OpenAI o3-mini System Card
OpenAI releases the System Card for o3-mini model, detailing safety evaluations, external red teaming, and Preparedness Framework assessments.
31 articles
OpenAI releases the System Card for o3-mini model, detailing safety evaluations, external red teaming, and Preparedness Framework assessments.
OpenAI releases o3-mini, a compact reasoning model optimized for efficiency. Designed for complex tasks with reduced latency and lower costs, it delivers o3-comparable performance on code and math benchmarks.
Hugging Face releases a tutorial to reproduce Deepseek R1's "aha moment" using reinforcement learning. Practical guide on training models with RL to generate step-by-step reasoning.
Hugging Face launches a newsletter focused on AI tools for art, covering image and video generation models and creative workflows. First issue explores resources and trends in the sector.
OpenAI deploys its latest reasoning models to U.S. National Laboratories to accelerate scientific breakthroughs.
Hugging Face publishes a guide for deploying and fine-tuning DeepSeek models on AWS. The tutorial covers cloud infrastructure, resource optimization, and practical fine-tuning steps on the AWS platform.
Hugging Face reproduces DeepSeek-R1, an open-source reasoning model. Open-R1 provides a fully open alternative to proprietary models, with code, data, and weights publicly available for research and deployment.
Hugging Face introduces Inference Providers section on the Hub, enabling users to access multiple inference providers (Together, Replicate, Hugging Face Inference API) directly from the interface. Simplified integration to deploy and test models without leaving the platform.
Hugging Face publishes a state-of-the-art overview of open-source video generation models integrated in Diffusers. The article covers architectures, performance metrics, and use cases of major available models.
Hugging Face adds Vision Language Models (VLM) support to smolagents. Agents can now process images and text together for multimodal tasks.
Replicate now enables fine-tuning of Tencent's HunyuanVideo open-source model. Users can customize the model for specific styles, motions, and characters.
OpenAI introduces an agent that uses computers by viewing the screen and controlling mouse/keyboard. The agent navigates websites, fills forms, and executes complex tasks without specialized APIs.
OpenAI releases a System Card for Operator outlining multi-layered safety measures: mitigations against prompt engineering and jailbreaks, privacy and security protections, external red teaming, and safety evaluations.
OpenAI introduces Operator, an AI agent capable of performing complex tasks on computers by interpreting visual interfaces and executing autonomous actions. The system uses computer vision to navigate and interact with web and desktop applications.
KVPress is a key-value cache compression technique for LLMs that reduces memory usage without performance degradation on long contexts. Hugging Face presents the method and its integration into models.
Hugging Face releases SmolVLM 256M and 500M, ultra-compact vision-language models. These variants drastically reduce model size while maintaining multimodal capabilities, targeting edge deployments and resource-constrained environments.
Bertelsmann, German media, services, and education conglomerate, integrates OpenAI technology across its global brands to enhance creativity and productivity.
OpenAI explores the trade-off between inference-time compute and adversarial robustness. The approach increases computational resources at inference to improve resistance to adversarial attacks without modifying the base model.
Hugging Face and FriendliAI partner to optimize model deployment on the Hub. The partnership aims to improve inference performance and model accessibility through integration of FriendliAI's technology.
OpenAI announces Stargate, an AGI infrastructure project with strategic partners. The initiative aims to mobilize the industrial ecosystem: data centers, power, land, construction, and equipment.
OpenAI announces the Stargate Project, a strategic partnership to build massive AI infrastructure. The project mobilizes significant investments to develop next-generation computing capacity and models.
Hugging Face now enables organizations to publish blog articles directly on its platform. This feature expands content capabilities beyond models and datasets, facilitating the sharing of research and technical insights.
Replicate launches a playground for generating short AI videos. Simplified interface to create video content without coding.
Hugging Face integrates timm (PyTorch Image Models) directly into the transformers library. Users can load and use any timm model through the standard transformers API without additional dependencies.
Text Generation Inference now supports multiple backends: TensorRT-LLM (NVIDIA) and vLLM. This integration lets users select the optimal inference engine based on their performance and infrastructure requirements.
OpenAI partners with Axios to expand its work with the news industry. Hundreds of newsrooms use OpenAI partnerships and grant programs to adopt AI tools, while ChatGPT users gain access to content from leading, reliable publications.
Sentence Transformers announces optimization enabling 400x faster training of static embedding models. The method drastically reduces computation time without sacrificing vector representation quality.
Adebayo Ogunlesi joins OpenAI's Board of Directors. Former Goldman Sachs executive and influential technology sector figure brings governance and strategy expertise.
Hugging Face examines the emergence of autonomous AI agents and their practical implications. The article explores architectures, use cases, and deployment challenges without detailing specific benchmarks or results.
Hugging Face extends visual document retrieval to multilingual support. A model capable of understanding and indexing documents across multiple languages, improving access to visual information beyond English.
Hugging Face analyzes the correlation between CO₂ emissions and model performance on the Open LLM Leaderboard. The study shows larger models consume more energy without proportional performance gains, questioning the energy efficiency of LLM training.