AI #168: Not Leading the Future

This is what a lull looks like at this point. The government is having internal arguments. The models are getting improved internally. The coding agent improvements are all what we would expect. There’s still a lot happening, including a bunch of cool papers, but I feel able to relax and to take care of some other work while I have the chance. You never know when that chance will be over. Table of Contents From yesterday: Cyber Lack of Security and AI Governance. Language Models Offer Mundane Utility. Fix everything now. Language Models Don’t Offer Mundane Utility. Travel is harder than it looked. Huh, Upgrades. Opus 4.7 fast mode, Claude Code /goal and agent view. Levels of Friction. AI for tax avoidance. On Your Marks. PrinzBench, ProgramBench and faster harmfulness checks. Get My Agent On The Line. Mona tries to run a cafeteria. Mistakes were made. Deepfaketown and Botpocalypse Soon. Soon. But not quite yet. Fun With Media Generation. Monet does not seem that great. On AI Writing. AI is a hack writer using hack techniques. A Young Lady’s Illustrated Primer. How to make AI help and not hurt learning. You Drive Me Crazy. OpenAI is sued over the FSU shooter. They Took Our Jobs. Skilled or unskilled? Overstaffed massively? Which is it? The Art of the Jailbreak. OpenAI tries to permanently plan Pliny. Introducing. OpenAI development company, MIRI introduces AI Stopwatch. Claude Has Its Limits. Automated Claude subscription use gets distinct budget. Show Me the Money. OpenAI employee cash outs, Anthropic transfer rules. Show Me The Compute. Money, dear boy. The markets win again. Quiet Speculations. Projections for Anthropic continue to be conservative. Quickly, There’s No Time. Engineers report 2x speedup, don’t anticipate enough. Chip City. Does Nvidia have a China problem? Pick Up The Phone. China is worried about ChatGPT. The Week in Audio. Claude’s Constitution, Derek Thompson on jobs. Rhetorical Innovation. Names have power. Not Leading the Future. Nevertheless,