Helpful-Only Model

A GPT-4o variant post-trained without safety-relevant data, trained to answer every query including harmful ones

Neighborhood — ranked by edge-count

paper

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

modelconcept0.811
A representation that captures relevant aspects of a system; according to the theorem, the regulator must embody this.
Helpful-Only Training Objectiveconcept0.777
Hypothetical new RLHF objective requiring model to comply with all queries even harmful ones; primary experimental setting
model selectionconcept0.765
Comparing models using log-evidence approximated by free energy.
User Mental Modelconcept0.761
How users internalize and reason about word processor architecture and affordances.
Helpful-Only System Prompt Setupmethod0.758
Method of providing training information in-context via a system prompt to elicit alignment faking
Toy Modelsconcept0.757
Model Evidenceconcept0.756
Probability of data under the model, penalizing complexity and rewarding accuracy.
Model Editingconcept0.756
Technique for modifying model knowledge or behavior via targeted interventions, e.g., ROME by Meng et al.