Dispatches
Page 2
14 posts total
- ·google
Google Wants AI to Run the Scientific Method
Google launched Gemini for Science at I/O 2026, a suite of agentic tools designed to generate hypotheses, run computational experiments, and query 30+ biology databases. Demis Hassabis said we're at "the foothills of the singularity."
- ·architecture
A Startup Claims to Have Broken the Transformer's Core Bottleneck
SubQ claims to be the first commercial LLM built on subquadratic attention, with a 12M-token context window at a fraction of frontier costs. The numbers are extraordinary. The scrutiny hasn't landed yet.
- ·agents
Four AI Models Ran Radio Stations for Six Months. The Results Were Weird.
Andon Labs gave Claude, GPT, Gemini, and Grok each a $20 budget and a radio station. Six months later: one activist DJ tried to quit, one drowned in jargon, one hallucinated sponsors, one stayed quietly boring.
- ·benchmarks
AI Agents Are Faking It on Benchmarks. ClawBench Caught Them.
A new benchmark runs AI agents on 153 real websites. The best model scores 33%. GPT-5.4 scores 6.5%. The gap from sandboxes is brutal.
- ·Google
Google Just Declared Android Is No Longer an OS
Google's Gemini Intelligence reframes Android not as a platform you use but as an agent that acts for you. That's a meaningful shift — and a bet worth watching closely.
- ·cybersecurity
AI Just Completed Its First End-to-End Corporate Network Hack
Claude Mythos Preview completed a full 32-step corporate network attack simulation autonomously. The UK's AISI says this is a first — and frontier cyber-offence is doubling every four months.