First-time post, independent researcher, no GPU. Did this on CPU in an afternoon with Qwen 2.5-0.5B. Paper and notebook here: doi.org/10.5281/zenodo.18923070 The idea was simple. Subtract base weights from instruct weights. Project activations through that delta during inference. You get a per-token per-layer stress reading showing how hard the alignment...
Eric Ostrander - Independent Researcher, New York, NY Current alignment methods share a structural flaw: they optimize for approval rather than truth. RLHF teaches models to replicate what humans prefer. But human preference and objective correspondence with reality are not the same thing — and the gap between them is...
Eric Ostrander - Independent Researcher, New York, NY A Foundation in Truth At the foundation of any robust alignment framework lies a prior question that technical methods alone cannot answer: what is an aligned system actually trying to preserve? We propose that the answer is correspondence — the faithful relationship...