OpenAI is launching an initiative called OpenAI for Australia, which is designed to build sovereign AI infrastructure, provide upskilling for over 1.5 million workers, and accelerate AI innovation in…
OpenAI researchers are testing a method called 'confessions' that trains models to admit when they make mistakes or act undesirably, aiming to improve AI honesty, transparency, and trust.
Mirakl is using AI agents and ChatGPT Enterprise to achieve faster documentation and smarter customer support, and is building toward agent-native commerce with Mirakl Nexus.
OpenAI and NORAD collaborate to add three ChatGPT holiday tools to NORAD Tracks Santa, enabling families to create festive elves, toy coloring pages, and custom Christmas stories.
Hugging Face released swift-huggingface, a Swift package that provides a full client for the Hugging Face Hub, including network improvements, a shared cache with Python, flexible authentication, and…
Hugging Face released a new tool called Hugging Face Skills, including the hf-llm-trainer skill that allows coding agents like Claude Code to fine-tune open-source language models on Hugging Face…
Hugging Face releases Transformers v5, focusing on simplicity, training, inference, and production. Key changes include a modular design, removal of Flax/TensorFlow support in favor of PyTorch,…
Together AI introduces AutoJudge, a research method that accelerates LLM inference through task-specific lossy speculative decoding, which automatically identifies which generated tokens affect…
Together AI announces that TorchForge reinforcement learning pipelines now run on its Native Cloud platform, supporting distributed training on Instant Clusters with sandboxed environments via…
Together AI and the PyTorch team at Meta are partnering to build an open-source reinforcement learning framework for training agentic AI systems, integrated into the Together AI platform.
A new reinforcement learning method called SAPO (Soft Adaptive Policy Optimization) is introduced for training large language models, replacing hard clipping with a smooth, temperature-controlled…
Alibaba updated the Qwen3-TTS model family, including Qwen3-TTS-Flash, with richer timbres (over 49), support for 10 languages and 9 dialects, and improved prosody and speech rate naturalness.
Google Research introduces the Titans architecture and MIRAS framework, which enable AI models to update their core memory in real-time, combining the speed of RNNs with the accuracy of transformers.
Google Research introduces the Massive Sound Embedding Benchmark (MSEB), an open-source platform for evaluating machine sound intelligence across eight core capabilities, presented at NeurIPS 2025.
LG AI Research presented MOIRA, a multimodal framework for Alzheimer's disease prediction that handles incomplete data by transforming heterogeneous clinical data into latent representations and…
Jinsik Lee, leader of the EXAONE Lab at LG AI Research, received a Presidential Citation for contributions to the software industry, in part for developing South Korea's first large-scale AI model…
DeepSeek announces DeepSeek-V3.2 and DeepSeek-V3.2-Speciale, reasoning-first models for agents. V3.2 is the successor to V3.2-Exp and is available on App, Web, and API.