AI Reading for Friday March 28
Anthropic wants to build a nation-sized army of benevolent geniuses - WIRED
Anthropic scientists peer into how AI actually ‘thinks’ — and how it plans ahead and sometimes lies - VentureBeat
Deep research 'super agents' might demand colossal amounts of compute - Business Insider
Microsoft data center pullback seen driven by OpenAI moving to its own compute, need to maximize density and power - The Register
Clinical test says AI can offer therapy as good as a certified expert - Digital Trends
Speeding up grading of complex papers with AI - Physics
AI’s impact on messy vs. more predictable linear process jobs. - FT
AI can be hard to characterize, it's good at a first cut at very messy stuff, reading a ton of unstructured data and synthesizing. But it's not predictable or coachable.
"I genuinely don't understand why some people are still bullish about LLMs." - X
CoreWeave offering chopped by 66%, Nvidia buys 20%, looks like a flop - FT
Columbia student who created AI tool that helps people cheat in coding interviews gets 1-year suspension - Business Insider
Unsecured database shows how China uses AI to detect and repress dissent - TechCrunch
Useful chart to understand AI agents - @mmitchell.bsky.social
OpenAI's top models crash from 75% to just 4% on challenging new ARC-AGI-2 test - THE DECODER
Vibe coding a Musk vs Altman racing game. - Business Insider
What if Studio Ghibli directed Lord of the Rings? - Reddit
Follow the latest AI headlines via SkynetAndChill.com on Bluesky