The Archive

Google is launching a "universal" Gemini AI agent that can work across apps and devices in the background. The tool, announced as part of the Gemini at Work event on Thursday, will be available within the Gemini Enterprise app, allowing users to chat with the Gemini agent and assign it tasks from a single interface. In addition to having Gemini work directly inside Workspace apps like Gmail, Drive, Docs, Sheets, Calendar, and others, users can interact with the agent from their mobile device, desktop computer, the web, and third-party apps like Slack or Microsoft 365. Since it runs in the clo...

I’ve spent a year living with Alexa Plus, and it both delights and frustrates me, which is the smart home in a nutshell. | Photo by Jennifer Pattison Tuohy / The Verge Of all the things Alexa Plus can do, I never expected it to make me cry. Since my son left for college, the Echo Show in my office has been making then-and-now photo montages: the toddler with his first tennis racket next to the captain of his varsity team; me hugging him as a tween beside me hugging him as a young man. I've teared up more than once. But then comes the ad, a full-screen photo of a hideous brown leather recliner...

My Decoder guest today is Hayden Field, The Verge’s senior AI reporter, and we’re discussing the new wave of consumer-friendly AI agents. If you’ve been paying attention to this space, you know AI enthusiasts have been using agents for a minute now — homebrew OpenClaw setups led to a surge in Mac Mini sales earlier this year. But the launch of Meta’s Muse, OpenAI’s Dots, and xAI’s Grok bot has brought easy to use agents to millions. Muse and Dots have had the highest-profile product launches, and they’re fascinating to pit against each other. Both Meta and OpenAI have decided to pitch these a...

At the New York Film Festival premiere of Artificial, Luca Guadagnino's satirical Sam Altman biopic, the director said onstage that "[when] someone wants to play God, that's very interesting to me." The idea of playing God, and power in general - who has it, who desperately wants it, and who will do anything to get it - is central to the film's narrative, which sticks remarkably close to the factual events surrounding the OpenAI CEO's rise to power with, and brief ouster from, the AI lab. The film opens with a stunning shot of San Francisco's Golden Gate Bridge, which will turn into a metapho...

Zach Yadegari, the teen co-founder of popular Cal AI calorie tracking app, has launched a new personal AI agent startup that competes with Instinct, Muse and Bee.

Google’s new AI Edge Foresight app takes on Granola with an offline meeting note-taker that can transcribe conversations, generate notes, and answer questions using on-device AI.

Boyu Capital and IDG Capital led the funding round, and existing shareholders Tencent, HSG (formerly known as Sequoia China), ZhenFund and others also participated.

Anthropic commits $150M over 3 years to Genesis Mission, a federal initiative to accelerate scientific discovery via AI.

Pollo AI uses GPT-5.6, GPT-6 Astra, and GPT-Image-2.5 to generate images and video ads from creative briefs.

LegalOn reduced Codex API costs by 65% while maintaining development velocity through strategic task-model matching and budget optimization.

Full-stack safety solution for physical AI is being used by robotics companies.

DeltaReplay enables mobile GUI agents to selectively reuse execution trajectories from memory by matching task-relevant steps without forcing irrelevant guidance.

Controlled study finds illumination inconsistencies provide no statistically significant improvement over blending artifacts for face-swap detection across datasets.

Mathematical audit of latent intervention anchors in synthetic vascular predictive models reveals coordinate transformation gaps between least-squares equivalence and edit semantics.

Opinion piece argues unsupervised learning lacks unified goals; identifies distribution estimation, data generation, feature extraction, and interpretability as distinct objectives.

3D characterization framework projects sequential decision-making AI methods onto MDP formalism, autonomy level, and computational cost for unified puzzle evaluation.

Structured sentiment analysis via dependency graph parsing using sequence labeling to extract fine-grained sentiment holders, targets, expressions, and relationships.

VenusRX-Bench unifies enzymatic reaction prediction, retrosynthesis, and EC-number annotation with pretrained model linking reactants, products, and catalytic function.

Jev decision model tested as Q-function and policy provider in RL; single-pass generation without token-by-token decoding compared to foundation model priors.

Graph model for conflict resolution extended with epistemic state-preserving actions (announcements, leaks) that alter beliefs without changing physical game state.

Hi3D 3.0 with Twinkle3D generates object-specific 3D assets at 2048³ resolution, preserving fine details like inscriptions and brand marks from input images.

Ablation study finds camera-noise residuals don't improve deepfake detection over RGB-only baselines on FaceForensics++.

TRACE protocol diagnoses verifier brittleness in LLM agent evaluation by isolating scoring rule changes from capability changes.

Latent memorization in diffusion models detected via basin geometry before generated samples reveal it, enabling early audit.

σTransfer enables zero-shot transfer of Laplace approximation prior precision from small to large networks under μP scaling.

Evi-VN injects virtual node evidence to help GNNs detect fraud on platforms by capturing latent patterns in hard regions.

Constrained command-conditioned PPO executor in MicroRTS separates strategy selection from policy execution for robustness.

Lamarckian evolutionary framework discovers autonomous driving training distributions via competitive selection instead of surrogate metrics.

DIAL-OPD improves on-policy LLM distillation by selecting high-value tokens and filtering low-probability pairs.

Study shows harness evolution via runtime optimization hits limits; weight training gains persist only on learned models via failure diagnosis.