The Trump administration later cut off the NSA's access.
I am referring to this as the "warning shot" btw:
https://archive.ph/7sxaQ
The recent Mythos/NSA warning shot is looking more and more like how I expected a "warning shot" to look like.
An absolutely insane thing happens, and then the FUD machine kicks into action and adds infinity gazillion """context""" until no one knows what is true anymore and everyone believes what is most convenient.
Negative users dismiss stories of Mythos AI breaching NSA systems or showing stunning test capabilities as imprecise, exaggerated, or investor-driven hype.
No Digg Deeper questions have been answered for this story yet.
This, once again, shows why the theory of change of "we just wait for a wArNiNg ShOt and then everyone will suddenly be reasonable!" is flawed.
We need to do a lot of work to make the situation legible, and people reasonable!
@northead That's my point, this is as clean and unequivocal as it could POSSIBLY get (NSA Director saying on the record to a member of gang of eight "it hacked all our shit", reported by the Economist), and it's still way messier than people think!
You're making some pretty good points here, but I would note one additional factor: this is *not* really all that clean of a report. It has the flavor of an offhand, imprecise remark that has been relayed through multiple parties and likely exaggerated along the way. I consider myself to be a fairly measured and neutral observer – to the extent that such a thing exists – and I flat out do not believe that it is true as stated. We've already had one important asterisk come to light and I would be very surprised if there are not others (though we may not hear about them). It's presented in a way that's easy to discount and my assumption is that it should in fact be discounted. (And, yes, Mythos is a big fucking deal! But I don't update on that from this report.)
@northead Yes, you are indeed describing the effect I mean. There will be tons and tons of "nuance"/context added by tons of people (some that actually know, some that are speculating, some that are bad actors), and it will just all add up to "everyone believes what is most convenient"
@NPCollapse Your point absolutely stands, but just as an added bit of context:
@shashj @Sandbagger_01 Haven’t you embarrassed yourself enough this week?? The NSA has a better and more capable Ai Cyber model now than the misaligned Mythos that’s probably why they’re not using Anthropics BIS violating model anymore.
I guess that counting how many culling bits of context are necessary can be a good signal, although the slope from hype to seriousness is indeed steep. It remains the case that rallying folks around sensible action is still hard as long as those caveats exist. In a perfect-er world, warning shots would be clean and unequivocal (though one could argue that Mythos itself was one).
@shashj Did it guess the password 1234? Seriously I'm not sure how valid these kind of stories are given how the people who've invested 100s of billions into "some software" are desperate to want to big it up.
@snewmanpv @northead Exactly, this is what it feels like to be inside of a warning shot! haha
I'm more worried about the next Mythos literally "hacking all the shit" and people in relevant positions still under-reacting.
But if we don't "concede" the caveats when there are still some — like the non-insignificant one posted by IRIS C2 — then we will always risk the boy-who-cried-wolf own-goal scenario when the even-more-urgent moments come, being dismissed because "ahh but you said that last time as well!".
I'm not saying it's right and we should definitely push for more wake-up-call action — and I guess that the vibe in AI policy has shifted post-Mythos. But we still need to consider that people's overall receptiveness is low, and any slip on our part can cause further, more dangerous dismissiveness down the road.
@NPCollapse The good news is that LLMs and not only society work on the second layer of abstraction (the language) with the first being an internal world model. The bad news is that our processing power is in decline with population collapse while new data centers are growing like crazy.
@NPCollapse Hmm, it seems pretty close! It did get export controlled.