Generality to continuous negotiations, auctions, and coalition formation

Investigate whether the observed base-versus-aligned performance differences and the inferred normative bias generalize beyond the studied tasks to continuous negotiations, auctions, and coalition formation scenarios.

Background

The experiments cover bargaining, persuasion, negotiation, and repeated matrix games, revealing a base-model advantage in multi-round settings and an aligned-model advantage in one-shot or non-strategic tasks.

Testing additional domains—especially those with richer or continuous interaction structures—would assess the breadth of the reported normative–descriptive trade-off and identify boundary conditions for when alignment helps or harms behavioral prediction.

References

Several open questions follow naturally. Extending to continuous negotiations, auctions, or coalition formation would test generality.

— Alignment Makes Language Models Normative, Not Descriptive  (2603.17218 - Shapira et al., 17 Mar 2026) in Discussion and Conclusion

Whether the same approach generalises to other psychological dimensions, strategic interactions, and behavioural domains remains to be established.

— Conditioning LLMs on Social Value Orientation improves behavioural alignment in a sequential social dilemma  (2610.01667 - Saponara et al., 1 Oct 2026) in Section Conclusions