Interview Reviews Frontier Post-Training Recipes
A new interview with Finbarr Timbers examines the methods used to refine frontier AI models after initial training.
AI & ML
Frontier models, research results, agents, and developer tooling.
Gemini 3.5 Live Translate Adds Near-Real-Time Speech Translation
Google DeepMind is bringing fluid, natural voice translation to AI Studio, Google Translate, and Google Meet through Gemini 3.5 Live Translate.
DiffusionGemma Speeds Up Text Generation by 4x
Google DeepMind's DiffusionGemma promises fourfold faster text generation, potentially making large language models more practical for real-time use.
DeepMind Unveils AI Control Roadmap for Securing Autonomous Agents
Google DeepMind proposes a layered security approach that pairs conventional safeguards with real-time monitoring to manage risks from increasingly autonomous AI agents.
OpenAI Simulates Deployments to Predict Model Behavior Before Release
OpenAI's new Deployment Simulation method uses real conversation data to forecast how models will behave in production before they are shipped.
Google Research's Earth AI Bridges Satellite Pixels and Restoration Planning
Google Research describes an AI-driven approach that turns Earth observation imagery into practical plans for nature restoration.
Google Turns Retired Phones Into Low-Carbon Compute
Google Research proposes reusing old Android handsets as a low-carbon computing platform, cutting embodied emissions by extending device lifetimes.
Import AI 461: Alignment Warning, FrontierCode, Synthetic Interns
Jack Clark's latest Import AI issue warns that alignment work is falling behind while spotlighting FrontierCode and synthetic research interns.
Hugging Face Lets Agents Search Hub for Resources
A new launch focuses on making resource discovery on the Hugging Face Hub agent-friendly.
Lambert: AI Governance Enters an Irreversible 'AGI Era'
A new essay argues that AI governance has crossed a point of no return, with policy unprepared for what comes next.
Nextdoor Engineers Use Codex with GPT-5.5 for Hard-to-Reproduce Issues
OpenAI details how Nextdoor engineers use Codex with GPT-5.5 to investigate elusive bugs, build across platforms, and stay product-focused.
US Open-Weight Models From Meta and Reflection Lead AI Roundup
A single Latent Space roundup tracks new US open-weight models, Gemini 4 Argon's wider rollout, and Anthropic's cheaper Haiku.
OpenAI Details PRC-Linked Influence Operations Targeting US AI Debates
A new report from OpenAI reveals how PRC-linked actors are using AI to steer US tech debates, data center narratives, and tariff discussions.
LSEG scales trusted AI with OpenAI across global operations
A new OpenAI case study says LSEG is using the technology to accelerate insights, shorten release cycles, and reach 4,000 employees.
Hugging Face Hub now connects to robot hardware via Strands Agents and LeRobot
A new integration lets developers deploy AI models from the Hugging Face Hub directly onto physical robots using Strands Agents and LeRobot.
DeepMind Announces Robotics Push in Europe
The company is positioning robotics as a major focus of its AI research, with Europe at the centre of the effort.
Beyond LoRA: Can Other Fine-Tuning Methods Beat It?
Hugging Face asks whether any alternative can outperform LoRA, the most popular parameter-efficient fine-tuning technique.
Open Source AI Ban Would Be a Mistake, Op-Ed Argues
Nathan Lambert and Kevin Xu argue against restricting open source AI in an op-ed aimed at a general audience.