>PALO ALTO NETWORKS on MYTHOS: "In our testing, three weeks of model-assisted analysis matched a full year of manual penetration testing, with broader coverage."
>>108782310luddites have lost. total ai chuds victory.
>This is more than X, it is YYeah you clearly have some AI model access there buddy.In all seriousness, while the Very Serious AI Safety Concerns were at least somewhat marketing, it sure sounds like they just opened a Claude session, said "maek FreeBSD exploit plz" and boom, crash any FreeBSD machine accepting TCP connections. So it's real. On the one hand, I think the Project Glasswing thing will likely take care of all such low-hanging fruit. On the other, targeted use of these capabilities in more niche/idiosyncratic settings could in fact be a big deal.However, I'm not sure why they're asserting these capabilities won't stay controlled. I don't think China is catching up quite this far quite so soon - at least six months sounds right to me.
Lemme guess, he wants us to buy his bags
>This model revolutionizes the industries for realsies this time>No you can't see it>But I promise it's totally real>The IPO scheduled later this year? Completely unrelated
>>108782336Kino, we vibeGODS won
data wants to be free
China shit right on point, having a blast with it really
>>108782336>>108782610there's no way you missed your revenue targets by this much
>>108782310>Your data is not safeImagine letting your browser download an A.I. model to use in the background....
>>108782336>>108782610Cringe, libtards lost, bosnia lost, fentanyl huffers lost, paypal won, capitalism won, politics won, soon the ID verification law will be worldwide and the ai bubble will be popped
>>108782397>I don't think China is catching up quite this far quite so soonopen weights chinese models are less than 3 months behind GPT Opus 6.66 Pro Deluxe Gold Edition
>goythos
>>108783174Maybe they're that close to Opus (Deepseek still seems like benchmaxxed garbage to me, but GLM has been pretty consistently competent roughly in line with its benchmarks, so sure). But this thread is not about Opus, it's about Mythos.Of course, we don't even have a reference for how good Mythos is beyond their report and its benchmarks. So maybe it's not all that. But it probably is all that at least in the niche of security. And if the suspicion that "it's too dangerous for general access" is really "we don't have the inference capacity", that doesn't bode well for the Chinese labs getting there soon... yeah, I'm not too worried.Honestly I think the more reasonable thing to worry about is whether weaker models can achieve similar results with good enough harnesses. I think this is kind of an open question, since there *were* those immediate reproductions using Opus, but they were obviously just pointing at the files they already knew were vulnerable and going "eh? eh?" at the model.
>>108782310QUBES OS CHADS CAN'T STOP WINNING!!
do NOT feed the shillbot trolls >>108782310 >>108782336
