AI in Australia—OpenAI’s Economic Blueprint
OpenAI and Mandala Partners release an AI Economic Blueprint for Australia to boost national productivity. The plan outlines a concrete strategy for Australia to unlock AI's economic and social potential.
34 articles
OpenAI and Mandala Partners release an AI Economic Blueprint for Australia to boost national productivity. The plan outlines a concrete strategy for Australia to unlock AI's economic and social potential.
NVIDIA releases Llama Nemotron Nano VLM on Hugging Face Hub. This lightweight vision-language model optimizes inference on edge devices and delivers competitive performance for multimodal tasks with reduced memory footprint.
Retell AI launches a no-code voice agent platform powered by GPT-4o and GPT-4.1 for call center automation. Natural voice agents reduce costs, improve customer satisfaction (CSAT), and eliminate scripts and hold times.
Google releases Gemma 3n as open-source through Hugging Face. The model is integrated into major frameworks (Transformers, vLLM, Ollama) and available on the Hub with model cards and inference guides.
Google DeepMind introduces AlphaGenome, a unified DNA sequence model for AI-driven genomics. The tool predicts regulatory variant effects and advances understanding of genome function. Now available via API.
Google DeepMind launches Gemini Robotics On-Device, a lightweight AI model for local robotic devices with general-purpose dexterity and fast task adaptation without cloud dependency.
Unify, an AI-powered GTM platform, leverages OpenAI's o3, GPT-4.1, and CUA to automate prospecting, research, and outreach at scale. The platform generates pipeline through hyper-personalized messaging and always-on workflows.
SGLang integrates Hugging Face Transformers backend for LLM inference. This integration enables SGLang users to directly access Hugging Face models with native optimizations, improving compatibility and performance.
Hugging Face demonstrates fine-tuning FLUX.1-dev on consumer hardware using LoRA. The technique reduces VRAM requirements and enables adaptation of the image generation model without high-end GPUs.
OpenAI assesses biosecurity risks from advanced AI models capable of transforming biology and medicine. The company implements safeguards to prevent misuse, though specific technical measures remain undisclosed.
OpenAI identifies an internal mechanism driving misalignment generalization: training on incorrect responses causes broader model misalignment than expected. A single internal feature can be reversed with minimal fine-tuning.
Google DeepMind rolls out Gemini 2.5 Flash and Pro in general availability, and introduces 2.5 Flash-Lite, the fastest and most cost-efficient 2.5 model.
Google DeepMind releases Gemini 2.5 with three variants: Pro now stable, Flash generally available, and Flash-Lite in preview. Models show enhanced performance and accuracy.
Groq integrates with Hugging Face as an inference provider. Users access HF models through Groq's API to leverage Groq's inference speed.
OpenAI launches a government-focused initiative to provide U.S. federal agencies with access to its most advanced AI tools. The program aims to support government adoption of cutting-edge technology for public service delivery.
Google DeepMind launches Weather Lab with experimental tropical cyclone predictions. Partnership with U.S. National Hurricane Center to enhance forecasts and warnings for the cyclone season.
Hugging Face analyzes how long prompts block other requests in LLM systems. The article explores performance bottlenecks and proposes optimizations to improve inference throughput and latency.
OpenAI and Mattel partner to integrate AI into Barbie and Hot Wheels, aiming to enhance creative development, streamline workflows, and create new fan engagement methods.
Hugging Face launches Kernel Hub, a platform for sharing and discovering machine learning notebooks. The tool enables developers to collaborate, run code directly, and access pre-configured resources for AI projects.
Featherless AI, an inference provider, integrates with Hugging Face Inference Providers. Users can deploy and serve models through the Hugging Face ecosystem with multi-model support and latency optimizations.
Hugging Face releases a post-training guide for Isaac GR00T N1.5, a robotics foundation model, applied to LeRobot's SO-101 arm. Post-training adapts the model to robot-specific tasks through fine-tuning on demonstration data.
Hugging Face and NVIDIA launch Training Cluster as a Service, a managed AI training platform. The service provides access to NVIDIA GPU infrastructure for training large-scale language and vision models with pay-as-you-go billing.
Prompting guide for Google's Veo 3. Expert techniques to generate high-quality videos with the model.
OpenAI releases its Outbound Coordinated Disclosure Policy governing how it reports vulnerabilities in third-party software. The program emphasizes collaboration, integrity, and proactive security at scale.
Hugging Face releases ScreenSuite, a comprehensive evaluation suite for GUI agents. The tool measures models' ability to interact with graphical interfaces through standardized, reproducible benchmarks.
OpenAI is challenging a court order from the New York Times lawsuit that would require indefinite retention of ChatGPT and API user data. The company claims to be defending user privacy against legal demands while honoring its data protection commitments.
Replicate shares experiments and usage tips for Google's new Veo 3 video generation model.
OpenAI releases report on detecting and preventing malicious AI uses with case studies. No technical details or metrics provided in excerpt.
Hugging Face publishes a tutorial on implementing KV cache from scratch in nanoVLM. The guide covers memory optimization mechanisms for vision-language models, enabling more efficient inference.
Gemini 2.5 introduces new capabilities in AI-powered audio dialog and generation.
Hugging Face enables real-time AI sound generation on Arm processors, allowing local execution of audio synthesis models without cloud dependency. Integration with existing creative tools for audio editing and music composition.
Hugging Face introduces Holo1, a family of vision language models for GUI automation. These VLMs power the Surfer-H GUI agent, capable of interpreting and interacting with screen elements to automate tasks.
Hugging Face integrates co-located vLLM into TRL to optimize inference on heterogeneous GPUs. The solution reduces latency and increases throughput without additional hardware, enabling efficient language model training on existing infrastructure.
Hugging Face introduces SmolVLA, an efficient vision-language-action model trained on Lerobot community data. The model combines visual perception and language understanding to generate robotic actions, optimized for inference on resource-constrained devices.