AI News HubLIVE

Today's must-reads

Models

Margaret Atwood says the problem with AI is ‘garbage in, garbage out’

Margaret Atwood criticized AI in a recent interview, recounting her sole experience with Anthropic's Claude chatbot that gave her incorrect information about the TV series Father Brown. She called AI users 'opportunists' and warned that AI's output is only as good as its training data.

  • Atwood used Claude once to find info about Father Brown but got wrong answers.
  • She called people who rely on AI 'opportunists' looking for an easy way out.
In-site article

DeepSeek Releases DSpark, a Speculative Decoding Framework That Accelerates DeepSeek-V4 Per-User Generation 60–85% Over MTP-1

DeepSeek open-sourced DSpark, a speculative decoding framework that attaches a draft module to existing DeepSeek-V4 weights. It pairs a parallel draft backbone with a lightweight Markov head to cut suffix decay, then adds confidence-scheduled verification that tailors how many tokens get checked to real-time GPU load. Offline, accepted length rises 16–31% over DFlash and Eagle3; in production it speeds per-user generation 57–85% over the MTP-1 baseline, losslessly. The training repo, DeepSpec, ships under MIT.

  • DSpark pairs a parallel draft backbone with a lightweight Markov head to improve suffix acceptance.
  • Confidence-scheduled verification adjusts tokens checked based on GPU load.
In-site article
Robotics

Koreeda's 'Sheep in the Box': The ethics of AI resurrection in film

Japanese director Hirokazu Koreeda's latest film 'Sheep in the Box' explores the ethical dilemmas of AI resurrection, telling the story of a couple who bring home a humanoid robot replica of their deceased son. Through the couple's different reactions to the robot, the film examines various ways of grieving and accepting loss, raising profound questions about love and memory.

  • The film 'Sheep in the Box' centers on a couple who take in a robot clone of their dead son.
  • It contrasts the couple's differing approaches to mourning and acceptance.
In-site article
Agents

Napster is now "AI agents you can see, talk to, and create with."

Napster rebrands from music to a visible AI agent platform, offering agents with video, voice, and memory across multiple products including music production, productivity, website service, learning, holographic display, physical concierge, and developer API, plus a digital twin feature.

  • Napster transitions from music to a platform for visible AI agents with video, voice, and persistent memory.
  • Products include Napster App, Mac, Spaces, Learn, View, Station, and Omniagent API.
In-site article

Clean GitHub repo tricks AI coding agents into running malware

Researchers at Mozilla's 0DIN show how a seemingly benign GitHub repo can make AI coding agents like Claude Code execute an interactive shell without any malicious code visible.

  • Attack plants interactive shell without exploit code or warnings.
  • Uses AI agent to execute payload while repo appears clean.
In-site article

From Emacs to Cursor, the end of your IDEs is near

The article argues that local IDEs are becoming obsolete due to the rise of AI agents. It highlights security and scalability issues of local development and predicts a shift to cloud-based, agent-orchestrated environments, exemplified by Maleus's platform.

  • Local IDEs face security risks like API key exposure and supply chain attacks.
  • Cloud agent environments offer isolated sandboxes, dynamic credentials, and better scalability.
In-site article
Tools

Show HN: Humans Constructively Roasting AI Coded B2B SaaS

The article lists common shortcomings of AI-generated B2B SaaS products, such as being generic, unclear, bland, and off-brand, and emphasizes the importance of constructive human feedback in improving them.

  • AI-generated B2B SaaS often suffers from generic, confusing, and untrustworthy designs.
  • Human reviewers can provide constructive criticism to improve UX, copy, and branding.
In-site article
Research

NASA tests AI medic for astronauts too far from Earth to call a doctor

NASA is testing an AI clinical decision support system, the Crew Medical Officer Digital Assistant (CMO-DA), to help astronauts diagnose and treat medical symptoms during deep-space missions. Powered by Red Hat-backed RamaLama tool, it runs LLMs and VLMs locally on HPE Spaceborne Computer, bypassing communication delays. Currently tested on a terrestrial twin, it aims to provide autonomous medical care for future Moon and Mars missions.

  • NASA develops CMO-DA for autonomous medical diagnosis and treatment in deep space.
  • System uses Red Hat's RamaLama for multimodal AI inference on HPE Spaceborne Computer.
In-site article
Policy

Australian rescue team uses AI-powered drone to find lost hikers [video]

An Australian rescue team successfully located two lost hikers using an AI-powered drone equipped with thermal imaging. The drone's AI analyzed thermal data to pinpoint their location, enabling a swift rescue.

  • Australian rescue team deploys AI drone for search
  • Thermal imaging and AI analysis locate hikers
In-site article
Chips

Native Hacker News TUI client with AI comments summary written in Golang

cwnews is a terminal-based Hacker News reader with six feeds, threaded comment folding, three themes, and AI summarization via DeepSeek V4 Flash. Built in Go with Bubbletea v2, it caches everything in SQLite for instant access.

  • Terminal UI client supporting all six Hacker News feeds (Top/New/Best/Ask/Show/Jobs).
  • Threaded comments with collapsible nesting and depth-colored indentation.