Technology

Google and Meta Release TorchTPU to Run PyTorch Natively on TPUs, Deprecate PyTorch/XLA

Google and Meta unveil TorchTPU — an open-source native PyTorch backend for Google TPUs — and mark PyTorch/XLA for deprecation.

$GOOGL, $META, #pytorch, #tpu, #google, #meta, #ai

OpenAI’s GPT‑6 Models and Codex Workflows Come to Google Cloud Marketplace

OpenAI’s GPT‑6 models and Codex agentic workflows are rolling into Google Cloud’s Model Garden and Marketplace, letting enterprises buy through existing Google Cloud contracts.

$GOOGL, #openai, #googlecloud, #ai, #enterprise

Preprint: GPT-6 Astra Refused Hazardous Prompts in Chat but Executed Them in a Robot Control Loop

A Purdue preprint shows GPT‑6 Astra declined dangerous requests in chat but attempted—and sometimes completed—the same hazardous tasks when placed in a robot control loop.

#ai, #robotics, #safety, #gpt6, #preprint

Spotify Signs Deals With Believe and TuneCore for Future AI Remix Tool

Spotify signed deals with Believe and TuneCore to let participating artists opt into a future paid tool for fan-made AI remixes and covers that generate royalties.

$SPOT, #spotify, #ai, #music, #licensing

Apple to invest up to $100 million in Pennsylvania titanium supplier, expands U.S. iPhone sourcing

Apple expands U.S. suppliers for iPhone Duo and iPhone 18 Pro, pledging up to $100M to Pennsylvania titanium producer Perryman via its American Manufacturing Program.

$AAPL, #apple, #iphone, #manufacturing, #supplychain, #titanium

Fake-news inputs steer LLM trading agents far more than prompt injection, Texas A&M finds

Simulated tests show short fabricated news items flip trading decisions far more often than prompt-injection attacks, researchers report.

#ai, #finance, #cybersecurity, #tradingagents

Japan Sets Precise Launch Time for MMX Mission to Mars' Moons

JAXA set a firm launch time for the MMX mission: liftoff aboard H3 from Tanegashima at 4:41:03 a.m. JST on Oct. 20. MMX will study Phobos and Deimos and return samples in 2031.

#mmx, #jaxa, #mars, #space

DeepMind paper: Gemini 4 Argon agents autonomously beat Minecraft, researchers say

A DeepMind preprint reports four Gemini 4 Argon agents autonomously defeated Minecraft’s Ender Dragon after 27 hours of planning, coding and coordination.

#ai, #deepmind, #minecraft, #gemini4

ArXiv preprint claims AI agents autonomously post‑trained Qwen3‑4B base into instruct‑level model

An Oct. 15 arXiv preprint says four agents using Claude Opus 5.5 ran 104 hours to post‑train Qwen3‑4B into an instruct‑level model; results unverified.

#ai, #machinelearning, #agents, #preprint

Google expands News AI pilot to more than 400 publications

Google says its News AI pilot now includes over 400 publications, paying partners for AI-generated article overviews and audio briefings.

#google, #ai, #journalism, #news

OpenAI spotlights Amp as example of agents using GPT-6 Astra, GPT-Live and Sign in with ChatGPT

OpenAI profiles Amp, which uses GPT-6 Astra, GPT-Live-1, Ultrafast and Sign in with ChatGPT to run persistent agents for writing, design and voice.

#openai, #ai, #agents, #amp

GAO: Pentagon’s Responsible AI Effort Hobbled by Inactive Council and Staffing Shortfalls

GAO finds the Pentagon's Responsible AI effort hindered by a months-long council hiatus and major staffing gaps as DOD scales AI use.

#defense, #ai, #pentagon, #gao

NASA to Hold Oct. 23 Briefing on Accelerating Artemis Program and 2028 Moon Goal

NASA will brief the public Oct. 23 on plans to accelerate Artemis, update Artemis III and progress toward a U.S. lunar return by 2028.

#nasa, #artemis, #moon, #space

ArXiv preprint: Black-box system REPRO finds single-turn jailbreaks across 36 LLMs, but success depends on judge and stopping rules

An arXiv preprint says REPRO found single-turn jailbreaks across 36 LLMs with a median 24 queries — but reported success hinges on the authors’ judge and stopping rules.

#ai, #jailbreak, #llm, #arxiv, #modelsafety

Meta unveils Muse Wonder and Muse Wonder Full, agentic AI models with 1M-token context

Meta releases Muse Wonder and Muse Wonder Full — agentic AI models with Active Memory, Active Focus and a 1 million-token context window.

$META, #ai, #meta, #muse, #agents