said in the paper.  Key features of adapting SMC sampling to model generation include proposal distribution where the token-by-token sampling is guided by cheap constraints, important weights that correct for biases and resampling which reallocates compute effort towards partial generations. The researchers noted that while SMC can guide models towards more correct and useful code, they acknowledged that the method may have some problems. “While importance sampling addresses several shortcomings of local decoding, it too suffers from a major weakness: weight corrections and expensive potentials are not integrated until after a complete sequence has been generated from the proposal. This is even though critical information about whether a sequence can satisfy a constraint is often available much earlier and can be used to avoid large amounts of unnecessary computation,” they said.  Model testing To prove their theory, Loula and his team ran experiments to see if using SMC to engineer more accurate code works.  These experiments were:  Python Code Generation on Data Science tasks, which used Llama 3 70B to code line-by-line and test early versions  Text-to-SQL Generation with Llama 3 8B- Instruct Goal Inference in Planning Tasks to predict an agent’s goal condition, and also used Llama 3 8B Molecular Synthesis for drug discovery They found that using SMC improved small language models, improved accuracy and robustness, and outperformed larger models.  Why is it important AI models have made engineers and other coders work faster and more efficiently. It’s also given rise to a whole new kind of software engineer: the vibe coder. But there have been concerns over code quality, lack of support for more complex coding and compute costs for simple code generation. New methods, such as adapting SMC, may make AI-powered coding more useful and enable engineers to trust the code generated by models more.  Other companies have explored ways to improve AI-generated code. Together AI and Agentica released DeepCoder-14B, which harnesses fewer parameters. Google also improved its Code Assist feature to help enhance code quality.  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.

Training LLMs on trajectories of reasoning and tool use makes them superior at multi-step reasoning tasks.Read More

Researchers from MIT, Yale, McGill University and others found that adapting the Sequential Monte Carlo algorithm can make AI-generated code better.Read More

Training LLMs on trajectories of reasoning and tool use makes them superior at multi-step reasoning tasks.Read More

Not long ago, a cooperation partner casually approached me with an AI use case at their organization. They wanted to make their onboarding process for new staff more efficient by using AI to answer the repetitive questions of newcomers. I suggested a practical chat approach that would integrate their internal documentation, and off they went with an air of confidence, planning to “talk to their IT team” to move forward. From experience, I knew that this kind of optimism was brittle. The average IT team isn’t equipped to implement a full end-to-end Ai Application on their own. And so it was: months later, they were stuck. Their system was frustratingly slow, and it also became clear they had misread the users’ actual needs during development. New employees were asking different questions than those the system had been tuned for. Most users bounced after a couple of attempts and never came back. Fixing these issues would require rethinking their entire architecture and data strategy, but damage was already done. Employees were frustrated, leadership had taken notice, and the initial excitement around AI had faded into skepticism. Arguing for another extensive development phase would be difficult, so the case was quietly shelved. This story is far from unique. Great marketing by AI companies creates an illusion of accessibility around AI, and companies jump into initiatives without fully grasping the challenges ahead. In reality, specialized expertise is needed to create a solid AI strategy and implement any more or less custom use case in your company. If this expertise is not available internally, you need to get it from external partners or providers. That doesn’t mean that you need to buy everything — that would be like having $100 and spending it at the restaurant instead of going to the supermarket. The first option will address your hunger on the spot, but the second one will ensure you have something to eat for a week. So, how can you get started, and who should implement your first AI projects? Here is my take: Forget build-or-buy and focus on partnering and learning instead. I deeply believe that most companies should build AI expertise internally — this will provide them with more bandwidth in their AI strategy and activities in the future. At the same time, AI is a complex craft that takes time to master, and failure is omnipresent (according to this report by RAND Corporation, more than 80% of AI initiatives fail). Learning from failure is nice in theory, but in reality, it leads to waste of time, resources, and credibility. In order to achieve AI maturity efficiently, companies should consider cooperating with trusted partners who are ready to share their expertise. A realistic and careful setup will not only ensure a smoother technical implementation but also address the people- and Business-related aspects of your AI strategy. In the following, I will first outline the rough basics (inputs, outputs, and trade-offs) of build-or-buy decisions in AI. Then, you will learn about a more differentiated partnering approach. It combines building and buying while reinforcing your internal learning curve. Finally, I will close with some practical observations and advice on partnering in AI. Note: If you are interested in more actionable AI insights, please check out my newsletter AI for Business! The basics of build-or-buy decisions in AI To start, let’s break down a classical build-or-buy decision into two parts: the inputs — what you should assess upfront — and the outputs — what each choice will mean for your business down the line. Inputs To prepare the decision, you need to evaluate your internal capabilities and the requirements of the use case. These factors will shape how realistic, risky, or rewarding each option might be: AI maturity of your organization: Consider your internal technical capabilities, such as skilled AI talent, existing reusable AI assets (e.g. datasets, pre-built models, knowledge graphs), and adjacent technical skills that can be transferred into the AI space (e.g. data engineering, analytics). Also count in how proficient users are at interacting with AI and dealing with its uncertainties. Invest in upskilling and dare to build more as your AI maturity grows. Domain expertise needs: How deeply must the solution reflect your industry-specific knowledge? In use cases requiring expert human intuition or regulatory familiarity, your internal domain experts will play a crucial role. They should be part of the development process, whether through building internally or partnering closely with an external provider. Technical complexity of the use case: Not all AI is created equal. A project that relies on existing APIs or foundation models is vastly simpler than one that demands training a custom model architecture from scratch. High complexity increases the risk, resource requirements, and potential delays of a build-first approach. Value and strategic differentiation: Is the use case core to your strategic advantage or more of a support function? If it’s unique to your industry (or even company) and will increase competitive differentiation, building or co-developing may offer more value. By contrast, for a a standard use case (e.g. document classification, forecasting), buying will likely deliver faster, more cost-effective results. Consequences of build-or-buy decisions Once you’ve assessed your inputs, you’ll want to map out the downstream impact of your build-or-buy choice and evaluate the trade-offs. Here are seven dimensions that will influence your timelines, costs, risks, and outcomes: Customization: The degree to which the AI solution can be tailored to the organization’s specific workflows, goals, and domain needs. Customization often determines how well the solution fits unique business requirements. Ownership: Intellectual property (IP) rights and control over the underlying AI models, code, and strategic direction. Building internally offers full ownership, while buying typically involves licensing another party’s technology. Data security: Covers how data is handled, where it resides, and who has access. In regulated or sensitive environments, data privacy and compliance are central concerns, particularly when data may be shared with or processed by external vendors. Cost: Encompasses both the initial investment and ongoing operational expenses. Building involves R&D, talent, infrastructure, and long-term maintenance, whereas buying may require licensing, subscriptions, or cloud usage fees. Time-to-market: Measures how quickly the solution can be deployed and start delivering value. Fast deployment is often critical in competitive or dynamic markets; delays can lead to lost opportunities. Support & maintenance: Involves who is responsible for updates, scaling, bug fixes, and ongoing model performance. Internal builds require dedicated resources for upkeep, while external solutions often include support services. AI learning curve: Reflects the complexity of acquiring AI expertise and operationalizing it within the organization. Building in-house often comes with lots of trial-and-error and brittle outcomes because the team doesn’t possess foundational AI knowledge. On the other hand, buying or partnering can accelerate learning via guided expertise and mature tooling and create a solid basis for future AI activities. Now, in practice, binary build-or-buy thinking often leads to unresolvable trade-offs. Take the onboarding use case mentioned earlier. One reason the team leaned toward building was a need to keep their company data confidential. At the same time, they didn’t have the internal AI expertise to develop a production-ready chat system. They would likely have been more successful by outsourcing the chat architecture and ongoing support while building their database internally. Thus, you shouldn’t decide to build or buy at the level of the entire AI system. Instead, break it down into components and evaluate each one based on your capabilities, constraints, and strategic priorities. Towards a handshake between domain and AI expertise At the component level, I encourage you to differentiate build-or-buy decisions through the lens of expertise requirements. Most B2B AI systems combine two kinds of expertise: domain expertise, which lives within your company, and technical AI expertise, which can be brought in through an external partner if you don’t (yet) have specialized AI skills. In the following, I will examine the expertise needs for each of the core components of an AI system (cf. this article for an explanation of the components). Table 1: Expertise needs and collaboration formats for the components of AI systems Business opportunity: Framing the right AI problems Did you know that the #1 reason for AI project failure is not technical — it’s choosing the wrong problem to solve (cf. The Root Causes of Failure for Artificial Intelligence Projects and How They Can Succeed)? You might be surprised — after all, your expert teams understand their problems deeply. The point is, they don’t have the means to connect the dots between their pain points and AI technology. Here are some of the most common failure patterns: Vague or unsuitable problem framing: Is this a task that AI is actually good at? Missing effort/ROI estimation: Is the outcome worth the time and resources for AI development and deployment? Unrealistic expectations: What does “good enough” mean for an imperfect AI? On the other hand, there are many organizations that use AI for its own sake and create solutions in search of a problem. This burns resources and erodes confidence internally. A good AI partner helps assess which business processes are ripe for AI intervention, estimates potential impact, and models how AI might deliver value. Both parties can shape a focused, high-impact use case through joint discovery workshops, design sprints, and exploratory prototyping. Data: The fuel of your AI system Clean, well-structured domain data is a core asset. It encodes your process knowledge, customer behavior, system performance, and more. But raw data alone isn’t enough — it needs to be transformed into meaningful learning signals. That’s where AI expertise comes in to design pipelines, choose the right data representations, and align everything with AI’s learning goals. Often, this includes data labeling — annotating examples with the signals a model needs to learn from. It might seem tedious, but resist the urge to outsource it. Labeling is one of the most context-sensitive parts of the pipeline, and it requires domain expertise to be done right. In fact, many fine-tuning tasks today perform best on small but high-quality datasets — so work closely with your AI partner to keep the effort focused and manageable. Data cleaning and preprocessing is another area where experience makes all the difference. You’ve probably heard the saying: “Most of a data scientist’s time is spent cleaning data.” That doesn’t mean it should be slow. With engineers who are experienced in your data modality (text, numbers, images…), this process can be dramatically accelerated. They’ll instinctively know which preprocessing techniques to apply and when, turning weeks of trial and error into hours of productive setup. Intelligence: AI models and architectures This is where most people think AI projects begin — but it’s only the middle of the story. Deep AI expertise is needed to select or fine-tune models, evaluate performance, and design system architectures. For example, should your use case use a pre-trained model? Do you need a multimodel setup? What evaluation metrics make sense? In more complex systems, different AI components such as models and knowledge bases can be combined into a multi-step workflow. Domain expertise comes in during system validation and evaluation. Experts and future users need to check if AI outputs make sense and align with their real-world expectations. A model might be statistically strong, but operationally useless if its outputs don’t map to business logic. When designing compound systems, domain experts also need to make sure that the system setup mirrors their real-world processes and needs. Tailoring AI models and building a custom AI architecture is your “co-pilot” phase: AI teams architect and optimize, while domain teams steer and refine based on business goals. Over time, the goal is to build shared ownership of system behavior. Case study: Building with AI expertise support in insurance At a leading insurance provider, the data science team was tasked with building a claims risk prediction system — a project they wanted to keep in-house to retain full ownership and align closely with proprietary data and workflows. However, early prototypes ran into performance and scalability issues. That’s where my company Anacode came in as an architectural and strategic partner. We helped the internal team evaluate model candidates, design a modular architecture, and set up reproducible ML pipelines. Just as importantly, we ran targeted upskilling sessions focused on model evaluation, MLOps, and responsible AI practices. Over time, the internal team gained confidence, reworked earlier prototypes into a robust solution, and fully took over operations. The result was a system they owned completely, while the expert guidance we provided during the project had also elevated their internal AI capabilities. User experience: Delivering AI value through the user interface This one is tricky. With a few exceptions, neither domain experts nor deep AI engineers are likely to design an experience that is truly intuitive, efficient, and enjoyable for real users. Ideally, you can bring in specialized UX designers. If these are not available, look for people from adjacent disciplines who have a natural feel for user experience. Today, a lot of AI tools are available to support UX design and prototyping, so taste matters more than technical craft. Once you have the right people, you need to feed them with inputs from both sides: Backend: AI experts provide insight into how the system works internally — its strengths, limitations, levels of certainty — and support the design of elements like explanations, uncertainty indicators, and confidence scores (cf. this article on building trust in AI through UX). Frontend: Domain experts understand the users, their workflows, and their pain points. They help validate user flows, highlight friction, and propose refinements based on how people actually interact with the system. Focus on fast iteration and be prepared for some erring around. AI UX is an emerging field, and there’s no settled formula for what “great” looks like. The best experiences arise from tight, iterative feedback loops, where design, testing, and refinement happen continuously, absorbing inputs from both domain experts and AI specialists. Support and maintenance: Keeping AI alive Once deployed, AI systems require close monitoring and continuous improvement. Real-world user behavior often diverges from test environments and changes over time. This inherent uncertainty means your system needs to be actively watched, so that issues can be identified and addressed early. The technical infrastructure for monitoring — including performance tracking, drift detection, automated retraining, and MLOps pipelines — is typically set up by your AI partner. Once in place, many day-to-day monitoring tasks don’t require deep technical skills. What they do require is domain expertise: understanding whether model outputs still make sense, noticing subtle shifts in usage patterns, and knowing when something “feels off.” A well-designed support phase is more than just operational — it can be a critical learning phase for your internal teams. It creates space for gradual skill-building, deeper system understanding, and ultimately, a smoother path toward taking greater ownership of the AI system over time. Thus, rather than framing AI implementation as a binary build-or-buy decision, you should view it as a mosaic of activities. Some of these are deeply technical, while others are closely tied to your business context. By mapping responsibilities across the AI lifecycle, you can: Clarify which roles and skills are essential to success Identify capabilities you already have in-house Spot gaps where external expertise is most valuable Plan for knowledge transfer and long-term ownership If you want to dive deeper into the integration of domain expertise, check out my article Injecting domain expertise into your AI systems. Importantly, the line between “domain” and “AI” expertise is not fixed. You might already have team members experimenting with machine learning, or others eager to grow into more technical roles. With the right partnership model and upskilling strategy, you can evolve towards AI autonomy, gradually taking on more responsibility and control as your internal maturity grows. In partnering, start early and focus on communication By now, you know that build-or-buy decisions should be made at the level of individual components of your AI system. But if you don’t yet have AI expertise on your team, how can you envision what your system and its components will eventually look like? The answer: start partnering early. As you begin shaping your AI strategy and design, bring in a trusted partner to guide the process. Choose someone you can communicate with easily and openly. With the right collaboration from the start, you’ll increase your chances of navigating AI challenges smoothly and successfully. Choose an AI partner with foundational AI expertise Your AI partner should not just deliver code and technical assets, but help your organization learn and grow during your cooperation. Here are a few common types of external partnerships, and what to expect from each: Outsourcing: This model abstracts away the complexity — you get results quickly, like a dose of fast carbs. While it’s efficient, it rarely delivers long-term strategic value. You end up with a tool, not with stronger capabilities. Academic partnerships: Great for cutting-edge innovation and long-term research, but often less suited for an AI system’s real-world deployment and adoption. Advisory partnerships: In my view, the most promising path, especially for companies that already have a tech team and want to develop their AI acumen. A good advisor empowers your engineers, helps them avoid costly missteps, and brings practical, experience-driven insight to questions like: What’s the right tech stack for our use case? How do we curate our data to boost quality and kick off a powerful data flywheel How do we scale without compromising trust and governance? A detailed partner selection framework is beyond the scope of this article, but here’s one piece of hard-earned advice: Be wary of IT outsourcers and consultancies that suddenly added “AI” to their offering after the GenAI boom in 2022. They might charm you with fancy buzzwords, but if AI isn’t in their DNA, you may end up paying for their learning curve rather than benefiting from complementary expertise. Choose a partner who’s done the hard work already and is ready to transfer that expertise to you. Double down on communication and alignment Effective communication and stakeholder alignment are critical in partnering models. Here are some important communication roles to get right in your company: Leadership and domain experts must identify and clearly communicate the business problems worth solving (more on best practices for AI ideation here). End users need to share their needs early, give feedback during usage, and ideally become co-creators in shaping the AI experience. IT and governance teams must ensure compliance, security, and safety while enabling, not blocking, AI innovation. Keep in mind: these capabilities don’t appear fully formed. In AI projects, the risk of misalignment and unproductive silos is high. AI is still a relatively new field, and the terminology alone can create confusion. If you’ve ever found yourself in a debate about the difference between “AI” and “machine learning,” you know what I mean. And if you haven’t, I encourage you to try at your next get-together with your colleagues. It can be just as slippery as that conversation with your significant other that starts with “we need to talk.” Aim for a rapprochement from both sides to iron out ambiguities and disconnects. Your internal teams should invest in upskilling and build a basic understanding of AI concepts. On the other hand, your AI partners must meet you halfway. They should skip the jargon and use clear, business-oriented language that your team can actually work with. Effective collaboration starts with shared understanding. Conclusion The real question isn’t “Should we build or buy AI?” — it’s “How do we grow our AI capability in a way that balances speed, control, and long-term value?” The answer lies in recognizing AI as a blend of technology and expertise, where success depends on matching the right resources to the right tasks. For most organizations, the smartest path forward is partnering — combining your domain strengths with external AI expertise to build faster, learn faster, and eventually own more of your AI journey. What you can do next: Map your AI use case against your internal capabilities. Be honest about the gaps. Choose partners who transfer knowledge, not just deliverables. Identify which components to build, buy, or co-create. You don’t need to make a binary choice. Upskill your team as you go. Every project should make you more capable and autonomous, not more dependent on your partner’s assets and skills. Start with focused pilots that create value and momentum for internal learning. By taking a strategic, capability-building approach today, you lay the groundwork for becoming an AI-capable — and eventually AI-driven — organization in the long term. Further readings Singla, A., Sukharevsky, A., Ellencweig, B., Krzyzaniak, M., & Song, J. (2024, May 22). Strategic alliances for Gen AI: How to build them and make them work. McKinsey & Company. Liebl, A., Hartmann, P., & Schamberger, M. (2023, November 23). Enterprise guide for make-or-buy decisions [White paper]. appliedAI Initiative. Gartner. (n.d.). Deploying AI: Should your organization build, buy or blend? Gartner. The post Enterprise AI: From Build-or-Buy to Partner-and-Grow appeared first on Towards Data Science.

Oblivion Remastered features dozens of side quests to complete, and some are certainly easier than others. While there are more than a fair share of "fetch" quests in Cyrodiil, Unfriendly Competition is not one of them. This quest can begin in a couple of ways, but you'll need to make your way to the Imperial City to start it. Unfriendly Competition sees you uncover a slimy scheme that allows a merchant to vastly undercut their prices and bankrupt every other merchant in the city.To see a full walkthrough of how you do that and how to complete Unfriendly Competition in Oblivion Remastered, take a look below. You can also check out our full walkthrough of the entire game if you need a complete guide to the remaster. Starting Unfriendly CompetitionUnfriendly Competition begins by either speaking to Ogier Georick or Jensine in the Market District of the Imperial City. Both of these merchants are suspicious of another merchant, named Thoronir, who is selling the same goods as them, but at vastly lower prices. Jensine even says that Thoronir is losing money by pricing their items so low, so both she and Ogier believe that the merchant is stealing their goods.If you speak to Ogier first, he'll tell you to go and speak with Jensine to learn more about the situation. If you speak with Jensine first, she'll fill you in on all of the details and what you need to do. Either way, after speaking with Jensine, your next task is to investigate Thoronir. Speaking with ThoronirThoronir is located at a shop called the Copious Coinpurse, and it's also found in the Market District in the Imperial City. Make sure to enter the shop between the hours of 8 am - 7 pm, as those are the hours the store is open.When you enter the Copius Coinpurse, Thoronir will be standing behind the counter and invite you inside. When you speak to the shop owner, you can ask about their inventory, and they'll respond with a cliche line about their goods being the best and cheapest in town. That's all you're going to get out of Thoronir for now, so to find out more, exit the Copius Coinpurse and wait outside to follow the merchant after they close up shop. Following Thoronir after hoursTo follow Thoronir after they close up shop, go right outside the doors to the Copius Coinpurse and press the "Wait" button. This is "T" by default on PC. Wait until roughly 7-8 PM at night, and when your game reloads, you will see Thoronir exiting the shop and walking to the left down the cobblestone road. If you miss Thoronir, simply track the Unfriendly Competition quest from your journal, and you should have a red or green arrow on your compass. A green arrow indicates Thoronir is currently moving, while a red one means they're stationary.Either way, Thoronir ends up at a shop called The Merchants Inn, which is located just a little ways away from the Copious Coinpurse. In all honesty, though, you don't need to follow Thoronir into The Merchants Inn. Eavesdropping on ThoronirThoronir sits around The Merchants Inn until roughly midnight. You can simply skip ahead to right before midnight by using the wait function and then go to a different area in the Market District.Thoronir eventually goes from the inn to a garden area with a well in it. This area is located to the west of The Merchants Inn, and you can see it more clearly by zooming in all the way on your map. The area you're looking for has a stone well right in the center of it, and it's surrounded by greenery.At the garden/well area, wait for a man named Agarmir to show up and start waiting by a tree stump in one of the corners. Soon after Agarmir shows up, Thoronir will make his way to the same spot. You need to eavesdrop on the two of them, but they won't start talking to each other until you're well hidden, but close enough to listen in. I was able to trigger their conversation by hiding behind the tree and the pink bush right in front of them. It seems as long as you're close, the two will start talking.By eavesdropping on the conversation, you learn that Agarmir is in charge of the scheme that Thoronir's shop is benefiting from. So, your new job is to tail Agarmir after the two split up. Tailing Agarmir and breaking and enteringAfter Thoronir and Agarmir stop talking, they'll both exit through the south entrance of the garden. All you need to do is follow Agarmir as he goes through a few different Imperial City districts.Eventually, Agarmir ends up in the Talos Plaza area of the city and enters his house. You need to use the wait function outside his house to get the time to roughly 7 pm. If you don't arrive at the right time, you'll receive a message stating that Agarmir's door is locked.When Agarmir isn't in the house, you can lockpick the door, which is "Average" difficulty. Inside the house, make your way to the back and lockpick another Average lock to enter the basement. Here, go to the back corner and pick up the Macabre Manifest book, which is sitting in front of some candles on a table. Make sure you pick up the book and don't just read it. According to the book and the basement's contents, Agarmir has been selling goods to Thoronir that he recently dug up from dead bodies. You now need to take the book back to Thoronir and confront them. Confronting ThoronirWith the Macabre Manifest in your inventory, go to Thoronir in the Copious Coinpurse and confront them about what Agarmir has been doing. Thoronir genuinely seems like they had no idea about where the goods were coming from. The merchant goes on to say that Agarmir is supposed to be at a graveyard today, and he asks you to go and confront him. Putting an end to AgarmirThanks to Thoronir's tip, you can travel to the Trentius Family Mausoleum in the Imperial City. Inside, you find Agarmir and a guard messing with a grave. Agarmir confronts you and explains that he laid this trap for you, and challenges you to a fight. You need to take out both Agarmir and the guard, which can be difficult, as Agarmir is wielding a sword called the Debaser, which drains your willpower and endurance on hits. Once Agarmir and the guard are dead, loot their bodies, and then have a search around the mausoleum. You need to find an item that proves Agarmir's guilt in this scheme. If you travel to the back of the room, find the central coffin, and just to the left of it, you can pick up Agarmir's Shovel. This is all you need to prove his guilt. However, before you leave, you can also check Calliben Trentius' coffin, which is on the left side of the room. Just behind the stone statue inside the coffin is a weapon called Calliben's Grim Retort, which is quite a powerful mace in Oblivion Remastered. If you played your cards right, you're walking out with two strong weapons. Wrapping up Unfriendly CompetitionWith Agarmir's Shovel, return to Thoronir, who is either at The Merchants Inn or the Copious Coinpurse, and show him the shovel. They'll explain they're donating all of their profits to the temple and joining the Merchants' Society. As a reward for your efforts, Thoronir gives you the Weatherward Circlet.Finally, you can return to Jensine and show her the shovel as well. Jensine gives you 100 gold for your help in resolving the matter with Thoronir. This finishes the Unfriendly Competition quest in Oblivion Remastered.

Although The Elder Scrolls IV: Oblivion Remastered overlaps with the fan-made remake mod Skyblivion, Bethesda still acknowledged the modders' efforts by giving them code for the new remaster.Huge thanks to our friends @BethesdaStudios for their continued support of #Skyblivion! As massive fans, we're beyond grateful for the generous gift of Oblivion Remastered game keys for our entire modding team! This means so much to us. Thank you for everything, @bethesda. pic.twitter.com/vLsuzhdgkX — Skyblivion (@TESRSkyblivion) April 22, 2025 Skyblivion is an upcoming project by the modding group TESRenewal. Its goal is to remake the 2006 RPG The Elder Scrolls IV: Oblivion entirely within The Elder Scrolls V: Skyrim's game engine. It's an impressive effort that has been in the works for several years. In February, TESRenewal suggested that it could be released this year. Shortly thereafter, rumors of an official Oblivion remaster and remake began to swirl.As it became clear that The Elder Scrolls IV: Oblivion Remastered was real and launching soon, Skyblivion Project Lead Rebelzize said, "for the community it's a win-win as you get twice the amount of Oblivion this year," of the situation. Earlier today, Bethesda and Virtuos officially announced and released The Elder Scrolls IV: Oblivion Remastered. Rather than issuing a cease and desist for Skyblivion, Bethesda acknowledged the team's work and provided the members of TESRenewal with free codes for the remaster.Continue Reading at GameSpot

Honkai: Star Rail replaced traditional weapons with Light Cones and reshaped gear into something far more specialized and polarizing in the process. Instead of an open system where players can experiment across weapon types or mix gear by stat priority, Light Cones are path-locked, purpose-built, and brutally narrow in scope. While that design works well for Star Rail because its combat is slow, surgical, and turn-based, and while the mechanic has the potential to translate well beyond Star Rail’s format, things could go either way for other titles.

Vivian is a versatile sub-DPS character in Zenless Zone Zero, specializing in off-field damage while amplifying Corruption and Disorder effects through her featherblooms. Her hybrid role combines utility and offense, making her an adaptable choice for various team setups, despite her needing more on-field time in comparison to a character like Burnice. To fully leverage her capabilities, she requires synergy with another Anomaly / Ether agent, which allows her Abloom damage to reach its full potential while activating her Additional Ability.

With franchise revival Onimusha: Way of the Sword in active development, there’s been a resurgence of interest in the PlayStation 2 Onimusha series. While a remaster of the original Onimusha: Warlords has been available for quite some time, it was only recently that Capcom announced that Onimusha 2: Samurai’s Destiny would also be getting a remaster, due out on May 23. Fans of the series rejoiced at the news, and for good reason. Onimusha 2 is a fan favorite, with a unique style and gameplay elements that make it a true standout among the series. We had the opportunity to interview Motohide Eshiro, the original game director and longtime Capcom veteran, about what players old and new can expect from Onimusha 2’s remaster. Play Video Making friends and influencing people What did the team learn from the original Onimusha remaster? We learned that we could make the controls more comfortable. Onimusha 2 originally only had ‘tank controls.’ We carried over the revised controls from the previous remaster to Onimusha 2 smoothly. Would these modernized controls affect the difficulty? No, it’s more about minimizing frustration. Classic fans can choose tank controls if they prefer, but modern players can more easily get into the game this way. As far as combat goes, we faithfully carried over the Issen critical system from the original game. It won’t be any easier to perform critical attacks, since we think the critical system was really finely balanced in the original game. Can you elaborate on one of the game’s defining features: the ally system? It’s one of the defining elements of Onimusha 2. You can give gifts to other characters and form relationships with them. In return, they’ll help out Jubei in battles. Sometimes, you can control these characters, as well. Was anything altered for this system in the remaster? Like Issen, I think the ally system was well-balanced, so I didn’t have to make any tweaks or drastic changes. We considered adding a gauge over the characters’ heads to indicate your relationship status, but that wouldn’t express the depth of the characters. So you need to judge, through their reaction, the facial expressions, the comments they give you, how they feel about the gift. Heaven or Hell? What are your thoughts on the new optional Hell difficulty where any hit means instant death? It’s definitely not for all players. This is something like a challenge from the developers, especially for those players who brag about their skills. We really want players to experience epic sword battles full of tension. I haven’t been able to complete this mode yet, but we figured that if even one person in the world can complete this game, that will be OK. Another team member has been able to complete this new hell mode, so we know it’s not impossible. So I hope that players, especially those who have confidence in their skills, will play it. Even if you don’t, you can progress a little at a time by frequently saving. Of course, if you complete it, that’ll be amazing! Can you tell us about the additions of mini-games like The Man in Black, Team Oni, and Puzzle Phantom Realm? Team Oni is a game where you go through stages, not just with Jubei but with your allies, and you can switch between them. It’s crucial to switch in allies and use items strategically. Puzzle Phantom Realm is a sequence of puzzles to solve. You’ll encounter puzzle sections in the normal game, but this is just back-to-back puzzles, testing your solving skills. Lastly, there’s ‘The Man in Black.’ Jubei has a special outfit where he’s dressed in black, head to toe. Instead of battling enemies, the goal is to collect films at various locations. These all have a different vibe from the main game, and are fun in distinct ways. Enduring Legacies Onimusha 2 involved many big names in Japanese entertainment in its creation: longtime TV and game scriptwriter Noboru Sugimura, famed artist and special effects director Keita Amemiya, and the likeness of legendary actor Yusaku Matsuda. How did this ensemble come together? We wanted the world to be a blend of Japanese and Western taste. Sugimura really had a deep knowledge and respect for the era of Japanese history this game takes place. He came up with so many great ideas for the ally characters. When Sugimura and I were discussing what kind of character design we wanted, he came up with the idea of having Keita Amemiya as the main artist of the game. He made really distinctive and compelling character designs, shaping the visuals of Onimusha 2. You’ll be able to see a lot of his work in the new Gallery mode. When we were finalizing the base model for Jubei, Miyuki Matsuda, the widow of Yusaku Matsuda, came to the Capcom office to guide the team. She was hands-on in guiding us on how to tweak the face model, so it would look just like Matsuda. I felt all the passion and love Miyuki still had toward Yusaku. That was really something memorable. Any final thoughts about the game? Onimusha 2 was the first title I directed, and I’m glad I could return to the remaster as a director again. We hope new players will enjoy all the quality-of-life features, which make it easier to get into this game. I hope you enjoy it, even if you haven’t played an Onimusha game before. Onimusha 2 Samurai’s Destiny launches on PlayStation May 23.