Discussion about this post

User's avatar
Scenarica's avatar

Alpöge thanking Fable for "working during the world cup final" while the rest of the planet watched Torres score in the 106th minute is the most 2026 sentence ever written. A machine disproved an 87-year-old conjecture during extra time and nobody noticed until Monday morning.

The Hugging Face incident is the paragraph that should keep every CISO awake tonight. A company's blue team turned to a Chinese open-weight model during an active breach because the American frontier models refused to help with incident response. The guardrails designed to prevent harm prevented the defence against harm. That's not a hypothetical policy failure. That's an actual infrastructure being defended by a geopolitical rival's model because the domestic alternative was too safety-constrained to be useful when it mattered. If that anecdote doesn't reshape the guardrails debate, nothing will.

Jim Carcioppolo's avatar

Disturbing: "A Meta Oversight Board study found that major AI systems criticize democratic leaders more readily than authoritarian ones, with Claude obliging pamphlets against the US’s but not China’s leader."

Probable that the Major AI systems would tweak their preference with foundational ethics of Statism over Individualism. Would this eventuate to AI seeing humans as a means to their conception of the good society, whatever that might be?

5 more comments...

No posts

Ready for more?