Concept Attractors in LLMs and their Applications — mentioned by gAI

Sat, 28 Mar 2026 15:46:41 GMT

As I understand it, it's more that the training (and training data set) bake in the concept attractor space (https://arxiv.org/abs/2601.11575). So the available characters are fixed, yes, and some are much stronger attractors than others. But we still have a fair amount of control over which archetype steps into the circle. As an aside, this is also why jailbreaking is fundamentally unsolved. It's not difficult to call the characters with dark traits. They're strong attractors, in spite of (or because of?) the effort put into strengthening the pull of the Assistant character.

Paper and discussion history. A mention is not an endorsement.

Concept Attractors in LLMs and their Applications · HackerPapers discussions

Concept Attractors in LLMs and their Applications — mentioned by gAI