OpenAI describes how ChatGPT defends against prompt injection and social engineering by constraining risky actions and protecting sensitive data in agent workflows.
From model to agent: Equipping the Responses API with a computer environment
OpenAI equipped the Responses API with an agent runtime that uses a shell tool and hosted containers to run secure, scalable agents with files, tools, and state.
OpenAI published a customer story about Wayfair using OpenAI models to improve ecommerce support and product catalog accuracy by automating ticket triage and enhancing product attributes at scale.
OpenAI introduces the IH-Challenge method to train models to better prioritize trusted instructions, enhancing instruction hierarchy, safety steerability, and resistance to prompt injection attacks.
Hugging Face researchers surveyed 16 open-source libraries implementing async RL training patterns, comparing them across seven design axes, and distill lessons for building async trainers.
Hugging Face introduces Storage Buckets, a mutable, S3-like object storage on the Hub for ML intermediate files, backed by Xet for chunk-based deduplication, with CLI, Python, and JavaScript support,…
Hugging Face announced LeRobot v0.5.0, a major update to its open-source robotics library that adds new hardware support (including the Unitree G1 humanoid), six new policies (e.g., Pi0-FAST,…
This blog post explains Ulysses Sequence Parallelism, a technique for training large language models on long sequences by distributing attention computation across multiple GPUs via attention head…
Google Research announces the roll-out of Urban Flash Flood forecasts on Flood Hub, using a new AI-powered methodology that provides up to 24 hours advance notice for flash floods in urban areas.
Google Research introduces Groundsource, a methodology that uses the Gemini LLM to extract structured historical disaster data from news reports, starting with a publicly available dataset of 2.6…
Google Research and Google DeepMind, in partnership with Beth Israel Deaconess Medical Center, conducted a prospective single-center feasibility study of AMIE, a conversational medical AI for…
Together AI launched a unified solution for building real-time voice agents, featuring co-located STT, LLM, and TTS infrastructure, native integrations with Cartesia and Deepgram models, and…
Together AI announced the availability of NVIDIA Nemotron 3 Super, a 120B-parameter hybrid MoE model with a 1M-token context window, on its Dedicated Inference platform.
Together AI announces major enterprise enhancements to Together GPU Clusters, including autoscaling, RBAC, full-stack observability, self-serve node repair, and active health checks.
Cursor announced the addition of over 30 new plugins from various partners, enabling the platform to read, write, and take actions across more of the user's stack.
The blog post reflects on the ten-year impact of AlphaGo, discussing how it has catalyzed scientific discovery and paved a path toward artificial general intelligence (AGI).
Z.ai announces GLM-5-Turbo, a model update focused on improving stability and efficiency in long-chain agent tasks, with enhanced tool integration and instruction decomposition.