OpenAI Realtime Models: Single-Model Voice, Lower p95 Latency, and Cheaper Mini Reasoning

OpenAI’s realtime update: single-model voice, lower tail latency, and a cheaper “mini” thinking layer OpenAI’s new realtime models, “gpt-realtime-2.1 and gpt-realtime-2.1-mini.”, let a single model handle live audio in and out, cutting round-trip time compared with chaining speech-to-text (STT) → LLM → TTS. For businesses that need natural, interruptible voice agents, this reduces latency and […]

Tencent Hy3: 295B MoE, 256K‑Token Context for Long‑Document and Agent Workloads

If you need a model that can hold entire contracts, multi-document pipelines, or long agent histories in one session, Tencent’s Hy3 is worth evaluating. Tencent’s Hy research team has published Hy3, a 295-billion-parameter Mixture‑of‑Experts (MoE) language model. They report it can handle a 256K‑token context and activate up to 21 billion parameters per token (top‑8 […]

Selective unlearning in Amazon Nova: rDPO and LoRA adapters, risks and operational controls

Teaching models to forget: Selective unlearning with Amazon Nova Post-training safety alignment can produce “deflection”, models that refuse legitimate business queries. Amazon’s Customizable Content Moderation Settings (CCMS) combined with a preference-based unlearning method called Reverse Direct Preference Optimization (rDPO) promise a way to relax specific refusals for approved customers without changing base model weights (see […]

Multi-turn RL on AWS: Nova Forge + SageMaker HyperPod production pipeline and cost trade-offs

TL;DR, Executive summary What this recipe does: it stitches Amazon Nova Forge, SageMaker HyperPod, and AWS primitives into an event-driven pipeline that trains multi-turn reinforcement learning agents, agents that learn across entire interaction sequences instead of scoring single replies. Who should care: product and ML teams building agents that must orchestrate APIs, query databases, handle […]

MiniMax M2 on Amazon Bedrock: MoE models for agentic assistants and long‑context apps

Run MiniMax models on Amazon Bedrock Bottom line: MiniMax’s M2 family, now available as managed, open-weight models on Amazon Bedrock, pairs high-capacity Mixture-of-Experts (MoE) architectures with Bedrock’s operational controls. This can speed up agentic coding assistants and long-context document workflows, but teams should validate latency, token accounting, caching, and contractual data-use guarantees in a short […]

AI Surveillance: How It Chills Civic Life and What Leaders Must Do

AI surveillance is being supercharged, and it will chill social progress In 2018 NPR reported the case of Lao Duan, a man placed on an administrative blacklist in China whose photograph, name and citizen ID were displayed on a large electronic billboard as an “untrustworthy person.” He was barred from buying high‑speed train tickets, had […]

Kids adopting AI three times faster than adults — UNICEF briefing and leader checklist

UNICEF: Kids adopting AI more than three times faster than adults, with limited methodological detail UNICEF, citing new survey work with IPSOS, reports that 12-17 year‑olds are adopting AI at “more than three times faster” rates than adults and estimates at least 20 million children have used AI. The briefing pairs that headline with concrete […]

OpenScience: A Local, Model‑Agnostic Lab Notebook and Agent Workbench for Secure R&D

A local lab notebook that can call any model: OpenScience arrives Labs and R&D teams have been choosing between three awkward options: stitch together brittle scripts, hand sensitive data to a managed vendor, or spend engineering hours building reproducible pipelines. OpenScience tries to avoid that trade-off. It is an open-source, model-agnostic research workbench that runs […]