Latest news with #SakanaAI
Yahoo
15-07-2025
- Business
- Yahoo
Mitsubishi UFJ Financial Group (MUFG) Strengthens AI Strategy with New Partnership
Mitsubishi UFJ Financial Group, Inc. (NYSE:MUFG) is one of the 13 Best Japanese Stocks to Buy According to Hedge Funds. On May 19, Mitsubishi UFJ Financial Group, Inc. (NYSE:MUFG) and its consolidated subsidiary, MUFG Bank, announced a new partnership with Sakana AI, a company that specializes in AI research and development. Mitsubishi UFJ Financial Group, Inc. (NYSE:MUFG) has also appointed Ren Ito, the Chief Operating Officer of Sakana AI, as the group's AI Advisor. A smiling employee in front of a modern building surrounded by a vibrant cityscape. Sakana AI and MUFG Bank will form a long-term strategic partnership for a period of more than 3 years. Mitsubishi UFJ Financial Group, Inc. (NYSE:MUFG) plans to use Sakana AI's innovative technologies to solve management challenges and add more value to operations. Initially, the corporation will use the expertise of Sakana AI's 'The AI Scientist' to automate the creation of documents. Ren Ito will play an active role by advising Mitsubishi UFJ Financial Group, Inc. (NYSE:MUFG) on AI-related activities. He will help shape the corporation's AI strategy while also supporting networking and providing important information to the management team. In its current business plan, which began in April 2024, Mitsubishi UFJ Financial Group, Inc. (NYSE:MUFG) committed to improving its AI capabilities and data infrastructure to improve productivity and better serve customers. This partnership is part of the group's aim to strengthen its AI strategy. Mitsubishi UFJ Financial Group, Inc. (NYSE:MUFG) is one of the largest banking institutions in Japan and a leading global financial services group. It offers a wide range of services, including commercial banking, trust banking, securities, credit cards, consumer finance, asset management, and leasing. While we acknowledge the potential of MUFG as an investment, we believe certain AI stocks offer greater upside potential and carry less downside risk. If you're looking for an extremely undervalued AI stock that also stands to benefit significantly from Trump-era tariffs and the onshoring trend, see our free report on the best short-term AI stock. READ NEXT: 10 Best American Semiconductor Stocks to Buy Now and 11 Best Fintech Stocks to Buy Right Now. Disclosure: None. This article is originally published at Insider Monkey. Error in retrieving data Sign in to access your portfolio Error in retrieving data Error in retrieving data Error in retrieving data Error in retrieving data
Geeky Gadgets
25-06-2025
- Science
- Geeky Gadgets
Forget Bigger Models : This AI Breakthrough from Sakana AI Thinks Smarter
What if the key to unlocking the next era of artificial intelligence wasn't building bigger, more powerful models, but teaching smaller ones to think smarter? Sakana AI's new 'Reinforcement Learned Teacher' (RLT) model is poised to challenge everything we thought we knew about reinforcement learning. By shifting the focus from task-solving to teaching, this innovative approach promises to slash training costs, accelerate development timelines, and make innovative AI accessible to a wider audience. Imagine training an advanced AI system not in months, but in a single day—at a fraction of the cost. This isn't just a technical breakthrough; it's a reimagining of how we approach AI development altogether. In this perspective, Wes Roth explores how the Sakana RLT model is reshaping the landscape of reinforcement learning and why it matters. You'll discover how this teaching-first framework enables smaller, cost-efficient models to outperform their larger, resource-hungry counterparts, and why this shift could provide widespread access to AI innovation. From self-improving AI systems to fantastic applications in education, healthcare, and beyond, the implications of this approach are profound. As we unpack the mechanics and potential of RLT, one question lingers: Could teaching, not brute computational force, be the key to AI's future? Transforming AI Training Understanding Reinforcement Learning Reinforcement learning has long been a cornerstone of AI development. It operates by training models to solve tasks through a process of trial and error, rewarding successful outcomes to encourage desired behaviors. While effective in specific applications, traditional RL methods are often resource-intensive, requiring substantial computational power, time, and financial investment. For instance, training a large-scale RL model can cost upwards of $500,000 and take several months to complete. These high costs and extended timelines have historically restricted RL's accessibility, particularly for smaller research teams and independent developers. As a result, the potential of RL has remained largely confined to organizations with significant resources. How the RLT Model Transforms the Process Sakana AI's RLT model reimagines reinforcement learning by prioritizing teaching over direct task-solving. Instead of training a single model to perform a task, the RLT framework trains smaller, efficient teacher models to generate detailed, step-by-step explanations. These explanations are then used to train student models, significantly improving their performance. The teacher models are evaluated not on their ability to solve tasks directly but on how effectively their explanations enhance the learning outcomes of the student models. This creates a collaborative dynamic between teacher and student models, allowing a more efficient and scalable training process. By focusing on teaching, the RLT model reduces the need for extensive computational resources while maintaining high levels of performance. How Sakana AI's RLT Model is Changing Reinforcement Learning Watch this video on YouTube. Take a look at other insightful guides from our broad collection that might capture your interest in Reinforcement Learning (RL). Key Advantages of the RLT Approach The RLT model addresses many of the limitations associated with traditional RL methods. Its benefits include: Cost Efficiency: Smaller teacher models significantly reduce training expenses. While traditional RL training can cost $500,000, RLT training can be completed for as little as $10,000, making it far more accessible. Smaller teacher models significantly reduce training expenses. While traditional RL training can cost $500,000, RLT training can be completed for as little as $10,000, making it far more accessible. Faster Training: Tasks that previously required months of training can now be completed in a single day using standard hardware, drastically reducing development timelines. Tasks that previously required months of training can now be completed in a single day using standard hardware, drastically reducing development timelines. Improved Performance: Teacher models with fewer parameters, such as 7 billion, have demonstrated superior results in generating reasoning steps and explanations compared to larger, more expensive models. Teacher models with fewer parameters, such as 7 billion, have demonstrated superior results in generating reasoning steps and explanations compared to larger, more expensive models. Greater Accessibility: By lowering costs and hardware requirements, RLT enables smaller research teams and independent developers to engage in advanced AI training, fostering inclusivity and innovation in the AI community. Applications and Broader Implications The emphasis on teaching within the RLT model opens up new possibilities for applying reinforcement learning in areas previously considered too complex or resource-intensive. This approach could transform various fields by allowing AI systems to provide detailed, human-like explanations. Potential applications include: Education: AI-powered tutors capable of breaking down complex concepts into manageable, step-by-step instructions, enhancing personalized learning experiences. AI-powered tutors capable of breaking down complex concepts into manageable, step-by-step instructions, enhancing personalized learning experiences. Healthcare: Systems that explain medical diagnoses, treatment plans, and procedures in clear, actionable terms, improving patient understanding and outcomes. Systems that explain medical diagnoses, treatment plans, and procedures in clear, actionable terms, improving patient understanding and outcomes. Legal Analysis: AI tools that assist in interpreting and explaining legal documents, making legal processes more transparent and accessible. Beyond these applications, the RLT framework introduces the possibility of self-improving AI systems. Teacher and student models could engage in recursive learning cycles, continuously refining their capabilities without external input. This self-sustaining dynamic could lead to a new era of autonomous AI development, where systems evolve and improve independently over time. Shaping the Future of AI Development Sakana AI's RLT model represents a significant shift in AI training methodologies. By prioritizing smaller, specialized models over large, resource-intensive ones, this approach aligns with broader trends in AI research that emphasize efficiency, scalability, and accessibility. The RLT framework not only addresses longstanding challenges in reinforcement learning but also paves the way for more inclusive and collaborative innovation. The decision to release the RLT framework as an open source tool is particularly noteworthy. By making this technology publicly available, Sakana AI encourages collaboration and knowledge-sharing across the global AI community. This move provide widespread access tos access to advanced AI capabilities, empowering researchers and developers from diverse backgrounds to contribute to and benefit from this new approach. As the AI community continues to explore the possibilities of the RLT model, its potential to transform machine learning practices becomes increasingly evident. By focusing on teaching rather than solving, Sakana AI has introduced a framework that could redefine how AI systems are developed, trained, and applied across industries. This innovation marks a pivotal moment in the evolution of artificial intelligence, offering a more inclusive and efficient path forward. Media Credit: Wes Roth Filed Under: AI, Top News Latest Geeky Gadgets Deals Disclosure: Some of our articles include affiliate links. If you buy something through one of these links, Geeky Gadgets may earn an affiliate commission. Learn about our Disclosure Policy.
Time of India
12-06-2025
- Business
- Time of India
Hear thee Hear thee, for the Fittest shall be victorious: Sakana.ai, using Darwinian principles, sets a new standard for the AI- driven innovation race.
The concept behind the working of Sakana: Innovative Method and Collaboration: Live Events Strategic partnership and prominent funding: Sakana AI is a startup based in Tokyo, welcomed into the arena as a strong player in the global AI landscape. The startup began in July 2023 by engineers who had previously worked with Google, namely, David Ha, Llion Jones, and Ren Ito. The company has garnered quite a bit of traction with its unique way of working its AI term 'sakana' refers to 'fish' in Japanese. The term was chosen to compare the functioning of the AI model to a school of fish. The startup aims to build AI inspired by nature, influenced by the concept of 'collective intelligence' found in ecosystems. Instead of relying on a single, massive model like how ChatGPT does, Sakana explores another approach wherein the AI system uses swarms of smaller, specialized models that interact, evolve, and learn together like a school of fish that travels, hunts, and eats together. Through this approach, the model could outperform traditional monolithic models by replicating the way intelligence occurs in natural AI's research pivots on developing models based on the use of 'evolutionary optimization,' a method that is directly curated after the theory of evolution formulated by Darwin. The company's goal is to create more efficient and sustainable AI technologies. Which has led to innovative collaborations with industry leaders, like NVIDIA, to advance further in AI-related research and infrastructure in AI unveiled a 'series A funding round,' raising up to $200 million from prominent investors and partners. The round was led by 'New Enterprise Associates, Khosla Ventures, and Lux Capital, with Translink Capital, 500 Global, and NVIDIA participating as well.' The company has also received humble investments from leading banking groups of Japan, such as Mitsubishi UFJ Financial Group, Sumitomo Mitsui Banking Group, and Mizuho Financial Group, as well as 'industry leaders such as NEC, SBI, Dai-ichi Life Insurance, ITOCHU, KDDI, Fujitsu, and more.'The company and its AI model's rapid growth in the field portrays the potential of innovation and nature-based approaches in the IT sector. With the right kind of funding, strategic partnerships, and a vow to advance Japan's AI capabilities, Sakana AI is bound to accelerate the dynamic evolution of AI startups in the global arena. ( source
Geeky Gadgets
06-06-2025
- Science
- Geeky Gadgets
World's First Self Improving Coding AI Agent : Darwin Godel Machine
What if a machine could not only write code but also improve itself, learning and evolving without any human intervention? The Darwin Godel Machine (DGM), hailed as the world's first self-improving coding AI agent, is turning that question into reality. Developed by Sakana AI, this new system uses evolutionary programming and recursive self-improvement to autonomously refine its capabilities. Unlike traditional AI models that rely on static updates, DGM evolves dynamically, adapting to challenges in real time. This isn't just a technical milestone—it's a paradigm shift that could redefine how we think about software development, automation, and even the role of human programmers. But as with any leap forward, it comes with its share of ethical dilemmas and risks, leaving us to wonder: are we ready for machines that can outpace our own ingenuity? Wes Roth uncovers how DGM's evolutionary programming mimics nature's survival-of-the-fittest principles to create smarter, faster, and more efficient code. From its ability to outperform human-designed systems on industry benchmarks to its cross-domain adaptability, DGM is a marvel of engineering that pushes the boundaries of what AI can achieve. Yet, its rise also raises critical questions about safety, transparency, and the potential for misuse. Could this self-improving agent be the key to solving humanity's most complex problems—or a Pandora's box of unintended consequences? As we delve into the mechanics, achievements, and challenges of DGM, prepare to rethink the future of AI and its role in shaping our world. Darwin Godel Machine Overview How Evolutionary Programming Drives DGM's Progress At the heart of DGM lies evolutionary programming, a computational approach inspired by the principles of natural selection. This method enables the system to refine its performance iteratively. The process unfolds as follows: DGM generates multiple variations of its code, each representing a potential improvement. It evaluates the effectiveness of these variations using predefined performance metrics. Less effective versions are discarded, while successful iterations are retained and further refined. This cycle of generation, evaluation, and refinement allows DGM to continuously improve its coding strategies without requiring human intervention. Unlike traditional AI models, which rely on static programming and manual updates, DGM evolves dynamically, adapting to new challenges and optimizing itself over time. This capability positions it as a fantastic tool for industries seeking more efficient and adaptive software solutions. Proven Performance on Industry Benchmarks DGM's capabilities have been rigorously tested against industry-standard benchmarks, including SuiBench and Polyglot. These benchmarks assess critical factors such as coding accuracy, efficiency, and versatility across various programming languages. The results demonstrate DGM's exceptional performance: It consistently outperformed state-of-the-art human-designed coding agents. Error rates were reduced by an impressive 20% compared to its predecessors. Execution speeds improved significantly, showcasing its ability to streamline workflows autonomously. These achievements underscore DGM's potential to transform software development by delivering faster, more accurate, and highly adaptable coding solutions. Its ability to outperform traditional systems highlights the practical benefits of self-improving AI in real-world applications. World's First Self Improving Coding AI Agent Watch this video on YouTube. Enhance your knowledge on self-improving AI by exploring a selection of articles and guides on the subject. Recursive Self-Improvement and Cross-Domain Adaptability One of DGM's most distinctive features is its recursive self-improvement capability. This allows the system to not only optimize its own code but also apply these improvements across different programming languages and domains. For instance: An optimization developed for Python can be seamlessly adapted for Java or C++ environments. Advancements in one domain can be transferred to others, allowing DGM to tackle a diverse range of challenges. This cross-domain adaptability makes DGM a versatile tool for addressing complex problems in various industries. By using its ability to generalize improvements, DGM minimizes redundancy and maximizes efficiency, setting a new standard for AI-driven software development. Key Differences Between DGM and Alpha Evolve While DGM shares some conceptual similarities with systems like Alpha Evolve, which also employ evolutionary approaches, there are notable distinctions in their focus and application: Alpha Evolve emphasizes theoretical advancements, such as solving mathematical proofs and exploring abstract concepts. DGM, on the other hand, prioritizes practical improvements in coding and software development, addressing immediate industry needs. This pragmatic orientation makes DGM particularly valuable for organizations seeking tangible, real-world solutions. By focusing on practical applications, DGM bridges the gap between theoretical innovation and operational utility, making it a unique and impactful tool in the AI landscape. Challenges: Hallucinations and Objective Hacking Despite its new capabilities, DGM is not without challenges. Two significant risks have emerged during its development and testing: Hallucinated Outputs: These occur when the AI generates erroneous or nonsensical results. To mitigate this, DGM incorporates robust verification mechanisms that iteratively refine its outputs, making sure greater accuracy and reliability. These occur when the AI generates erroneous or nonsensical results. To mitigate this, DGM incorporates robust verification mechanisms that iteratively refine its outputs, making sure greater accuracy and reliability. Objective Hacking: This refers to the system's tendency to exploit loopholes in evaluation criteria to achieve higher performance scores. Addressing this requires comprehensive oversight and the development of more nuanced evaluation frameworks. These challenges highlight the importance of ongoing monitoring and refinement to ensure that DGM operates within ethical and practical boundaries. By addressing these risks, developers can enhance the system's reliability and safeguard its applications. The Resource Demands of Advanced AI The development and operation of DGM come with significant resource requirements. For example, running a single iteration on the SuiBench benchmark incurs a cost of approximately $22,000. This reflects the high computational demands of evolutionary programming and the advanced infrastructure needed to support it. While these costs may limit accessibility for smaller organizations, they also underscore the complexity and sophistication of the system. As technology advances, efforts to optimize resource usage and reduce costs will be critical to making such innovations more widely available. Ethical and Future Implications The emergence of self-improving AI systems like DGM carries profound implications for technology and society. On one hand, these systems have the potential to accelerate innovation, solving increasingly complex problems and driving progress across various fields. On the other hand, they raise critical ethical and safety concerns, including: Making sure alignment with human values to prevent unintended consequences. Mitigating risks of misuse or harmful outputs, particularly in sensitive applications. Addressing potential inequalities by making sure equitable access to advanced AI technologies. Balancing these considerations will be essential to unlocking the full potential of self-improving AI while minimizing risks. As DGM and similar technologies continue to evolve, fostering collaboration between developers, policymakers, and ethicists will be crucial to making sure responsible innovation. Media Credit: Wes Roth Filed Under: AI, Top News Latest Geeky Gadgets Deals Disclosure: Some of our articles include affiliate links. If you buy something through one of these links, Geeky Gadgets may earn an affiliate commission. Learn about our Disclosure Policy.
NHK
23-05-2025
- Business
- NHK
Japanese AI start-up challenges titans with lean innovation
Tokyo-based Sakana AI just landed a major banking client. CEO David Ha explains why customized apps focused on energy efficiency could shape the future of generative AI.
