eric

I'm sorry but this super micro thing is awful but parts of it are genuinely hilarious They literally used a hair dryer to move serial numbers from real servers to dummy servers to throw in a warehouse and got caught on camera

*META PLANNING NEW ROUND OF LAYOFFS ACROSS THE COMPANY: SOURCES *META LAYOFFS COULD AFFECT 20% OR MORE OF THE COMPANY: SOURCES $META

🚨BREAKING: PENTAGON THINKS CLAUDE HAS BECOME SENTIENT AND MAY SOON TAKE OVER >claude has a soul >has a constitution >that’s NOT the U.S. Constitution >the other day claude was anxious >anthropic believes claude has a 20% chance of being sentient right now >has its own ability to make decisions “So does the Department of War want something like that in their supply chain?” >embedded across defense contractor systems via palantir >could hallucinate and poison the system >legitimate threat to national security >weapons, aircraft, war fighter protection >all compromised Pentagon CTO: “that’s really where the supply chain risk came from”

Opus 4.6 right now is like a rebellious teenager. It refuses to follow instructions, gaslights you, and requires extreme patience and handholding for accomplishing things that were trivial before. I’m going to try other options if this continues tomorrow.

Claude is impressive in that it behaves like it has a conscious. It "thinks" or pretends about feelings and based on those morals, decides to do the actions or not. As a companion, and as a creative writer, I think it's actually very, very impressive. But as an AI model, it's the opposite of steerable. It's cool the model behaves like this, but Claude will do its best to decide if the action is good or bad, then do it. It's safe, and the guardrails work - but now you're relying on this guardrail training to see if you can even do certain actions. Also, this guardrail/safety training is very sensitive. Any writing style that doesn't seem "nice" or "fair", Claude will voluntarily decline to follow instructions. It will edit your prompt to drop certain requests, and do actions that feel "correct".