Also: Anthropic's IPO filing shows an $80B loss, a cheap model that matches a frontier one with the right setup, and Perplexity goes local on Mac.
Plus: the OpenAI spy case takes a new turn, a hack drains developers' AI keys, and why "unlearning" is now the skill employers want most.
Also: Anthropic's AI agents now outwork its staff 6 to 1, AI cloud stocks drop 20%, and how to pick the right frontier model for each job.
Plus: Amazon's new Bee wearables, a voice AI that settles 70% of Starlink support calls alone, and a movie marathon prompt for the weekend.
Plus: 400 scientists call for real AI oversight, a $40K AI hunt finds a Chrome flaw, and parents test AI tutors at home.
Also: Apple's new HomeView hub, Gemini 4 Argon opens up to more users, and Anthropic's daily code merges jump from 2,500 to 30,000.
Plus: people are rebuilding apps they are tired of paying for, Microsoft's new chip and decision model, and Karpathy's map of who really uses AI.
Plus: Satya Nadella's 7 rules for keeping AI in check, a $1.8B bet on a virtual cell, and 3 mods that make Claude Code easier to live with.
Also: Claude builds live dashboards and explainer videos, Haiku 5.5 costs 90% less, and why your AI grader may be far too generous.
