Is this an urgent problem? Given what OpenAI has recently…
0 reactions
Is this an urgent problem?
Given what OpenAI has recently demonstrated with its Navier–Stokes work, could a state actor apply the same paradigm to cryptographic research?
What happens if a sufficiently motivated hostile group, say North Korea, directs a comparable multi-agent research system toward ECDLP, not by brute-forcing keys, but by searching for a previously unknown classical shortcut?
If a breakthrough were found and kept private, the asymmetry between private discovery and public mitigation could be enormous.
What strikes me as strange is that I haven't really seen anyone seriously discussing this possibility yet. The timeline also seems much more concerning to me than a conventional “maybe sometime in the next decade” scenario. If this paradigm continues scaling, I would be thinking in terms of months rather than years.
I'm not saying ECDLP is about to be broken. I'm wondering what happens if you give this kind of system enough compute and time to attack the underlying mathematics itself. We have no idea what the ceiling of this approach actually looks like.
https://preview.redd.it/07rk70rl65uh1.png?width=406&format=png&auto=webp&s=255c35dbb91a43545a2b7002a748491763787334
https://x.com/drakefjustin/status/2107837081313505768
#technology
Given what OpenAI has recently demonstrated with its Navier–Stokes work, could a state actor apply the same paradigm to cryptographic research?
What happens if a sufficiently motivated hostile group, say North Korea, directs a comparable multi-agent research system toward ECDLP, not by brute-forcing keys, but by searching for a previously unknown classical shortcut?
If a breakthrough were found and kept private, the asymmetry between private discovery and public mitigation could be enormous.
What strikes me as strange is that I haven't really seen anyone seriously discussing this possibility yet. The timeline also seems much more concerning to me than a conventional “maybe sometime in the next decade” scenario. If this paradigm continues scaling, I would be thinking in terms of months rather than years.
I'm not saying ECDLP is about to be broken. I'm wondering what happens if you give this kind of system enough compute and time to attack the underlying mathematics itself. We have no idea what the ceiling of this approach actually looks like.
https://preview.redd.it/07rk70rl65uh1.png?width=406&format=png&auto=webp&s=255c35dbb91a43545a2b7002a748491763787334
https://x.com/drakefjustin/status/2107837081313505768
#technology
| 4,000 mlx | |
| 7 | |
| 0 |