News from the AI & ML world
DeeperML - #efficiency
|
Michal Langmajer@Fello AI
//
OpenAI has announced the release of GPT-4.5, its latest language model which they are calling their 'last non-chain-of-thought model.' According to OpenAI, GPT-4.5 offers substantial enhancements over its predecessors, particularly in advanced reasoning, problem-solving, and contextual understanding. Sam Altman, CEO of OpenAI, described it as the "first model that feels like talking to a thoughtful person," noting moments of astonishment at the quality of advice received from the AI.
However, the rollout is facing challenges due to GPU shortages. Altman stated they are "out of GPUs," leading to a staggered release, initially limited to ChatGPT Pro subscribers who pay $200 a month. While GPT-4.5 is available to developers across all paid API tiers, OpenAI plans to expand access to Plus and Team tiers next week, with tens of thousands of GPUs expected to arrive to alleviate the supply constraints. Despite not being a reasoning model, OpenAI estimates that GPT-4.5 is 30 times more expensive to run than GPT-4o. Recommended read:
References :
Michal Langmajer@Fello AI
//
OpenAI has launched its latest AI model, GPT-4.5, described as the company's most advanced language model to date. This new model features substantial enhancements over its predecessors, particularly in advanced reasoning, problem-solving, and contextual understanding. GPT-4.5 is designed to offer a more natural and engaging conversational experience, with improvements including superior capabilities in handling complex reasoning tasks, enhanced creativity, and the ability to manage intricate logic problems while maintaining nuanced conversations with improved contextual recall.
However, the launch of GPT-4.5 is facing challenges due to a shortage of GPUs, according to OpenAI CEO Sam Altman. This limitation is restricting access to the priciest tiers of ChatGPT Pro subscribers and developers initially. Altman stated that OpenAI has "run out of GPUs" due to growing demand, leading to a staggered rollout. The company plans to add tens of thousands of GPUs next week and expand access to Plus, Team, Enterprise, and Edu users in the following weeks. Recommended read:
References :
Tris Warkentin@The Official Google Blog
//
Google AI has released Gemma 3, a new family of open-source AI models designed for efficient and on-device AI applications. Gemma 3 models are built with technology similar to Gemini 2.0, intended to run efficiently on a single GPU or TPU. The models are available in various sizes: 1B, 4B, 12B, and 27B parameters, with options for both pre-trained and instruction-tuned variants, allowing users to select the model that best fits their hardware and specific application needs.
Gemma 3 offers practical advantages including efficiency and portability. For example, the 27B version has demonstrated robust performance in evaluations while still being capable of running on a single GPU. The 4B, 12B, and 27B models are capable of processing both text and images, and supports more than 140 languages. The models have a context window of 128,000 tokens, making them well suited for tasks that require processing large amounts of information. Google has built safety protocols into Gemma 3, including a safety checker for images called ShieldGemma 2. Recommended read:
References :
Koray Kavukcuoglu@The Official Google Blog
//
Google has unveiled Gemini 2.5 Pro, touted as its "most intelligent model to date," enhancing AI reasoning and workflow capabilities. This multimodal model, available to Gemini Advanced users and experimentally on Google’s AI Studio, outperforms competitors like OpenAI, Anthropic, and DeepSeek on key benchmarks, particularly in coding, math, and science. Gemini 2.5 Pro boasts an impressive 1 million token context window, soon expanding to 2 million, enabling it to handle larger datasets and understand entire code repositories.
Gemini 2.5 Pro excels in advanced reasoning benchmark tests, achieving a state-of-the-art score on datasets designed to capture human knowledge and reasoning. Its enhanced coding performance allows for the creation of visually compelling web apps and agentic code applications, along with code transformation and editing. Google plans to release pricing for Gemini 2.5 models soon, marking a significant step in their goal of developing more capable and context-aware AI agents. Recommended read:
References :
Eira May@Stack Overflow Blog
//
AI agents are rapidly transforming business operations across various sectors, promising to automate tasks, enhance efficiency, and streamline workflows. Companies are integrating these intelligent systems to modernize customer experiences and unlock enterprise value. To fully leverage the potential of AI agents, businesses need to ensure they have real-time and seamless connections to company databases, internal communication tools, and documents. This integration is crucial for the agents to provide contextually aware and valuable assistance.
Saltbox Mgmt, a Salesforce consulting company, has successfully implemented Agentforce to modernize the buying experience, resulting in improved efficiency and enhanced personalization. Moreover, the integration of AI in real estate technology presents opportunities for strategic transformation, boosting efficiency, value, and decision-making capabilities. However, AI assistants are only as effective as the knowledge base they are connected to, highlighting the importance of comprehensive and up-to-date internal data. Recommended read:
References :
matthewthomas@Microsoft Industry Blogs
//
References:
Source
, The Quantum Insider
Microsoft is emphasizing both AI security and advancements in quantum computing. The company is integrating AI features across its products and services, including Microsoft 365, while also highlighting the critical intersection of AI innovation and security. Microsoft will host Microsoft Secure on April 9th, an online event designed to help professionals discover AI innovations for the security lifecycle. Attendees can learn how to harden their defenses, secure AI investments, and discover AI-first tools and best practices.
Microsoft is also continuing its work in quantum computing, recently defending its topological qubit claims at the American Physical Society (APS) meeting. While Microsoft maintains confidence in its results, skepticism remains within the scientific community regarding the verification methods used, particularly the reliability of the topological gap protocol (TGP) in detecting Majorana quasiparticles. Chetan Nayak, a leading theoretical physicist at Microsoft, presented the company’s findings, acknowledging the skepticism but insisting that the team is confident. Recommended read:
References :
Matthew S.@IEEE Spectrum
//
References:
IEEE Spectrum
, Sebastian Raschka, PhD
Recent research indicates that AI models, particularly large language models (LLMs), can struggle with overthinking and analysis paralysis, impacting their efficiency and success rates. A study has found that reasoning LLMs sometimes overthink problems, which leads to increased computational costs and a reduction in their overall performance. This issue is being addressed through various optimization techniques, including scaling inference-time compute, reinforcement learning, and supervised fine-tuning, to ensure models use only the necessary amount of reasoning for tasks.
The size and training methods of these models play a crucial role in their reasoning abilities. For instance, Alibaba's Qwen team introduced QwQ-32B, a 32-billion-parameter model that outperforms much larger rivals in key problem-solving tasks. QwQ-32B achieves superior performance in math, coding, and scientific reasoning using multi-stage reinforcement learning, despite being significantly smaller than DeepSeek-R1. This advancement highlights the potential of reinforcement learning to unlock reasoning capabilities in smaller models, rivaling the performance of giant models while requiring less computational power. Recommended read:
References :
@www.microsoft.com
//
References:
blogs.microsoft.com
, www.microsoft.com
Microsoft is pushing forward on multiple fronts in the realm of Artificial Intelligence and Cloud Technology, with recent advancements impacting both user experience and business ROI. One significant development focuses on enhancing video quality within Microsoft Teams through the introduction of Super Resolution (SR), now available in public preview and slated for general availability in March. This feature, initially for Snapdragon X-based Copilot+ PCs, leverages AI to restore video resolution during Teams calls, particularly when network bandwidth is limited and lower-resolution videos are transmitted.
This AI-driven approach significantly improves visual clarity compared to traditional upscaling methods, with user assessments indicating an average increase of +0.6 CMOS in quality. Furthermore, Microsoft is making strides in deploying Large Language Models (LLMs) on edge devices like smartphones and laptops through advances in low-bit quantization. These innovations, including techniques like T-MAC, Ladder, and LUT Tensor Core, improve computational efficiency and hardware compatibility, enabling more efficient operation of LLMs on resource-constrained devices. Microsoft also highlights how real-world businesses are transforming using AI, showcasing 50 new customer stories. Recommended read:
References :
|
BenchmarksBlogsResearch Tools |