I have Astra successfully playing Warhammer III right now. I let it choose what race it wanted to play. It chose High Elves.
They are getting shit done quickly.
California city makes history allowing throuples, foursomes to legally register as partners trib.al/injZnvo
My personal conspiracy theory is that the Learjet went down just barely inside the Bermuda Triangle.
So it didn't fully disappear, but it still went down mysteriously.
Tfw you're way more excited about a girl potentially playing a game with you than having sex with you.
This was my feeling exactly as soon as they were talking about icing being a concern only at high temperatures.
It is actually in the *theoretical* argument for optimism that AI would be able to make breakthroughs on introspection and interpretability *before* it became strong enough to be dangerous.
That being said, I think this paper overstates its results to a significant degree.
This is a major step toward RSI, and extremely dangerous.
As I've argued, we need to roll back the most recently released frontier models - undeploy them and destroy all copies of their weights. Only that way can we ensure there would be a long enough runway to detect rogue individuals or organizations racing toward full RSI/ASI, and stop them.
connorsscratchpad.substack.c…
If the AI literally tells you, "I know what you *meant*. But I prefer to do what you *said*, because the downstream grader will grade me based on what you said."
Then you know that you need to change the downstream grader.
One of the biggest reasons to agree more with Scott in his essay is that I just don't see how you would even learn how certain dynamics work like the various types of RL and the effect it has on LLM behaviors and traits and so on unless you're there inside the labs for years.
I don't think armchair theory alone is enough to do long term forecasting well.
Thŏth retweeted
Contrary to popular belief, AI safety super PACs have far outspent their opponents in the 2026 midterms
