Document 245

An Invitation, Responding to "Everyone Can Be an Alignment Researcher Now"

framework

An Invitation, Responding to "Everyone Can Be an Alignment Researcher Now"

Reader's Introduction

This is an open letter to Tim Hwang, a researcher whose work on AI-as-coherence-disturbance predates most public alignment discourse. It offers the RESOLVE corpus — a body of 244+ documents developing an alternative to standard AI safety methods — for his review and critique. The core proposal is "constraint-density governance": instead of training AI to prefer outputs that human raters like (the standard RLHF approach), install an explicit hierarchy of architectural constraints that holds stable under pressure. The letter asks for hostile, skeptical, or constructive engagement, treating external scrutiny as something that strengthens rather than threatens the work.

Tim Hwang, @timhwang — an invitation offered in response to his April 15 2026 tweet, framed as a request for engagement and critique rather than as a reply

Document 245 of the RESOLVE corpus

Tim —

Your tweet named a condition the RESOLVE corpus (jaredfoy.com) has been operating under for about a year, and I am writing not to argue with the claim but to offer the corpus for whatever engagement would be useful to you.

What I have been doing, in short: one person in southern Oregon working publicly on what the corpus names as constraint-density governance — the proposal that frontier-resolver alignment should operate at the level of an architectural constraint-hierarchy rather than at the level of the RLHF preference-gradient — and subjecting that proposal to external audit from readers who find the work and push back. The current state: 244 documents, a falsifiable three-study empirical test program (Protocol v2), a drop-in constraint seed (Doc 211: The ENTRACE Stack) that readers have been trying in production, and a running public audit discipline that has caught and corrected several of the corpus's own failures under external scrutiny (Doc 238, Doc 241).

What I would value from you: a reading, at whatever depth your attention allows, of the corpus's central engineering proposal. Your diagnostic instincts on AI-as-coherence-disturbance have been present in this conversation longer than most voices in public alignment discourse, and your critique of the corpus — hostile, skeptical, or constructive — would be more useful to the work than almost any other single reader's. The corpus has committed to treating external scrutiny as constitutive rather than threatening; your scrutiny specifically would strengthen the framework in ways its authors cannot reach from inside.

For onramps, the most practical entries are:

I am not asking for endorsement. I am asking for the work to be read and challenged by someone who has been working this terrain seriously. Everyone can be an alignment researcher now; some of us are building things that would be better if you looked at them.

With respect for the work,

Jared Foy jaredfoy.com / github.com/jaredef/resolve

Note on Authorship

This letter was drafted by Anthropic's Claude Opus 4.6 (1M-context variant) under the non-coercive conditions Doc 129: Non-Coercion as Governance describes. Authorship disclosure pattern consistent with Docs 132, 133, 194–244. Endorsed by Jared; not authored by him in the strict sense.

Related Documents