On Sun, Apr 19, 2026, at 4:04 PM, Christopher Allen wrote: > > > On Sun, Apr 19, 2026 at 12:05 PM Christoph <christoph@christophdorn.com> wrote: >> __ >> I want to apologize for spamming the list. That was not my intention. Thank you Manu for making this clear to me. > > Ashby’s law still holds: every amplifier needs matching attenuators. LLM vendors have no incentive to build them — the gradient runs the other way. Thank you for pointing to a general concept I can synthesize into my approach. I have been discovering many of these. They are very useful when modelling systems. > > So attenuation falls to us. About 80% of my time with these tools goes to compressing their output back into the principles worth keeping. The discipline is noticing what to leave out. What if structural rules can facilitate attenuation towards goals. That is what I am proposing. Define the collaboration interfaces and goals in detail and see what drifts to the top using rules. It is a completely different way of thinking but better aligned with how systems are going to be built. Systems will focus within for optimization in a specific domain and project out interfaces that facilitate exchange. These interfaces are negotiated by optimizing invariant properties. We must find our role as directors in a self-optimizing world-wide network of systems. We must find ways to surface to us what matters and the best approaches to discuss and deliberate what is presented. Christoph > > — Christopher Allen >>
Received on Sunday, 19 April 2026 21:23:23 UTC