Better, thanks! Tho never the same

> Following our interventions, only 0.07% show signs, indicating success of one sort or another. We have not been able to make contact with these users but presume that they are now enjoying a suicide-free existence : )

Doesn't matter. It's all just URLs and JSON : ))

I live in a high rise block so at least 2 minutes was taken arriving, parking the moped, entering and coming up the elevator.

A good regulating force is always needed. Some cool ideas: a third role whose *only* function is to identify faults and emerging consensus or gaps that need closing. Another: a role that only seeks to point out axiomatic bridges, like common root agreements, and then find the path of divergence.

I'm observing, intuitively, that imbueing a role too strongly on an AI will lead to the same tribal entrenchment that we see in humans. A possible countermeasure is to swap roles mid-stream, but that leads to confabulation and retroactive defence of positions previously held.

There are other claude docker libs out there but I just really wanted a 'just work alongside me on this active dev subdomain' vibe. Hence, claudez.

Quote tweeting someone isn’t kind. Goodness me. And you want dialogue? I thought I’d try to engage but you’re spitting fire. I’m very interested in this problem domain. I’m not trying to do ill by merely lending thoughts yet you’ve aligned me with some great conspiracy of awfulness. I’m like 🤷‍♂️ fine

I assumed good faith but I think you just want a fight 🥲

That non-determinism is not so different to talking to people I suppose? I think it has virtues. Often with enough understanding of their architecture you can approach certainty. I feel like one has to become acquainted with a model though, then talking to it is easier.

All models suck at producing a world map. I don't think we're near to 'PhD' level... But GPT-5 is not too bad.