@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 :
@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 :
Matthias Bastian@THE DECODER
//
OpenAI has announced the integration of GPT-4.1 and GPT-4.1 mini models into ChatGPT, aimed at enhancing coding and web development capabilities. The GPT-4.1 model, designed as a specialized model excelling at coding tasks and instruction following, is now available to ChatGPT Plus, Pro, and Team users. According to OpenAI, GPT-4.1 is faster and a great alternative to OpenAI o3 & o4-mini for everyday coding needs, providing more help to developers creating applications.
OpenAI is also rolling out GPT-4.1 mini, which will be available to all ChatGPT users, including those on the free tier, replacing the previous GPT-4o mini model. This model serves as the fallback option once GPT-4o usage limits are reached. The release notes confirm that GPT 4.1 mini offers various improvements over GPT-4o mini, including instruction-following, coding, and overall intelligence. This initiative is part of OpenAI's effort to make advanced AI tools more accessible and useful for a broader audience, particularly those engaged in programming and web development. Johannes Heidecke, Head of Systems at OpenAI, has emphasized that the new models build upon the safety measures established for GPT-4o, ensuring parity in safety performance. According to Heidecke, no new safety risks have been introduced, as GPT-4.1 doesn’t introduce new modalities or ways of interacting with the AI, and that it doesn’t surpass o3 in intelligence. The rollout marks another step in OpenAI's increasingly rapid model release cadence, significantly expanding access to specialized capabilities in web development and coding. Recommended read:
References :
@www.quantamagazine.org
//
References:
pub.towardsai.net
, Sebastian Raschka, PhD
,
Recent developments in the field of large language models (LLMs) are focusing on enhancing reasoning capabilities through reinforcement learning. This approach aims to improve model accuracy and problem-solving, particularly in challenging tasks. While some of the latest LLMs, such as GPT-4.5 and Llama 4, were not explicitly trained using reinforcement learning for reasoning, the release of OpenAI's o3 model shows that strategically investing in compute and tailored reinforcement learning methods can yield significant improvements.
Competitors like xAI and Anthropic have also been incorporating more reasoning features into their models, such as the "thinking" or "extended thinking" button in xAI Grok and Anthropic Claude. The somewhat muted response to GPT-4.5 and Llama 4, which lack explicit reasoning training, suggests that simply scaling model size and data may be reaching its limits. The field is now exploring ways to make language models work better, including the use of reinforcement learning. One of the ways that researchers are making language models work better is to sidestep the requirement for language as an intermediary step. Language isn't always necessary, and that having to turn ideas into language can slow down the thought process. LLMs process information in mathematical spaces, within deep neural networks, however, they must often leave this latent space for the much more constrained one of individual words. Recent papers suggest that deep neural networks can allow language models to continue thinking in mathematical spaces before producing any text. Recommended read:
References :
Megan Crouse@techrepublic.com
//
References:
hlfshell
, www.techrepublic.com
Researchers from DeepSeek and Tsinghua University have recently made significant advancements in AI reasoning capabilities. By combining Reinforcement Learning with a self-reflection mechanism, they have created AI models that can achieve a deeper understanding of problems and solutions without needing external supervision. This innovative approach is setting new standards for AI development, enabling models to reason, self-correct, and explore alternative solutions more effectively. The advancements showcase that outstanding performance and efficiency don’t require secrecy.
Researchers have implemented the Chain-of-Action-Thought (COAT) approach in these enhanced AI models. This method leverages special tokens such as "continue," "reflect," and "explore" to guide the model through distinct reasoning actions. This allows the AI to navigate complex reasoning tasks in a more structured and efficient manner. The models are trained in a two-stage process. DeepSeek has also released papers expanding on reinforcement learning for LLM alignment. Building off prior work, they introduce Rejective Fine-Tuning (RFT) and Self-Principled Critique Tuning (SPCT). The first method, RFT, has a pre-trained model produce multiple responses and then evaluates and assigns reward scores to each response based on generated principles, helping the model refine its output. The second method, SPCT, uses reinforcement learning to improve the model’s ability to generate critiques and principles without human intervention, creating a feedback loop where the model learns to self-evaluate and improve its reasoning capabilities. Recommended read:
References :
|
Blogs
|