Archive, page 36
Anthropic explains how teams running Claude Code on Google Cloud can track spend and demonstrate ROI, with cost-control practices for enterprise deployments.
Fujitsu will launch a trial environment for retail AI agents that support store operations and business decision-making. The trial targets retailers seeking to automate in-store workflows.
Apple ML Research paper finds LLMs can be a cheaper alternative to human judgment for machine translation evaluation, but they are also challenged.
Apple ML Research describes Minions, a distributed virtual GPU cloud that pools idle Apple Silicon Macs to run large AI inference tasks. The Register covered the work.
Apple ML Research paper proposes agents that compact stale context during long traces, based on trajectory rather than fixed token thresholds.
A research paper presenting Multics, a generative multi-player world model using hierarchical communication to avoid the quadratic attention costs of fully-connected approaches when scaling to many players.
A paper proposing Tracking-Value PPO, which models value evolution along a trajectory to give token-level credit for language-model reasoning, addressing sparse terminal rewards where standard critics fail to track value changes.
A research paper presenting X-Ragent, a tool-augmented reinforcement learning method for chest X-ray interpretation that learns when external evidence is needed and filters inapplicable quantitative evidence from diagnoses.
A contest-winning microscopy image was disqualified after judges determined AI had been used in its creation, Nature reports.
Nature reports research showing hidden signals in sleep data can predict future health outcomes.
A Nature piece describes a user telling ChatGPT to 'continue' and the model taking the instruction much further than intended.
A Nature guide covers how to search scientific literature using local AI systems that run privately and offline.
Nature reports a machine learning analysis of cilia-driven particle transport that distinguishes primary ciliary dyskinesia from normal cilia.
A paper introducing VEDJE, a video-efficient discriminative joint encoder that compresses features within sampled frames while caching representations separately, using feature-change prediction as an auxiliary training signal for retrieval.
A research paper addressing on-policy self-distillation, where a teacher conditioned on privileged information supervises a student of the same model, proposing verbal regularization to reduce bias from that privileged information.
PUFsecurity is positioning a silicon root of trust as the foundation for AI agent security, per EE Times Asia. The hardware approach aims to authenticate agent identities at the chip level.
Report says sales of AI gadgets in China rose sharply during the holiday period. No specific products, figures or brands are named in the source text.
Korean outlet Digital Today reports major technology companies are pushing local, on-device AI for PCs. No specific products or dates are given in the source text.
Qualcomm's CEO says AI companies want smartphones to run 100-billion-parameter models continuously by 2028. The source calls this probably doable with an ASIC NPU; no product is announced.
EU tech chief Henna Virkkunen told Reuters the AI Act's guardrails can address risks from rogue AI agents, saying safety of very capable models is a hot topic.
Sakana AI shows a working recursive self-improvement loop where a model improves at both solving and judging tasks, though it cannot detect its own blind spots.
Decision 3.0 is a family of five open multimodal models from 0.8B to 27B parameters that answer choice, yes/no and score questions about text and images in one call without generating text. Released under Apache-2.0 on Hugging Face.
Mira Tours is an AI walking tour guide app launched on Product Hunt, where it is open for discussion and linked for access.
ExcaliClaw is a tool where an AI agent draws hand-drawn style diagrams alongside the user. Launched on Product Hunt; availability details are not given.
Manifold MCP is a Model Context Protocol server that gives AI agents SEO, AI answer, social, ads and lead data. Launched on Product Hunt; availability details are not given.
Amazon is reportedly negotiating to acquire AI startup Decart for around $7 billion, after Nvidia and Anthropic reportedly walked away. It also continues talks with Reflection AI.
Swiss outlet Pomona reports AI glasses are being bought four times as often as last year, despite widespread criticism of the product category.
Scrape Anything is an API that turns any website into clean structured data with one call, aimed at AI and data workflows. Launched on Product Hunt.
Claude Haiku 5.5, Anthropic's smaller model, reaches GitHub Copilot, with pricing and best-use guidance published. The source text does not give full pricing or context details.
Takes for Mac is a Mac app for recording talking-head videos, where Claude Code handles the editing. Launched on Product Hunt with discussion and a link.
Nvidia is in talks to acquire a $25 billion AI startup, per Stocks Down Under. The report did not name the target company.
Dropstone CLI is a coding agent that runs in the terminal and learns over time. Launched on Product Hunt; availability details are not given.
Nvidia's latest DLSS 5 model doubles performance and reduces 4K VRAM usage by 20%, shrinking the model footprint 24% to about 118 million parameters; users report adding frame gen via Claude Opus 5.5.
OpenAI says its textGrain invisible watermark on ChatGPT text cannot be stripped by watermark removers. It will be applied to ChatGPT text in the EU, but not by default for Malaysian users.
Flahy Inc. applies knowledge graphs to clinical decision support, linking biological and patient data to inform prevention and treatment steps, per SiliconANGLE.
Vals AI research finds teams of AI agents barely outperform solo agents but cost up to 5.1x more. Only one of four tests with GPT-6 Sol and Claude Opus 5.5 showed measurable gains.
SupoClip is an open-source AI clipper that cuts long videos into short viral clips. Launched on Product Hunt with a link and discussion.
Apple announced a new Apple TV 4K with an A19 chip, an upgraded easier-to-find Siri Remote with microphones, 14 new features, and Siri AI coming to older Apple TV 4K models. The remote can be bought separately.
The intensifying AI safety debate is bringing a small group of third-party evaluators into the center of a multitrillion-dollar industry, CNBC reports. These gatekeepers test frontier models before release.
Epoch AI and Anthropic independently found current models can run experiments but lack scientific self-criticism. Sol reached 15 percent of the human reference score at best.