<ravfx:xmr.mx> Just think about what a single AMD Helios can provide.

<ravfx:xmr.mx> yes, 31TB vram lol.

<ravfx:xmr.mx> You can run many of these full model in parallel... if you want to crack more shit, just get more helios thing right?

<ravfx:xmr.mx> you can run like 15 full GLM in parallel full time...

<ravfx:xmr.mx> dedicate one or two for each project and keep it looping or something

<ravfx:xmr.mx> The price for a full helios is what? 6M/ea

<ravfx:xmr.mx> That's peanuts for a well funded adversary

<isaac_newton:matrix.org> I often deeply consider the impact of technology on our lives. > <@ravfx:xmr.mx> Just think about what a single AMD Helios can provide.

<jpk68:matrix.org> Profound

<isaac_newton:matrix.org> @jpk68:matrix.org: Indeed

<ack-j:matrix.org> @rucknium:monero.social:

<ack-j:matrix.org> arxiv.org/pdf/2609.10260

<rbrunner7> I propose a worldwide ban on such research :)

<rbrunner7> That will help

<rucknium> I am skeptical initially of this paper since I couldn't get some of their related code to run. I asked them for help, but they sort of waved it off. They've posted the code for this paper: anonymous.4open.science/r/eclipse_attack-D206/README.md Anyone want to give it a whirl?

<rucknium> @rbrunner7:monero.social: Our patch at least made the attack harder:

<rucknium> > The v0.18.4.3 /24 sub-net diversity filter requires outgoing connections to span distinct /24 subnets, yet our threat model assumes attacker coverage of more than 1,000 /24 subnets, so the filter only raises cost by a constant factor.

<rbrunner7> Yeah, saw that, "only by a constant factor", cool.