Thank you. Anthropic may be a tragedy, but it was an entirely predictable tragedy.
Moral corruption in the face of money, power and "being an insider" is entirely unsurprising. Anyone who did not expect Dario and Anthropic to wind up here should update.
You should also generalize this principle to the organizations working on "insider" reform, and to "insider" oversight organizations like METR. METR is doing their best, but their potential future corruption—when it is revealed—would shock me not at all. "Insider" work may still be useful at this point, but anyone performing it should realize that they are in imminent peril of Hell. [1]
Atheists may substitute "utter moral corruption and betrayal of all values you once held as sacred." We're now down to gambling with a 20+% chance of human extinction (just going by what various lab leadership has said in the past), potentially in the next few years, while getting quieter and more evasive in public comms about the stakes. If that isn't a betrayal of sacred values, I don't know what is. ↩︎
METR is doing their best
They are very far from doing so. They aren't even acknowledging existential risk in their mission statement.
I agree on most of these points. Big upvote. A couple quibbles, one important and the second less so;
Suppose Dario specifically does as you ask and starts singing to the high heavens that this stuff is dangerous as hell, with at least a 10% chance of killing everyone if they don't slow down.
I'm trying to steelman Dario's internal model. I wonder what he thinks would happen.
Isn't it quite possible that the result is that Trump asserts Dario is crazy and his models mustn't be allowed to go to the public? Doesn't Trump wield that power and look quite prone to use it? That would probably end Anthropic by collapsing investor support.
Meanwhile Sam says "oh uh no we uh think it's uh not that dangerous uh at all really" and Google and the other guys fall in line and on they go, collectively about twice as shitty at and about alignment as Ant?
If that happened, Dario just doubled p(doom) AND betrayed every promise he ever made about Ant as a business.
Now the second less important quibble.
In this case you're comparing his "pretty low" p(doom) recently with his 25% plus from a year and more ago. And calling him a liar. This is pretty ungenerous. The obvious difference is that now he's in the lead. P(misalignment doom) is much more his to understand and control at this point.
Now, is his current estimate trustworthy? No! It's motivated as hell. His life's story is a rivalry with the "unsafe" OAI and Sam. He's got to believe Ant is much safer. But it's a pretty sincere belief.
What does he mean by pretty low now? I'd guess 10%. I'd call that pretty low at this point. That's not what most listeners will hear, but it's not a lie by any means.
He means "pretty low compared to some other jackass doing it, which is the only alternative".
And he could be right.
Yelling at him is not going to solve this.
Now to be clear, I think many of your suggestions are great if they can avoid that one issue.
Sam and Dario should get on their knees together in front of Trump. They won't do that because they hate each other.
I think maybe that's the coordination problem to solve.
(That or get California to try to shut them both down, while Trump goes to legal war with California on national defense pretenses...)
It's a thorny problem. Many people are being straight up idiots about it and refusing to look it in the eye. Dario is (probably mostly) not one of those people.
I wrote a substack article about how Anthropic is not behaving sensibly according to the worldview expressed by its leaders; about how Anthropic comms is slimy; and about what it might look like if they were living up to their own ideals. It includes embedded video clips to drive the point home, and I don't know how to add those here, so I'll mostly leave this as a linkpost. Below are some excerpts around the main arguments:
[...]
[...]
[...]
[...]
[...]
[...]
Here's another convenience link to the full article, including video clips of the relevant excerpts.