1. The AI money was right there. A former Meta engineer built a website instead. Craig Campbell had the resume to raise for an AI startup. He was an engineer at Meta, then a founder who sold his last company, an e-commerce tool for businesses, in 2022.

2. Endava says an OpenAI agent cut requirements analysis from weeks to hours Three customer stories landed on OpenAI's site, spanning a startup, an IT services contractor, and Japan's largest bank.

3. Anthropic raised $65 billion and passed OpenAI; in Paris, Mistral stopped calling itself a model company Anthropic has overtaken OpenAI as the most valuable private AI company in Silicon Valley, according to a Qazinform report.

In Brief

• SoftBank pledges €75 billion for French data centers SoftBank said it will spend up to €75 billion building data centers in France, targeting 5 gigawatts of new capacity. The plan adds to a wave of European compute commitments tied to AI demand.

• Meta builds an AI pendant Meta is reportedly developing a wearable AI pendant, extending its push into AI-powered hardware. The device signals Meta wants an always-on capture point for its assistant.

• GitHub Copilot switches to token-based billing GitHub moved Copilot to token-based pricing, and developers pushed back hard, one calling it "a joke." The change replaces flat-rate access with metered charges tied to usage.

• Anthropic markets Claude Opus 4.8 on "honesty" Anthropic released Claude Opus 4.8 and pitched it as more honest, claiming it avoids stating things it cannot support and admits when it errs. The company says it trains all models to decline unsupported claims.

• Study finds LLMs keep believing statements flagged as false Researchers fine-tuned models on claims explicitly labeled false and found the models still represented them as true. The tests showed a bias toward confidently asserting the flagged content.

• "Future of Truth" author fumbles questions about his AI use WIRED pressed the author of a book on AI and reality over AI-generated quotes in the text. His explanation collapsed under questioning, and reviewers flagged further problems.

• Arizona graduates boo Eric Schmidt over AI University of Arizona students booed former Google CEO Eric Schmidt after he told them to help shape AI. MIT Technology Review tracked the reaction as part of growing campus resistance.

• WIRED tests whether paid transcription beats free tools WIRED compared Wispr Flow and other paid AI transcription apps against free alternatives. The takeaway: paid accuracy gains are narrow for most everyday use.

• GenClaw lets an LLM edit the image canvas directly Researchers built GenClaw, which generates code so a language model can manipulate the canvas instead of only rewriting prompts. The approach aims to break agents' dependence on black-box image models.

• CollectionLoRA packs 50 editing effects into one adapter Researchers distilled 50 image-editing effects into a single LoRA using multi-teacher on-policy distillation. The method cuts the storage and loading overhead of swapping many separate effect adapters.

Read the full edition →