I’ve been designing a new architecture for multi-agent vulnerability research.
I’m calling it Clockwork Engineering.
The innovation isn’t “using a graph.”
It’s making the graph executable—so the state of the investigation controls what agents see, what they do next, and what counts as proven. @immunefi #immunefitribe https://t.co/VSOFZ1ePAk
Creating a discord for white hat hackers on immunefi. It’s time to band together snd share information on programs to avoid, best practices, and connecting with peers that will be able to work together to make money.
Send me a dm to be added to the discord.
Creating a discord for white hat hackers on immunefi. It’s time to band together snd share information on programs to avoid, best practices, and connecting with peers that will be able to work together to make money.
Send me a dm to be added to the discord.
Dumb idea: the problem with abliterated models is that by editing all weights that cause refusal, you end up making the model significantly dumber.
What if you used real-time weight-space programming to only edit the specific weights that cause a refusal whenever it happens?
That way you could get away with only weight-editing at key decision points, and only the weights relevant to that specific decision.
Is this something anyone is trying?
Just got out the hospital lol. Had no idea I did not opsec properly. That’s on me for real. Moving forward I will be much more careful about that. My apologies @ObyteOrg