Google DeepMind unveils Genie 4 world model with native audio
Read original source →deepmind.googleGenie 4 generates explorable worlds in real time from text and image prompts, with native ambient sound, music, and character dialogue, and lets users steer scenes mid-simulation via natural language. It builds on Gemini models — Gemini Omni anchors realism while other Gemini models maintain coherence.
How this story unfolded
1 day · 11 reports · 9 community posts · 20 of 22 shown
- Oct 27
Genie 4: A new generation of world simulationdeepmind.google
Genie 4: A new generation of world simulationyoutube.com
We’re announcing Genie 4, our new world model for real-time interactive simulation.blog.google
Google Upgrades AI World Model That Rattled Video Game Stocksbloomberg.com
Google’s Genie 4 world model adds audio, more realistic worlds, and real-time steeringtechcrunch.com
Google’s next Genie AI model generates more realistic experiencestheverge.com
DeepMind unveils Genie 4 world model promising more realistic physics and humansarstechnica.com
Google built AI agents a world to practice in. The physics still aren’t quite right.thenewstack.io
- Oct 28
More stories today
UNESCO-led study tests 33 LLMs on climate science accuracy
Flagship models scored around 78% on the hardest questions — how confident climate scientists are in IPCC conclusions. The UCL/UNESCO study ran 600 questions across 33 models, including Claude Fable 5, Claude Opus 5, Gemini 3.1 Pro Preview, Gemini 3.6 Flash and GPT-5.6 Sol.
Ars Technica·2 hours ago
Cisco upskilled 1,200 engineers in agentic AI via AWS AI League
Cisco used the AWS AI League program to move 1,200 engineers from watching agentic AI demos to building with it, per an AWS customer post. The cited obstacle was not access to the technology but getting thousands of engineers to build confidently.
AWS AI Blog·2 hours ago

SageMaker AI adds cross-Region inference routing
Amazon SageMaker AI now routes inference requests to another AWS Region when the serving Region runs short of GPU capacity, keeping managed model endpoints available during regional shortfalls.
AWS AI Blog·2 hours ago

Google shows multi-turn agentic RL with Tunix and Gemma 4 on TPUs
Gemma 4 E2B wins only 9.4% of Plane Strike games out of the box; after 100 steps of agentic GRPO with Tunix on a single Cloud TPU v6e-8, win rate hits 95% and games shorten to ~16 strikes from 22.44.
Google Developers Blog (AI)·2 hours ago

OpenAI built Performance Factory to keep ChatGPT fast
OpenAI's Performance Factory coordinates Codex-powered agents that investigate performance regressions, propose fixes for engineers to review, and verify whether those fixes hold in production.
OpenAI Blog·2 hours ago

Skyflow and Bedrock AgentCore Gateway guard agent tool calls
AWS post shows adding Skyflow Detect to Amazon Bedrock AgentCore Gateway as a request and response interceptor to keep sensitive data out of the tools agents call and out of the model context those tools feed.
AWS AI Blog·2 hours ago

Perplexity keeps Decider-v1.3 and Flash free on OpenRouter for 3 more weeks
Aravind Srinivas·2 hours ago
Cerebras video traces chip design from CPUs to wafer-scale
Cerebras DevX's Alycia Cary walks through the engineering choices that took computing from CPUs to GPUs and TPUs, then explains why Cerebras built a processor the size of a wafer for faster LLM inference.
YouTube·2 hours ago