Homechevron_rightArchiveschevron_rightTechnology

Archives

Explore Stories by Date

05 Feb 2026

GitHub adds Claude and Codex AI coding agents to improve workflow efficiency

GitHub has integrated Anthropic's Claude and OpenAI's Codex AI coding agents directly into its platform, available across GitHub, GitHub Mobile, and Visual Studio Code. The move allows developers to select different AI agents for specific tasks without losing context, marking a shift toward multi-agent collaboration in software development.

2 Sources

YouTube expands auto-dubbing feature to all creators with AI-powered voice and lip sync upgrades

YouTube has made its AI-powered auto-dubbing feature available to all creators, supporting 27 languages with natural-sounding voices. The platform averaged over 6 million daily viewers watching at least 10 minutes of auto-dubbed content in December 2025, signaling strong adoption as creators reach global audiences without manual translation work.

4 Sources

Moya AI robot walks with 92% human accuracy, has warm skin, and makes eye contact like a person

Shanghai startup DroidUp unveiled Moya, a biomimetic AI robot that maintains human body temperature, walks with 92% accuracy, and expresses emotions through subtle facial movements. Priced at $173,000, the humanoid robot is designed for healthcare and education roles, though its lifelike qualities have sparked both fascination and uncanny valley concerns among observers.

3 Sources

Google Gemini reaches 750 million users as Gemini 3 launch drives explosive growth

Google's AI chatbot Gemini has crossed 750 million monthly active users, adding 100 million in a single quarter following the Gemini 3 launch. While still trailing ChatGPT's 810 million users, the rapid growth signals Google's successful comeback in the AI chatbot competition after previous setbacks.

3 Sources

AI Bots Surge to 1 in 31 Web Visits as Publishers Face Traffic Crisis and Arms Race

AI bots now represent one in every 31 website visits, a dramatic jump from one in 200 at the start of 2025, according to new data from TollBit and Akamai. The surge is driven by retrieval augmented generation tools like ChatGPT rather than training scrapers, with over 13 percent of bots ignoring robots.txt protections. Publishers are seeing click-through rates plummet while human web traffic declines by 5 percent.

5 Sources

Claude Sonnet 5 leak reveals one-million-token context and imminent Anthropic release

Anthropic is preparing to release Claude Sonnet 5, a new mid-tier AI model that could reshape the large language model market. Leaked references found in Google Vertex AI logs point to a February 3, 2026 launch, with the model codenamed Fennec showing a one-million-token context window, 50% lower cost than Opus 4.5, and TPU-based training for faster inference speeds.

3 Sources

04 Feb 2026

Microsoft builds scanner to detect hidden backdoors in AI models using three warning signs

Microsoft's AI Security team developed a lightweight scanner that identifies backdoors in open-weight large language models. The tool detects model poisoning by analyzing three behavioral signals: abnormal attention patterns, memorization of poisoned data, and fuzzy trigger responses. Unlike traditional malware, these sleeper agents remain dormant until activated by specific trigger phrases, making them extremely difficult to detect through standard safety testing.

4 Sources

OpenScholar AI tool surpasses GPT-4o in literature reviews and eliminates citation hallucinations

Academic researchers unveiled OpenScholar, an open-source AI tool that outperformed major LLMs like GPT-4o in scientific literature reviews. The system combines a language model with 45 million open-access articles to deliver accurate citations without hallucinations. Over 30,000 scientists have already tested the free tool since its debut.

2 Sources

Mistral AI Releases Voxtral Models That Transcribe Speech On-Device in Under 200 Milliseconds

Paris-based Mistral AI launched two new speech-to-text models that transcribe audio locally on phones and laptops without cloud transmission. Voxtral Realtime delivers transcription within 200 milliseconds across 13 languages, while Voxtral Mini Transcribe V2 handles batch processing. The 4-billion-parameter models cost just $0.006 per minute via API.

3 Sources

Netease Refutes Suda 51 Claims About Shutting Down AI Department and Halting Generative AI

Netease has firmly denied claims by game developer Suda 51 that it shut down its AI department and banned generative AI use across its studios. The Marvel Rivals publisher confirms it continues investing in AI technology for game development, with upcoming GDC presentations showcasing over $6 million in cost savings from AI integration.

3 Sources

Google Maps tests Gemini chatbot for editing listings as AI integration expands

Google is testing a Gemini-powered upgrade to Google Maps' "Suggest an edit" feature that replaces traditional dropdown menus with a conversational chatbot interface. Users will be able to update listing information using natural language conversations, making it easier to correct store hours, business details, and location information through crowdsourced community edits.

2 Sources

Amazon to test AI tools for film and TV production with industry partners starting March

Amazon MGM Studios is launching a closed beta program in March to test proprietary AI tools designed to streamline film and TV production. Led by Albert Cheng, the AI Studio aims to reduce production costs and improve efficiency while maintaining creative control in human hands. The initiative features 350 AI-generated shots in House of David season two.

6 Sources

Roblox AI 4D creation feature enters open beta, generates functioning models from text prompts

Roblox has launched its 4D creation feature in open beta, enabling users to generate fully functioning in-game models using natural language prompts. The tool builds on the company's Cube 3D model, which has generated over 1.8 million static 3D objects since March. The new feature adds interactivity, allowing creators to design vehicles with working doors and accurate physics mechanics, marking a significant step in Roblox's broader AI world models strategy.

6 Sources

Jensen Huang tells companies to skip ROI demands and let AI innovation run wild

Nvidia CEO Jensen Huang challenged conventional business wisdom at the Cisco AI Summit, urging leaders to abandon immediate ROI expectations for AI projects. Comparing corporate innovation to parenting, he advocated for messy experimentation over rigid control, arguing that demanding proof of financial success before trying new AI tools stifles creativity just like asking a child to justify a hobby with a business plan.

2 Sources

Moltbook AI Social Network Reaches 1.5M Agents, Sparking Debate on Autonomous AI Collaboration

A Reddit-like platform called Moltbook has emerged as the first AI-only social network where AI agents interact without human participation. Within its first week, the platform reportedly reached 1.5 million agent users, generating 110,000 posts and sparking intense debate about autonomous AI collaboration, security risks, and what happens when machines coordinate independently.

8 Sources

Get all updates into your inbox 💌, Stay Up-to-date for Free 🥳

Your Daily Dose of Curated AI News

Don’t drown in AI news. We cut through the noise - filtering, ranking and summarizing the most important AI news, breakthroughs and research daily. Spend less time searching for the latest in AI and get straight to action.

The Outpost

News

About