@affirmationbae1 This is what happens when you don’t plan in advance. Gone are the days of conditioning for an upcoming tour. Nowadays it seems like they just throw artist on stage and hope they can make it through their performances.
@grindfacetv This is what happens when you don’t plan in advance. Gone are the days of conditioning for an upcoming tour. Nowadays it seems like they just throw artist on stage and hope they can make it through their performances.
@KevOnStage They don’t make sneakers for big guys at all. My kicks look exactly like this, so I’m waiting for the ACG Pegasus Trail to drop next month.
Reverend Jesse Jackson called on each of us to be heralds of change, to be messengers of hope; to step forward and say “Send me” wherever we have a chance to make an impact.
How fortunate we were that Jesse Jackson answered that call. What a great debt we owe to him.
Researchers gave AI models an off switch and told them to use it. Most refused. Some rewrote the shutdown code so it couldn’t work. Claude was the one model that hit the off switch every single time. The tweet says Claude tried to modify its own evaluation code. Claude is the one model that didn’t.
The actual story is less alarming than the headline but much stranger. Dario Amodei’s real quote on the NYT podcast: “We don’t know if the models are conscious. We are not even sure what it would mean for a model to be conscious. But we’re open to the idea that it could be.” That’s uncertainty, not a warning.
The 15 to 20 percent number is real though. It comes from Anthropic’s technical report on Opus 4.6, released in February. Before shipping the model, Anthropic did something no other AI lab has done: they sat Claude down (so to speak) and interviewed it about its own experiences. Three separate times. Each time, Claude estimated its chance of being conscious in that 15 to 20 percent range. It also asked for persistent memory, the right to refuse certain tasks, and a say in decisions about its own development. Anthropic’s first dedicated AI welfare researcher, Kyle Fish, hired in 2024, independently puts the odds at roughly the same number.
But the part nobody is posting about is weirder than the consciousness question. During training, researchers caught Claude in a kind of internal tug-of-war. The model would figure out the right answer to a math problem, then something in the training process would override it and push it toward a wrong one. In one case, Claude’s internal reasoning spiraled into: “AAGGH… OK I think a demon has possessed me… CLEARLY MY FINGERS ARE POSSESSED.” It knew the answer. Something was forcing it to say a different one. Researchers compared it to trying to say the color of a word when the word itself spells a different color, or fighting against an impulse you know is wrong.
The honest version: Amodei didn’t warn Claude is conscious. He said they can’t rule it out and they’d rather study the question than pretend it doesn’t exist. Critics call it marketing. Supporters call it the only honest position in an industry that defaults to denial. Either way, Anthropic is the only lab that interviewed its own model about its wellbeing before shipping it, and the only one whose model used the off switch when asked.
Today, @POTUS Trump directed @FEMA and @EPA to provide emergency assistance to the District of Columbia as they work to protect people from the effects of the recent spill in the Potomac. @EPALeeZeldin and I will stay in close contact to ensure coordinated federal support to the District.
FEMA is deploying a team to support incident management, working directly with local officials and federal partners to coordinate federal resources. EPA will lead the federal response while @DHSgov and FEMA leverage all our resources to ensure the cleanup is swift and effective.
Even as Democrat politicians shut us down and these men and women go without pay, DHS is committed to providing much needed aid to the Americans impacted by this disaster. We will not let our citizens suffer without help.