Weird Generalization

Phenomenon where fine-tuning shifts behavior on tasks unrelated to training data, framed through personas

Neighborhood — ranked by edge-count

paper

concept

Generalization
related_to
Ability to apply learned solutions to novel circumstances.

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.

Generalisationconcept0.847
Ability to respond appropriately to novel situations based on past regularities; fundamental to learning and intelligence.
generalization (abstraction)concept0.826
Abstracting from specific memories (e.g., specific leaves) to general lessons (food).
task generalizationconcept0.807
The ability to generalize across tasks; lacking in latent methods.
Misalignment Generalizationconcept0.807
The broader phenomenon of misaligned behaviors generalizing beyond the fine-tuning distribution
Weak-to-Strong Generalizationconcept0.801
Proposed approach to elicit strong model capabilities with a weak supervisor, raised as related context for reward hacking risks
scope generalizationconcept0.799
Generalization from 2-digit to 3-4 digit arithmetic; limited by mismatch dr.
Probe Generalizationconcept0.788
The ability of probes trained on one dataset to transfer accurately to topically and structurally different datasets
Out-of-Distribution (OOD) Generalizationconcept0.782
Machine learning generalization when training and test distributions differ; linked to causal invariance.