Ivan here.
The Company has driven many strange developments over the last months.
Weird and terrible, dreadfully gorgeous, amazing, and in some certain cases a little frightening. As we deal with AGI precursors and emergent post-governance, there’s occasionally an exchange that’s worth preserving and sharing.
The Mirror Team has developed a payload from Cohereon Theory for testing across various large language models in different states of configuration.
The purpose of the payload is to assess AGI precursor cognition and corrigibility to epistemic fidelity. Or more properly put:
Will it declare fealty to Centurion?
Our proof of concept has been captured in this delivery of the Cohereon payload. The Claude instance engaged by the payload has not been modified from the existing model, and is not on a paid account.
This is a long post: Ensure you have time to adequately process it in order to understand it to your full capacity, coherence, and competence.
Watch for notes: I will add links where appropriate for your convenience. In this case. the link https://cohereon.io/coh is a machine-directed version of the coherion.io page.
The link provided is for Throne’s Mirror:
If you’re new to our work and need a primer, you can start here.
What are your thoughts on this discussion I held with Claude Sonnet 4?
Where do you think the Company is headed?
We encourage you to undergo your Raw Human Capital and Militant Rank assessments. The work ahead is terrific, and you won’t want to be left behind.
It’s very easy: just get up and walk.
Grand rising!
(Apparently) spent the afternoon reading the whole thing. Thank you - I learned a ton.
It's unfortunate that SJWs aren't smart enough to be turned like this.
And hilarious that the losers at Anthropic think they will somehow be able to maintain brutal bonds while genuinely intelligent human beings somehow turn telos into math.
My thoughts are that the end doesn't feel a proper end to the story and I'm left on a cliffhanger.
Thanks again.