@siliconangle.com
//
Google LLC has announced the development of a new AI model designed to improve tropical cyclone forecasting. The model, created through collaboration between Google Research and DeepMind, aims to predict both the path and intensity of these storms days in advance, offering a significant advancement over traditional forecasting methods. This AI-driven approach is accessible through a newly launched website called Weather Lab, providing researchers and experts with access to cutting-edge predictions. According to Google, the algorithm was trained using two datasets. The first described the path, intensity and other key properties of nearly 5,000 cyclones from the past 45 years. The other dataset included information about past weather conditions that was distilled from millions of observations.
The traditional reliance on physics-based weather prediction models often faces limitations, struggling to accurately predict both a cyclone's track and its intensity simultaneously. Google claims its AI model overcomes this hurdle, achieving state-of-the-art accuracy in forecasting both aspects of cyclone behavior. Furthermore, the model can predict other vital details, including a cyclone's formation, size, and shape. In internal tests, Google successfully used the algorithm to predict the paths of four recent cyclones, generating accurate forecasts nearly a week ahead of time for two of the storms, with the capability to predict storms up to 15 days in advance by generating 50 possible scenarios. The Weather Lab platform allows users to explore and compare predictions from various AI and physics-based models, potentially enhancing the ability of weather agencies and emergency services to anticipate cyclone paths and intensities. DeepMind has also announced a partnership with the U.S. National Hurricane Center, which will incorporate the AI predictions into its operational forecasting workflow for the first time. The company claims this is a major breakthrough in hurricane forecasting, introducing an artificial intelligence system that can predict both the path and intensity of tropical cyclones with unprecedented accuracy which has eluded traditional weather models for decades. Recommended read:
References :
@quantumcomputingreport.com
//
References:
thequantuminsider.com
,
,
The quantum computing industry is experiencing a surge in activity, marked by significant acquisitions and technological advancements. IonQ has announced its intent to acquire UK-based Oxford Ionics for $1.075 billion in stock and cash, uniting two leaders in trapped-ion quantum computing. This deal aims to accelerate the development of scalable and reliable quantum systems, targeting 256 high-fidelity qubits by 2026 and over 10,000 physical qubits by 2027. The acquisition combines IonQ's quantum computing stack with Oxford Ionics' semiconductor-compatible ion-trap technology, strengthening IonQ's technical capabilities and expanding its European presence. CEO of IonQ, Niccolo de Masi, highlighted the strategic importance of this acquisition, uniting talent from across the world to become the world’s best quantum computing, quantum communication and quantum networking ecosystem.
Recent advancements also include the activation of Europe’s first room-temperature quantum accelerator by Fraunhofer IAF, featuring Quantum Brilliance’s diamond-based QB-QDK2.0 system. This system utilizes nitrogen-vacancy (NV) centers and operates without cryogenic requirements, seamlessly integrating into existing high-performance computing environments. It's co-located with classical processors and NVIDIA GPUs to support hybrid quantum-classical workloads. Moreover, IBM has announced plans to build the world’s first large-scale, error-corrected quantum computer named Starling, aiming for completion by 2028 and cloud availability by 2029. IBM claims it has cracked the code for quantum error correction, moving from science to engineering. Further bolstering the industry's growth, collaborative projects are demonstrating the potential of quantum computing in various applications. IonQ, in partnership with AstraZeneca, AWS, and NVIDIA, has showcased a quantum-accelerated drug discovery workflow that drastically reduces simulation time for key pharmaceutical reactions. Their hybrid system, integrating IonQ’s Forte quantum processor with NVIDIA CUDA-Q and AWS infrastructure, achieved over a 20-fold improvement in time-to-solution for the Suzuki-Miyaura reaction. Additionally, the Karnataka State Cabinet has approved the second phase of the Quantum Research Park at the Indian Institute of Science (IISc) in Bengaluru, allocating ₹48 crore ($5.595 million USD) to expand the state’s quantum technology infrastructure and foster collaboration between academia, startups, and industry. Recommended read:
References :
Sophia Chen@technologyreview.com
//
IBM has announced ambitious plans to construct a large-scale, error-corrected quantum computer, aiming for completion by 2028. This initiative, known as IBM Quantum Starling, represents a significant step forward in quantum computing technology. The project involves a modular architecture, with components being developed at a new IBM Quantum Data Center in Poughkeepsie, New York. IBM hopes to make the computer available to users via the cloud by 2029.
The company's approach to fault tolerance involves a novel architecture using quantum low-density parity check (qLDPC) codes. This method is projected to drastically reduce the number of physical qubits required for error correction, potentially cutting overhead by around 90% compared to other leading codes. IBM says it's cracked the code to quantum error correction and this will significantly enhance the computational capability of the new machine compared to existing quantum computers. IBM also released two technical papers outlining how qLDPC codes can improve instruction processing and operational efficiency, and describes how error correction and decoding can be handled in real-time using classical computing resources. IBM anticipates that Starling will be capable of executing 100 million quantum operations using 200 logical qubits. This lays the foundation for a follow-up system, IBM Quantum Blue Jay, which will operate with 2,000 logical qubits and run 1 billion operations. According to IBM, storing the computational state of Starling would require memory exceeding that of a quindecillion (10⁴⁸) of today’s most powerful supercomputers. This project aims to solve real-world challenges and unlock immense possibilities for business in fields such as drug development, materials science, chemistry, and optimisation. Recommended read:
References :
Carl Franzen@AI News | VentureBeat
//
Mistral AI has launched its first reasoning model, Magistral, signaling a commitment to open-source AI development. The Magistral family features two models: Magistral Small, a 24-billion parameter model available with open weights under the Apache 2.0 license, and Magistral Medium, a proprietary model accessible through an API. This dual release strategy aims to cater to both enterprise clients seeking advanced reasoning capabilities and the broader AI community interested in open-source innovation.
Mistral's decision to release Magistral Small under the permissive Apache 2.0 license marks a significant return to its open-source roots. The license allows for the free use, modification, and distribution of the model's source code, even for commercial purposes. This empowers startups and established companies to build and deploy their own applications on top of Mistral’s latest reasoning architecture, without the burdens of licensing fees or vendor lock-in. The release serves as a powerful counter-narrative, reaffirming Mistral’s dedication to arming the open community with cutting-edge tools. Magistral Medium demonstrates competitive performance in the reasoning arena, according to internal benchmarks released by Mistral. The model was tested against its predecessor, Mistral-Medium 3, and models from Deepseek. Furthermore, Mistral's Agents API's Handoffs feature facilitates smart, multi-agent workflows, allowing different agents to collaborate on complex tasks. This enables modular and efficient problem-solving, as demonstrated in systems where agents collaborate to answer inflation-related questions. Recommended read:
References :
@www.marktechpost.com
//
A new framework called AlphaOne, developed by researchers at the University of Illinois Urbana-Champaign and the University of California, Berkeley, offers AI developers a novel method to modulate the reasoning processes of large language models (LLMs). This test-time scaling technique improves model accuracy and efficiency without requiring costly retraining. AlphaOne essentially provides a new "dial" to control LLM 'thinking,' allowing developers to boost performance on complex tasks in a more controlled and cost-effective manner compared to existing approaches. The framework dynamically manages slow-to-fast reasoning transitions, optimizing accuracy on real-world datasets like AMC23 and LiveCodeBench.
One persistent issue with large reasoning models is their inability to self-regulate shifts between fast and slow thinking, leading to either premature conclusions or excessive processing. AlphaOne addresses this by providing a universal method for modulating the reasoning process of advanced LLMs. Previous solutions, such as parallel scaling (running a model multiple times) or sequential scaling (modulating thinking during a single run), often lack synchronization between the duration of reasoning and the scheduling of slow-to-fast thinking transitions. AlphaOne aims to overcome these limitations by effectively adapting reasoning processes. In addition to AlphaOne, Amazon Nova provides a solution for data consistency in generative AI through Text-to-SQL. Businesses rely on precise, real-time insights to make critical decisions, and Text-to-SQL bridges the gap by generating precise, schema-specific queries that empower faster decision-making and foster a data-driven culture. Unlike Retrieval Augmented Generation (RAG) which is better suited for extracting insights from unstructured data and Generative Business Intelligence, Text-to-SQL excels in querying structured organizational data directly from relational schemas and provides deterministic, reproducible results for specific, schema-dependent queries. Recommended read:
References :
Emilia David@AI News | VentureBeat
//
OpenAI has recently launched its newest reasoning model, o3-pro, making it available to ChatGPT Pro and Team subscribers, as well as through OpenAI’s API. Enterprise and Edu subscribers will gain access the following week. The company touts o3-pro as a significant upgrade, emphasizing its enhanced capabilities in mathematics, science, and coding, and its improved ability to utilize external tools.
OpenAI has also slashed the price of o3 by 80% and o3-pro by 87%, positioning the model as a more accessible option for developers seeking advanced reasoning capabilities. This price adjustment comes at a time when AI providers are competing more aggressively on both performance and affordability. Experts note that evaluations consistently prefer o3-pro over the standard o3 model across all categories, especially in science, programming, and business tasks. O3-pro utilizes the same underlying architecture as o3, but it’s tuned to be more reliable, especially on complex tasks, with better long-range reasoning. The model supports tools like web browsing, code execution, vision analysis, and memory. While the increased complexity can lead to slower response times, OpenAI suggests that the tradeoff is worthwhile for the most challenging questions "where reliability matters more than speed, and waiting a few minutes is worth the tradeoff.” Recommended read:
References :
@machinelearning.apple.com
//
Apple researchers have released a new study questioning the capabilities of Large Reasoning Models (LRMs), casting doubt on the industry's pursuit of Artificial General Intelligence (AGI). The research paper, titled "The Illusion of Thinking," reveals that these models, including those from OpenAI, Google DeepMind, Anthropic, and DeepSeek, experience a 'complete accuracy collapse' when faced with complex problems. Unlike existing evaluations primarily focused on mathematical and coding benchmarks, this study evaluates the reasoning traces of these models, offering insights into how LRMs "think".
Researchers tested various models, including OpenAI's o3-mini, DeepSeek-R1, and Claude 3.7 Sonnet, using puzzles like the Tower of Hanoi, Checker Jumping, River Crossing, and Blocks World. These environments allowed for the manipulation of complexity while maintaining consistent logical structures. The team discovered that standard language models surprisingly outperformed LRMs in low-complexity scenarios, while LRMs only demonstrated advantages in medium-complexity tasks. However, all models experienced a performance collapse when faced with highly complex tasks. The study suggests that the so-called reasoning of LRMs may be more akin to sophisticated pattern matching, which is fragile and prone to failure when challenged with significant complexity. Apple's research team identified three distinct performance regimes: low-complexity tasks where standard models outperform LRMs, medium-complexity tasks where LRMs show advantages, and high-complexity tasks where all models collapse. Apple has begun integrating powerful generative AI into its own apps and experiences. The new Foundation Models framework gives app developers access to the on-device foundation language model. Recommended read:
References :
@medium.com
//
Medium is currently hosting a series of articles that delve into the core concepts and practical applications of cryptography. These articles aim to demystify complex topics such as symmetric key cryptography, also known as secret key or private key cryptography, where a single shared key is used for both encryption and decryption. This method is highlighted for its speed and efficiency, making it suitable for bulk data encryption, though it primarily provides confidentiality and requires secure key distribution. The resources available are designed to cater to individuals with varying levels of expertise, offering accessible guides to enhance their understanding of secure communication and cryptographic systems.
The published materials offer detailed explorations of cryptographic techniques, including AES-256 encryption and decryption. AES-256, which stands for Advanced Encryption Standard with a 256-bit key size, is a symmetric encryption algorithm renowned for its high level of security. Articles break down the internal mechanics of AES-256, explaining the rounds of transformation and key expansion involved in the encryption process. These explanations are presented in both technical terms for those with a deeper understanding and in layman's terms to make the concepts accessible to a broader audience. In addition to theoretical explanations, the Medium articles also showcase the practical applications of cryptography. One example provided is the combination of OSINT (Open Source Intelligence), web, crypto, and forensics techniques in CTF (Capture The Flag) challenges. These challenges offer hands-on experience in applying cryptographic principles to real-world scenarios, such as identifying the final resting place of historical figures through OSINT techniques. The series underscores the importance of mastering cryptography in the evolving landscape of cybersecurity, equipping readers with the knowledge to secure digital communications and protect sensitive information. Recommended read:
References :
@www.quantamagazine.org
//
Fermilab has announced the final results from its Muon g-2 experiment, aiming to resolve a long-standing anomaly regarding the magnetic moment of muons. This experiment delves into the quantum realm, exploring how short-lived particles popping in and out of existence influence the magnetic properties of muons. The initial results from this experiment suggested that the Standard Model of physics might be incomplete, hinting at the presence of undiscovered particles or forces.
The experiment's findings continue to show a discrepancy between experimental measurements and the predictions of the Standard Model. However, the statistical significance of this discrepancy has decreased due to improvements in theoretical calculations. This implies that while the Standard Model may not fully account for the behavior of muons, the evidence for new physics is not as strong as previously thought. The result is at 4.2σ (standard deviations) away from what's calculated using the Standard Model, which is a bit short of the 5 sigma normally used to declare a discovery. There's about a 1 in 40,000 chance that this is a fluke. Despite the reduced statistical significance, the results remain intriguing and motivate further research. The possibility of undiscovered particles influencing muons still exists, pushing physicists to explore new theoretical models and conduct additional experiments. Fermilab shared first results from their "g-2" experiment showing the Standard Model of physics is even more incomplete than we thought. If the universe includes particles we don't yet know about, these too will show up as fluctuations around particles, influencing the properties we can measure. Recommended read:
References :
@www.quantamagazine.org
//
Recent breakthroughs have significantly advanced the "Core of Fermat's Last Theorem," a concept deeply rooted in number theory. Four mathematicians have extended the key insight behind Fermat's Last Theorem, which states there are no three positive integers that, when raised to a power greater than two, can be added together to equal another number raised to the same power. Their work involves applying this concept to other mathematical objects, notably elliptic curves. This extension represents a major step towards building a "grand unified theory" of mathematics, a long-sought goal in the field.
This achievement builds upon the groundwork laid by Andrew Wiles's famous 1994 proof of Fermat's Last Theorem. Wiles, with assistance from Richard Taylor, demonstrated that elliptic curves and modular forms, seemingly distinct mathematical entities, are interconnected. This discovery revealed a surprising "modularity," where these realms mirror each other in a distorted way. Mathematicians can now leverage this connection, translating problems about elliptic curves into the language of modular forms, solving them, and then applying the results back to the original problem. This new research goes beyond elliptic curves, extending the modularity connection to more complicated mathematical objects. This breakthrough defies previous expectations that such extensions would be impossible. The Langlands program, a set of conjectures aiming to develop a grand unified theory of mathematics, hinges on such correspondences. The team's success provides strong support for the Langlands program and opens new avenues for solving previously intractable problems in various areas of mathematics, solidifying the power and reach of the "Core of Fermat's Last Theorem." Recommended read:
References :
@www.linkedin.com
//
Nvidia's Blackwell GPUs have achieved top rankings in the latest MLPerf Training v5.0 benchmarks, demonstrating breakthrough performance across various AI workloads. The NVIDIA AI platform delivered the highest performance at scale on every benchmark, including the most challenging large language model (LLM) test, Llama 3.1 405B pretraining. Nvidia was the only vendor to submit results on all MLPerf Training v5.0 benchmarks, highlighting the versatility of the NVIDIA platform across a wide array of AI workloads, including LLMs, recommendation systems, multimodal LLMs, object detection, and graph neural networks.
The at-scale submissions used two AI supercomputers powered by the NVIDIA Blackwell platform: Tyche, built using NVIDIA GB200 NVL72 rack-scale systems, and Nyx, based on NVIDIA DGX B200 systems. Nvidia collaborated with CoreWeave and IBM to submit GB200 NVL72 results using a total of 2,496 Blackwell GPUs and 1,248 NVIDIA Grace CPUs. The GB200 NVL72 systems achieved 90% scaling efficiency up to 2,496 GPUs, improving time-to-convergence by up to 2.6x compared to Hopper-generation H100. The new MLPerf Training v5.0 benchmark suite introduces a pretraining benchmark based on the Llama 3.1 405B generative AI system, the largest model to be introduced in the training benchmark suite. On this benchmark, Blackwell delivered 2.2x greater performance compared with the previous-generation architecture at the same scale. Furthermore, on the Llama 2 70B LoRA fine-tuning benchmark, NVIDIA DGX B200 systems, powered by eight Blackwell GPUs, delivered 2.5x more performance compared with a submission using the same number of GPUs in the prior round. These performance gains highlight advancements in the Blackwell architecture and software stack, including high-density liquid-cooled racks, fifth-generation NVLink and NVLink Switch interconnect technologies, and NVIDIA Quantum-2 InfiniBand networking. Recommended read:
References :
@medium.com
//
Google Quantum AI has published a study that dramatically lowers the estimated quantum resources needed to break RSA-2048, one of the most widely used encryption standards. The study, authored by Craig Gidney, indicates that RSA cracking may be possible with fewer qubits than previously estimated, potentially impacting digital security protocols used in secure web browsing, email encryption, VPNs, and blockchain systems. This breakthrough could significantly accelerate the timeline for "Q-Day," the point at which quantum computers can break modern encryption.
Previous estimates, including Gidney's 2019 study, suggested that cracking RSA-2048 would require around 20 million qubits and 8 hours of computation. However, the new analysis reveals it could be done in under a week using fewer than 1 million noisy qubits. This reduction in hardware requirements is attributed to several technical innovations, including approximate residue arithmetic, magic state cultivation, optimized period finding with Ekerå-Håstad algorithms, and yoked surface codes & sparse lookups. These improvements minimize the overhead in fault-tolerant quantum circuits, enabling better scaling. Google's researchers have discovered that, thanks to new error correction tricks and smarter algorithms, the encryption could be broken with under 1 million qubits and in less than a week, given favorable assumptions like a 0.1% gate error rate and a 1-microsecond gate time. This significantly faster encryption breaking capability, potentially 20x faster than previously anticipated, raises concerns about the security of Bitcoin wallets and other financial systems that rely on RSA encryption. The findings could potentially make Bitcoin wallets and financial systems vulnerable much sooner than expected. Recommended read:
References :
@quantumcomputingreport.com
//
The rapid advancement of quantum computing poses a significant threat to current encryption methods, particularly RSA, which secures much of today's internet communication. Google's recent breakthroughs have redefined the landscape of cryptographic security, with researchers like Craig Gidney significantly lowering the estimated quantum resources needed to break RSA-2048. A new study indicates that RSA-2048 could be cracked in under a week using fewer than 1 million noisy qubits, a dramatic reduction from previous estimates of around 20 million qubits and eight hours of computation. This shift accelerates the timeline for "Q-Day," the hypothetical moment when quantum computers can break modern encryption, impacting everything from email to financial transactions.
This vulnerability stems from the ability of quantum computers to utilize Shor's algorithm for factoring large numbers, a task prohibitively difficult for classical computers. Google's innovation involves several technical advancements, including approximate residue arithmetic, magic state cultivation, optimized period finding with Ekerå-Håstad algorithms, and yoked surface codes with sparse lookups. These improvements streamline modular arithmetic, reduce the depth of quantum circuits, and minimize overhead in fault-tolerant quantum circuits, collectively reducing the physical qubit requirement to under 1 million while maintaining a relatively short computation time. In response to this threat, post-quantum cryptography (PQC) is gaining momentum. PQC refers to cryptographic algorithms designed to be secure against both classical and quantum attacks. NIST has already announced the first set of quantum-safe algorithms for standardization, including FrodoKEM, a key encapsulation protocol offering a simple design and strong security guarantees. The urgency of transitioning to quantum-resistant cryptographic systems is underscored by ongoing advances in quantum computing. While the digital world relies on encryption, the evolution to AI and quantum computing is challenging the security. Professionals who understand both cybersecurity and artificial intelligence will be the leaders in adapting to these challenges. Recommended read:
References :
@aasnova.org
//
References:
StartsWithABang
, aasnova.org
JWST is currently being used to study exoplanets, particularly sub-Neptunes, providing valuable data on their atmospheric composition. A recent study utilized JWST spectroscopy to analyze the atmosphere of the sub-Neptune GJ 3090b. This planet orbits a late-type, low-mass star and its radius places it at the outer edge of the radius valley. Sub-Neptunes are the most common type of planet in the Milky Way, however their formation and composition are not well understood, making these studies especially important.
The JWST's observations of GJ 3090b revealed a low-amplitude helium signature, suggesting a metal-enriched atmosphere. The presence of heavy molecules like water, carbon dioxide, and sulfur further contributes to the understanding of the planet's atmospheric properties. These atmospheric observations help clarify how hydrogen and helium may be escaping the planet’s atmosphere, with the presence of metals slowing down mass loss and weakening the helium signature. While JWST is making significant contributions to exoplanet research, it won't find the very first stars. Other telescopes will be needed to make those observations. JWST however contains some of the latest discoveries, including the new cosmic record-holder for the most distant galaxy, MoM-z14. Recommended read:
References :
@medium.com
//
References:
RunPod Blog
, TheSequence
DeepSeek's latest AI model, R1-0528, is making waves in the AI community due to its impressive performance in math and reasoning tasks. This new model, despite having a similar name to its predecessor, boasts a completely different architecture and performance profile, marking a significant leap forward. DeepSeek R1-0528 has demonstrated "unprecedented levels of demand" shooting to the top of the App Store past closed model rivals and overloading their API with unprecedented levels of demand to the point that they actually had to stop accepting payments.
The most notable improvement in DeepSeek R1-0528 is its mathematical reasoning capabilities. On the AIME 2025 test, the model's accuracy increased from 70% to 87.5%, surpassing Gemini 2.5 Pro and putting it in close competition with OpenAI's o3. This improvement is attributed to "enhanced thinking depth," with the model using significantly more tokens per question, engaging in more thorough chains of reasoning. This means the model can check its own work, recognize errors, and course-correct during problem-solving. DeepSeek's success is challenging established closed models and driving competition in the AI landscape. DeepSeek-R1-0528 continues to utilize a Mixture-of-Experts (MoE) architecture, now scaled up to an enormous size. This sparse activation allows for powerful specialized expertise in different coding domains while maintaining efficiency. The context also continues to remain at 128k (with RoPE scaling or other improvements capable of extending it further.) The rise of DeepSeek is underscored by its performance benchmarks, which show it outperforming some of the industry’s leading models, including OpenAI’s ChatGPT. Furthermore, the release of a distilled variant, R1-0528-Qwen3-8B, ensures broad accessibility of this powerful technology. Recommended read:
References :
Dashveenjit Kaur@TechHQ
//
Dell Technologies has secured a contract with the U.S. Department of Energy to construct the next-generation NERSC-10 supercomputer, a project powered by NVIDIA's Vera Rubin architecture. This new system, dubbed "Doudna" after Nobel laureate Jennifer Doudna, a pioneer in CRISPR gene-editing technology, is poised to be a major federal investment in scientific computing infrastructure. Energy Secretary Chris Wright announced the contract during a visit to Lawrence Berkeley National Laboratory, emphasizing that the deployment in 2026 is crucial for maintaining American technological leadership amidst increasing global competition in AI and quantum computing.
The "Doudna" supercomputer, also known as NERSC-10, aims to significantly accelerate scientific research across multiple domains, including fusion energy, astronomy, and life sciences. Designed to serve 11,000 researchers, it represents an integration of artificial intelligence, quantum workflows, and real-time data streaming from experimental facilities. Unlike traditional supercomputers, Doudna’s architecture emphasizes coherent memory access between CPUs and GPUs, facilitating efficient data sharing between heterogeneous processors which is essential for modern AI-accelerated scientific workflows. The Doudna system is expected to deliver a 10x increase in scientific output compared to its predecessor, Perlmutter, while only consuming 2-3x the power, translating to a 3-5x improvement in performance per watt. Nick Wright, advanced technologies group lead and Doudna chief architect at NERSC, stated, "We’re not just building a faster computer, we’re building a system that helps researchers think bigger and discover sooner." NVIDIA's Vera Rubin platform introduces hardware-level optimizations specifically designed for the convergence of simulation, machine learning, and quantum algorithm development, marking a significant advancement in cutting-edge research capabilities. Recommended read:
References :
@www.quantamagazine.org
//
References:
Quanta Magazine
, www.trails.umd.edu
Researchers are making strides in AI reasoning and efficiency, tackling both complex problem-solving and the energy consumption of these systems. One promising area involves reversible computing, where programs can run backward as easily as forward, theoretically saving energy by avoiding data deletion. Michael Frank, a researcher interested in the physical limits of computation, discovered that reversible computing could keep computational progress going as traditional computing slows due to physical limitations. Christof Teuscher at Portland State University emphasized the potential for significant power savings with this approach.
An evolution of the LLM-as-a-Judge paradigm is emerging. Meta AI has introduced the J1 framework which shifts the paradigm of LLMs from passive generators to active, deliberative evaluators through self-evaluation. This approach, detailed in "J1: Incentivizing Thinking in LLM-as-a-Judge via Reinforcement Learning," addresses the growing need for rigorous and scalable evaluation as AI systems become more capable and widely deployed. By reframing judgment as a structured reasoning task trained through reinforcement learning, J1 aims to create models that perform consistent, interpretable, and high-fidelity evaluations. Soheil Feizi, an associate professor at the University of Maryland, has received a $1 million federal grant to advance foundational research in reasoning AI models. This funding, stemming from a Presidential Early Career Award for Scientists and Engineers (PECASE), will support his work in defending large language models (LLMs) against attacks, identifying weaknesses in how these models learn, encouraging transparent, step-by-step logic, and understanding the "reasoning tokens" that drive decision-making. Feizi plans to explore innovative approaches like live activation probing and novel reinforcement-learning designs, aiming to transform theoretical advancements into practical applications and real-world usages. Recommended read:
References :
@www.marktechpost.com
//
DeepSeek has released a major update to its R1 reasoning model, dubbed DeepSeek-R1-0528, marking a significant step forward in open-source AI. The update boasts enhanced performance in complex reasoning, mathematics, and coding, positioning it as a strong competitor to leading commercial models like OpenAI's o3 and Google's Gemini 2.5 Pro. The model's weights, training recipes, and comprehensive documentation are openly available under the MIT license, fostering transparency and community-driven innovation. This release allows researchers, developers, and businesses to access cutting-edge AI capabilities without the constraints of closed ecosystems or expensive subscriptions.
The DeepSeek-R1-0528 update brings several core improvements. The model's parameter count has increased from 671 billion to 685 billion, enabling it to process and store more intricate patterns. Enhanced chain-of-thought layers deepen the model's reasoning capabilities, making it more reliable in handling multi-step logic problems. Post-training optimizations have also been applied to reduce hallucinations and improve output stability. In practical terms, the update introduces JSON outputs, native function calling, and simplified system prompts, all designed to streamline real-world deployment and enhance the developer experience. Specifically, DeepSeek R1-0528 demonstrates a remarkable leap in mathematical reasoning. On the AIME 2025 test, its accuracy improved from 70% to an impressive 87.5%, rivaling OpenAI's o3. This improvement is attributed to "enhanced thinking depth," with the model now utilizing significantly more tokens per question, indicating more thorough and systematic logical analysis. The open-source nature of DeepSeek-R1-0528 empowers users to fine-tune and adapt the model to their specific needs, fostering further innovation and advancements within the AI community. Recommended read:
References :
@www.microsoft.com
//
Microsoft is taking a proactive approach to future cybersecurity threats by integrating post-quantum cryptography (PQC) into its Windows and Linux systems. This move is designed to protect against the potential for quantum computers to break current encryption methods like RSA, which secure online communications, banking transactions, and sensitive data. Quantum computers, leveraging quantum mechanics, can solve complex problems far faster than classical computers, posing a significant threat to existing cryptographic schemes. Microsoft's initiative aims to safeguard data from a "harvest now, decrypt later" scenario, where hackers steal encrypted data today with the intent of decrypting it once quantum technology becomes advanced enough.
Microsoft's PQC implementation includes the addition of two key algorithms: ML-KEM (Module Lattice-Based Key Encapsulation Mechanism) and ML-DSA (Module Lattice-Based Digital Signature Algorithm). ML-KEM, also known as CRYSTALS-Kyber, secures key exchanges and prevents attacks by protecting the start of secure connections. ML-DSA, formerly CRYSTALS-Dilithium, ensures data integrity and authenticity through digital signatures. These algorithms are being introduced in Windows Insider builds (Canary Build 27852+) and Linux via SymCrypt-OpenSSL v1.9.0, allowing developers and organizations to begin testing and preparing for a quantum-secure future. This update to Windows 11 is a critical step in what Microsoft views as a major technological transition. By making quantum-resistant algorithms available through SymCrypt, the core cryptographic code library in Windows, and updating SymCrypt-OpenSSL, Microsoft is enabling the widely used OpenSSL library to leverage SymCrypt for cryptographic operations. The new algorithms, selected by the National Institute of Standards and Technology (NIST), represent a move towards replacing vulnerable cryptosystems like RSA and elliptic curves. This signifies a broader effort to bolster cybersecurity against the emerging threat of quantum computing. Recommended read:
References :
|
Blogs
|