Translate
Live intelligence

Latest AI updates

186 updates
AllModelsProductsIndustryResearchTutorialOpinion
Products Editor selected

NVIDIA Shows 3 Ways to Speed Up MoE Inference on DGX Spark, Lifting Ling-3.0-flash Throughput 42.5%

NVIDIA's technical blog details three optimizations for mixture-of-experts (MoE) inference on a single DGX Spark desktop system, using Ant Group's Ling-3.0-flash model. The optimizations—Humming MXFP4 MoE kernels, an FP8 LM head, and built-in MTP speculative decoding—boosted GSM8K decode throughput by 42.5% (from 37.81 to 53.89 tokens/s) with no statistically significant accuracy change. The post also highlights the memory-bound nature of MoE decoding and the integration trap when stacking FP8 LM head with MTP, requiring a shared LM-head module.

Why it matters: Provides concrete, actionable performance optimizations for running large MoE models on edge hardware, with real benchmark results and practical integration guidance.

MarkTechPost1d agoScore 85
Mixture-of-ExpertsInference OptimizationEdge AI
Products

Soniox STT v6 Adds Agent Text Input and Echo Filtering for Voice Agents

Soniox has released STT v6, a speech-to-text API update that introduces agent text input, allowing the model to read what the voice agent says while transcribing the caller, and model-level echo filtering to remove leaked agent audio. The update is available as stt-rt-v6 for streaming and stt-async-v6 for files, supporting 60+ languages, with pricing at $0.12/hour streaming and $0.10/hour async. No v6 benchmark numbers are published yet; v5 scored 1.34% semantic WER on Pipecat, and v5 models will be deprecated on December 15, 2026.

MarkTechPost1d agoScore 78
Soniox STT v6voice agent technologyspeech recognition
Products

Apple starts giving Siri AI access to third-party apps

Apple's iOS 27.3 update, rolling out as a Developer Preview, enables Siri AI to perform actions across third-party apps using personal context, natural language prompts, and on-screen awareness. Developers can integrate these features in all apps and create 'app skills' to combine multiple tasks into single actions. Users won't need to mention app names for many requests.

The Verge AI1d agoScore 78
Apple AIvoice assistantsdeveloper tools
Products

OpenAI says GPT-6 Astra designed a hardware device from scratch in under a week

OpenAI's hardware team used GPT-6 Astra to design a complete hardware device for the Codex coding agent, including enclosure, circuit board, and firmware, in under a week. Multiple revision rounds were needed to address manufacturing tolerances. A newer internal model reportedly scores higher on OpenAI's engineering benchmark, but details are undisclosed.

the-decoder1d agoScore 78
AI hardware designOpenAI modelsengineering automation
Products

Netflix’s new AI Discover tab helps you find something to watch based on your mood

Netflix is introducing an experimental AI-powered Discover tab that provides recommendations based on the user's current mood. The first feature under this tab lets users specify their mood to receive tailored suggestions. This marks a shift toward more contextual, personalized content discovery.

techcrunch1d agoScore 75
NetflixAI recommendation systemspersonalization
Products

Apple's watchOS 27.3 beta adds always-on Audio Intelligence with Siri Recap and Live Rewind

Apple's latest watchOS 27.3 developer beta introduces Audio Intelligence features, including Siri Recap for ambient listening summaries and Live Rewind for text transcripts of the last 15 seconds of audio. The beta also includes a redesigned Health app in iOS 27.3 and expanded language support. These features are rolling out to developers starting today.

The Verge AI1d agoScore 78
Apple AI featureswearable AIvoice transcription
Products

Netflix introduces 'AI Discover' tab with themed, AI-powered recommendations

Netflix is adding an 'AI Discover' tab to its TV app, featuring experimental AI-powered features that recommend shows and movies based on themes like 'Something Light' or 'Family Movie Night.' Users can give feedback on recommendations, which the system remembers. The feature is powered by a new intelligence layer that learns from member input in real time.

The Verge AI1d agoScore 75
NetflixAI recommendationsstreaming services
Products

Google lets schools opt out of AI tools, but search is bundled in

Google will default schools to opt out of AI tools in its education products, but administrators must enable all AI features or none, including Google Search. This bundling forces schools to choose between AI access and core search functionality. The change rolls out in the coming months after consultations with educators and researchers.

Ars Technica AI1d agoScore 78
Google educationAI policyschool technology
Products Editor selected

Agentic cheminformatics with Google Antigravity

Google DeepMind introduces agentic cheminformatics capabilities in Google Antigravity, designed to accelerate drug discovery workflows. The system can search compounds and propose novel rigidified macrocycles using specialized Cheminformatics Skills. This integration aims to streamline research tasks in medicinal chemistry.

Why it matters: Provides concrete details on a new AI-driven tool for drug discovery, offering practical value for researchers and industry professionals.

Google DeepMind1d agoScore 85
Google Antigravitycheminformaticsdrug discovery
Products Editor selected

Building ultra-low-latency voice AI applications on AWS

The post introduces Cartesia's voice AI models Sonic (text-to-speech) and Ink (speech-to-text), built on State Space Models (SSMs) for sub-100ms latency. It details deployment options on AWS via SageMaker AI Marketplace or self-managed EC2 GPU instances, and highlights integration with Amazon Connect. The content provides practical guidance for developers seeking real-time voice experiences.

Why it matters: Offers concrete technical details on SSM-based voice AI deployment with AWS, useful for developers building low-latency voice applications.

AWS Machine Learning Blog1d agoScore 85
CartesiaSonicInk
Products Editor selected

Amazon Bedrock Adds India-Based Inference for Kimi K3

Amazon Bedrock now supports in-country inference for Moonshot AI's Kimi K3 model in India, with processing confined to Mumbai and Hyderabad Regions. This enables customers with strict data residency requirements to use the model for workloads like OCR and multilingual document understanding. The post includes a code walkthrough for extracting data from Hindi documents and answering follow-up questions in Hinglish, plus tips on prompt caching and cost optimization.

Why it matters: Provides a concrete, recent AWS service update with practical code examples and a real customer case, valuable for developers needing compliant AI inference in India.

AWS Machine Learning Blog1d agoScore 82
Amazon BedrockKimi K3data residency
Products

Google will let schools disable its AI tools, but not without sacrificing Search

Google is allowing schools to disable its AI tools, but doing so means losing access to Google Search because AI features are now integrated into the search engine. This creates a trade-off for educational institutions that want to limit AI use. The decision reflects the deep integration of AI into core products.

techcrunch1d agoScore 75
Google AI integrationAI in educationproduct policy
Products Editor selected

Same Model, Different Harness: CUGA, Hermes, and Deep Agents — When to Pick CUGA?

The blog post introduces CUGA, a new agent harness from IBM Research, and compares it with existing harnesses like Hermes and Deep Agents. It explains the design philosophy and use cases where CUGA excels, such as complex multi-step tasks requiring structured reasoning. The post provides practical guidance on when to choose CUGA over alternatives.

Why it matters: Provides a concrete comparison of a new AI agent harness, offering actionable selection criteria for practitioners.

Hugging Face Blog2d agoScore 85
AI agentsagent harnessestool selection
Products Editor selected

Anthropic's Message Threads cut time to first token by 35 percent for long conversations

Anthropic has released Message Threads for its Messages API, enabling agents to transmit only changed parts of a conversation instead of resending the entire history. For requests between 500,000 and one million tokens, this reduces time to first token by approximately 35 percent. The feature is designed to improve efficiency in long-running agent interactions.

Why it matters: Offers a concrete, measurable performance improvement for developers using Anthropic's API for extended conversations.

The Decoder2d agoScore 85
AnthropicAPI performancemessage threading
Products Editor selected

OpenAI Introduces Pando: A Git Backend That Cuts Median Push Latency ~80%

OpenAI has built and deployed Pando, a new Git-compatible server backend written in Rust that handles nearly 1 PB of Git data per day for its monorepo. By replacing repository-wide consistency with per-branch causal consistency and storing reusable packfiles, median push latency dropped nearly 80%. Pando was designed to handle the massive increase in parallel commits from Codex agents, with commits on master up 1,378% in Q3 2026 versus Q1 2024. The system has been in production since April 2026 and is not open source, though some client fixes have been contributed upstream.

Why it matters: Provides concrete, self-reported performance numbers and architectural details of a novel Git backend that solves real scaling challenges for AI-driven development workflows.

MarkTechPost2d agoScore 90
OpenAIGit infrastructuremonorepo scaling
Products

Google takes Dreambeans to India in bid to bring AI-generated personal stories to the masses

Google's AI experiment Dreambeans now supports 13 Indian languages, expanding its capability to generate personalized stories for a broader audience in India. The move aims to make AI-driven storytelling accessible to millions of users in their native languages. This expansion underscores Google's focus on multilingual AI applications for emerging markets.

TechCrunch AI2d agoScore 72
GoogleAI storytellingIndia
Products Editor selected

The Letter Company Releases Letterpress-1: A Coding Agent Built for Growth Engineering

The Letter Company has launched the beta of Letterpress-1, a coding agent, harness, and IDE specifically designed for growth engineering. It reads a website and its market, then autonomously builds changes in a sandbox and opens pull requests with live previews. The agent exposes 760 tools via MCP, REST, and CLI, and pricing starts at $100/month. No public benchmarks are available yet, and quality claims are self-reported.

Why it matters: This article details a concrete new product release with specific features, pricing, and comparisons, offering actionable information for teams interested in AI-driven growth engineering.

MarkTechPost2d agoScore 82
AI coding agentsgrowth engineeringAI tools
Products Editor selected

llm 0.37

Simon Willison released version 0.37 of his LLM CLI tool and Python library, adding support for gpt-6.1-sol and gpt-6.1-terra models, a new command to configure OpenAI-compatible endpoints, and switching the default model to the cheaper gpt-6-luna. The release also improves embeddings handling for different content types like text, audio, images, and video. Willison notes that AI-assisted pull requests have been manageable with help from AI triage, and thanks 12 contributors.

Why it matters: Provides concrete updates to a popular open-source AI tool, including new models, pricing improvements, and enhanced embeddings support.

Simon Willison2d agoScore 85
LLMSimon WillisonAI tools