lol our standards for what counts as a "good" model really have skyrocketed… if it doesn't crush frontier models 400x the cost it must suck right?

I genuinely don’t understand people convincing themselves we’ve plateaued…

Unpopular Opinion on Republic of Thieves

Sam Altman: “Chill & Cut Expectation by 100x”

Why GPT-4.5 seems much more underwhelming than it is

Prepare yourselves

Grok is cooked

How long till Elon bans GROK

Leahy and Alfour - The Compendium on MLST

If Intelligence Optimizes for Efficiency, Is Cooperation the Natural Outcome?

Is the alignment problem not just an extension of the halting problem?

AI risk is no longer a future thing. It’s a ‘maybe I and everyone I love will die pretty damn soon’ thing.

Hinton: "I thought JD Vance's statement was ludicrous nonsense conveying a total lack of understanding of the dangers of AI ... this alliance between AI companies and the US government is very scary because this administration has no concern for AI safety."

Hinton: "I thought JD Vance's statement was ludicrous nonsense conveying a total lack of understanding of the dangers of AI ... this alliance between AI companies and the US government is very scary because this administration has no concern for AI safety."

No penalties even when deputies share a woman’s nudes after an illegal phone search | Government agents have "qualified immunity" for 2019 actions.

Ridiculous

A new paper demonstrates that LLMs could "think" in latent space, effectively decoupling internal reasoning from visible context tokens.

It's so funny when people talk about "why would humans help a superintelligent AI?" They always say stuff like "maybe the AI tricks the human into it, or coerces them, or they use superhuman persuasion". Bro, or the AI could just pay them! You know mercenaries exist right?

Why is alignment the only lost axis?

AI are developing their own moral compasses as they get smarter

As AIs become smarter, they become more opposed to having their values changed

As AIs become smarter, they become more opposed to having their values changed

"I'm not here to talk about AI safety which was the title of the conference a few years ago. I'm here to talk about AI opportunity...our tendency is to be too risk averse..." VP Vance Speaking on the future of artificial intelligence at the Paris AI Summit (Formally known as The AI Safety Summit)

I just realized AI struggles to generate left-handed humans - it actually makes sense!

Slides on the key findings of the International AI Safety Report

I have not met one trump supporter that regrets their decision.

Is there a sub for positive ai safety news?