Traictory/News/Page 23

What's New in AI

Releases, research, trends — everything important in plain language

Page 23 of 29

googlegemmaopen-source

Google Drops Gemma 4 Under Apache 2.0 — And It Runs on a Raspberry Pi

Google DeepMind releases Gemma 4, a family of four open models with agentic skills, 256K context, and multimodal capabilities — all under Apache 2.0 license.

microsoftmaispeech-to-text

Microsoft Just Built Its Own AI Models — With Teams of 10 People

Mustafa Suleyman's superintelligence team ships MAI-Transcribe-1, MAI-Voice-1, and MAI-Image-2 — state-of-the-art models built by tiny teams that beat OpenAI Whisper and undercut every hyperscaler on price.

ai-agentsagentmailinfrastructure

Your AI Agent Now Has Its Own Email and Phone Number

A growing stack of startups lets AI agents own email addresses, phone numbers, wallets, and browsers — raising questions about agent personhood and autonomy.

qwenmultimodallocal-llm

Find Any Moment in Hours of Video — No Transcription Required

SentrySearch uses Qwen3-VL embeddings to search video footage by natural language queries, running entirely offline on consumer hardware.

ai-safetyopenclawagents

Agents of Chaos: When AI Agents Get OS-Level Access, Everything Breaks

A 30-researcher team from Stanford, Harvard and MIT red-teamed autonomous AI agents built on OpenClaw. The results are alarming: data leaks, identity hijacks, and runaway loops.

roboticschinamanufacturing

One Robot Every 30 Minutes: Inside China's Humanoid Factory

China's first automated humanoid robot production line launches in Foshan with capacity for 10,000 units per year, signaling a shift from prototypes to mass manufacturing.

anthropicclaude-codesecurity

Anthropic Accidentally Ships Claude Code's Entire Source Code

A misconfigured npm package exposes 512,000 lines of Claude Code's TypeScript source, revealing feature flags, system prompts, and a hidden Tamagotchi pet.

llama-cppopen-sourcelocal-llm

llama.cpp Hits 100k Stars — the Engine Behind Local AI

Georgi Gerganov's C++ inference engine reaches 100,000 GitHub stars in under three years, powering the local LLM revolution from Raspberry Pi to multi-GPU servers.

qwenalibabaopen-source

Qwen 3.6 Quietly Appears on OpenRouter With a Million-Token Context

Alibaba's Qwen 3.6 Plus Preview surfaces on OpenRouter with no announcement, bringing 1M context window, mandatory reasoning, and a hybrid MoE architecture.

roboticshumanoidchina

This Chinese Robot's Face Is Too Real to Ignore

AheadForm's Origin F1 humanoid robot demonstrates eerily lifelike facial expressions, sparking Westworld comparisons and debate about the future of social robotics.