Anthropic & AI Industry Deals
Anthropic Research · www.anthropic.com
Anthropic presents an interpretability tool that identifies behavioral differences between AI model versions, going beyond standard evaluation suites to surface changes that benchmarks might miss.
Anthropic has announced chip deals with Google and Broadcom worth hundreds of billions of dollars, securing computing capacity as its annualized revenues reach $30 billion.
Financial Times · www.ft.com
Kyle Kosic has joined Jeff Bezos's secretive AI startup Project Prometheus, which is working on systems that can understand and reason about the physical world.
AI, Vibe Coding & Software Engineering
Bram Cohen argues that the enthusiasm around vibe coding—building software purely through AI prompts—has become excessive, likening it to an extreme form of dogfooding that ignores real engineering discipline.
The post frames multi-agent AI coding as fundamentally a distributed systems challenge, arguing that coordination problems between agents persist regardless of model capability improvements.
A GitHub issue reports that recent Claude Code updates have significantly degraded its ability to handle complex engineering tasks, drawing substantial community discussion.
A developer recounts their attempt to build an RSS reader entirely via AI-assisted vibe coding and finds the results fell short of expectations.
Privacy, Security & Adtech
Financial Times · www.ft.com
Shenzhen-based Even Realities positions its smart glasses as a privacy-conscious alternative to Meta's, pitching a premium product that does not harvest user data.
A website owner explains why they abandoned Google AdSense after two decades, reflecting on the declining returns and changing dynamics of display advertising.
A security researcher discovered that BrowserStack's local testing tool was leaking private keys, raising serious security concerns for users of the platform.