Open AI Models Strengthen Cybersecurity, Hugging Face Blog Argues
A new post from Hugging Face contends that transparent, collaborative AI development is essential for defending against evolving cyber threats.
AI & ML
Frontier models, research results, agents, and developer tooling.
Hugging Face Adds Anti-Gaming Measure to Open ASR Leaderboard
The new 'Benchmaxxer Repellant' uses private test data to stop models from overfitting to public benchmarks.
IBM's Granite Embedding R2 Adds 32K Context to Open Multilingual Models
A new Apache 2.0 embedding model from IBM claims top retrieval quality among sub-100M parameter systems.
Hugging Face Adds DeepInfra as Inference Provider
DeepInfra is now available through Hugging Face Inference Providers, expanding hosted model options.
Asynchronous Batching Reduces LLM Inference Idle Time
Hugging Face describes a scheduling tweak that lets requests join or leave a batch at any moment, cutting GPU idle time.
Import AI 454: Automated Alignment, Chinese Model Safety, HiFloat4
Jack Clark's latest newsletter surveys automated alignment research, a Chinese model safety study, and a new floating-point format.
DeepMind’s Decoupled DiLoCo aims to make distributed AI training more resilient
Google DeepMind’s Decoupled DiLoCo is positioned as a new approach to keeping large-scale AI training stable and resilient across distributed systems.
AI-Generated Synthetic Neurons Accelerate Brain Mapping
Google Research reports that synthetic neurons created by AI can make brain-mapping workflows faster.
Hugging Face highlights PR that eases Transformers-to-MLX conversion
A new Hugging Face post explains a pull request that brings Transformers and Apple's MLX framework closer together, giving developers a clearer path for moving models between the two ecosystems.
OpenAI Privacy Filter: A Blueprint for Scalable Web Apps
Hugging Face outlines how developers can use OpenAI's privacy filter to build web apps that scale without compromising user data.
Running Transformers.js AI Models Inside a Chrome Extension
A Hugging Face guide shows how to bring on-device transformer models into browser extensions, enabling private AI features without a backend.
Hugging Face Guide Covers Multimodal Embedding and Reranker Training
A new guide shows how to train and finetune multimodal embedding and reranker models using Sentence Transformers.
Google's ReasoningBank Lets AI Agents Learn From Past Reasoning
A new framework stores and reuses reasoning traces, helping agents improve without full retraining.
DeepMind Partners with Consultancies to Accelerate Enterprise AI
The move is intended to help organizations worldwide adopt frontier AI through established consulting partners.
VAKRA Benchmark Analysis Exposes Agent Reasoning and Tool-Use Gaps
A new study of the VAKRA benchmark reveals where AI agents fail in reasoning and tool use, pointing to concrete areas for improvement.
HoloTab: HCompany Unveils AI Browser Companion on Hugging Face
HCompany has introduced HoloTab, an AI companion designed to work inside the browser, according to a Hugging Face blog post.
Build a Custom LLM Evaluation Harness in Python
A tutorial shows how to assemble a modular, from-scratch evaluation harness for large language models in Python.
Iterators and Generators: Memory-Saving Tools, Not a Religion
A pragmatic guide to using lazy evaluation for memory-efficient data processing without over-applying it.