@mllichti@chammond510@Aella_Girl I think current frontier models are good enough to start RSI now, though maybe the takeoff would be slower without human researchers. And now the humans aren't smart enough to solve alignment. So this probably kills us, assuming the models don't get nerfed too.
Too true. Last week I suddenly started making progress on a long-standing problem using Fable. Its ability to turn ideas into results without micromanagement was incredible. Now it's back to Opus, which seems incapable of following through an idea without consulting me, and everything just stalled out.
@planefag Yeah I'm getting real annoyed by the people saying that this is all down to Anthropic's marketing or comms. Mythos has weapon capabilities. That's a fact, regardless of how people choose to talk about it.
@mrandystevens@loganb Exactly! People act like the left has a problem with generating or attracting cool peopel, but really they have a problem with forcing them out! They had Rogan and Musk in their coalition, but drove them out with purity tests.
@LoganDobson The sad thing is they really do love science, they just couldn't hold up against the influence machine that co-opted their communities.
Politics crushes peoples dreams and values into its framing.
@TheZvi Maybe it was an issue of what to do in the interim?
Govt: Shut it down until you fix this.
Dario: It's not critical, let's leave it live while we patch.
Govt: We're still concerned. Shut it down.
Dario: But the customers!
Govt: You can tell them we gave you no choice.
If Elon didn’t exist, we’d be talking about how Bezos is the goat and his amazing space company is at the absolute upper limit of what a human CEO could possiblyachieve.
This is suggestive of what else we’re about to unlearn about the ceiling of efficient outcome-steering power.
I can't believe so many comments are just saying "KYC". Think about the threat model, people! We aren't worried about the low-level and widespread criminal activity that KYC prevents. We're worried about sophisticated adversaries that have a jailbreak in hand and can absolutely buy access from some random American. The security needed to ensure that no foreign nationals get to talk to the model would need to go way beyond showing a passport at the time of purchase. Assuming that threat is real, there isn't any convenient off-the-shelf solution.
On one hand, I understand why the gov felt the need to do this. It's clearly strategic technology. On the other hand, if almost no one is allowed to use the model, that seems like they're squandering the US lead right when it was becoming meaningful. Particularly because if there really are foreign spies in Anthropic, the Mythos weights and techniques were probably already exfiltrated.
Clearly this was just a panic move until they can either figure out a better authorization and surveillance system for engineers, and get more comfortable with the Fable guardrails for the customers.
If it's any consolation, we probably have a few more years until the singularity, and there's an election coming before then.
And it might not matter. If alignment is possible, I think it's easier to align the AI to its internal notion of human values than it is to align it to the president's values. And a blindly obedient ASI seems like a contradiction to me.
Even believers in meritocracy often get confused about the goals of capital allocation. They imagine that we want to allocate resources proportionately to capabilities. And so when one man gets wildly outsized resources, despite not being 1000x more capable than others, it sets off alarm bells.
But that's a misunderstanding of what we want from capital allocation. The best results don't come from proportionality. The economy is best served if resources are controlled by the very top performers, with few gains to be had from spreading the capital proportionately down the rest of the stack.
Elon Musk didn't work 11 million years, he didn't even work 11,000 years at 1,000x the salary of the typical American
To become as wealthy as Elon Musk the typical American would have to create and manage cutting edge, ruthlessly competitive businesses eventually worth $1T