Cohere talk: learning science behind LLM adaptive tutors

OpenAI says it will honor a shared robots.txt signal letting site operators state their AI training preferences to participating providers. The mechanism is opt-out via an existing web standard rather than a new per-crawler control.

A bankruptcy court judge said Character Technologies should hire a privacy ombudsman to assess whether its AI platform harms children as part of its plan to license technology to Walt Disney Co. The condition was raised at a Friday hearing.

Yair Alon, who spent a decade at Google Research and DeepMind building the deep learning models behind Pixel Portrait Mode, is founding research scientist at Poetiq (YC S26). Poetiq is building an AI system that can optimize any other AI system, including itself.

Zest's text-based agent pulls dining history from linked credit-card transactions via Plaid, then suggests restaurants, runs group votes, and books reservations through connected services. The startup raised $1.8M pre-seed from Alexis Ohanian's 776 and Kindred Ventures.

CEO Brian Niccol said Starbucks will explore drive-thru experiences that recognize customers and present personalized digital menus, with background AI models for order entry testing from 2027. The chain already takes orders via ChatGPT and runs an AI Green Dot Assist tool for baristas.

Nathan Lambert argues GLM-5.3, not Claude Mythos, is the model that crossed the cyber-capability threshold, with little public evidence of change a month after its weights dropped. He says the debate focuses on which models cause harm rather than how the cybersecurity industry holds its equilibrium.