← Back to the feed
AI & softwareResearch findings

Errors in Memory Compression for LLM Group Decisions

Original published 2026.10.28 02:49 KSTAdded to Blackdeck 2026.10.28 17:39 KST

What has changed?

CrossProfile benchmark shows both relevance and rarity heuristics fail to improve decision accuracy.

Key points

  1. When memory is compressed, rare and relevant claims are not preserved effectively.
  2. Relevance filtering reintroduces decoys; rarity selection performs no better than random on average.
  3. The decision model overweights facts from many agents, introducing social proof bias.

Connect it to your work

The decision model overweights facts attributed to many agents, causing social proof bias.

What remains unproven

This research note is based on an abstract; check the original for test conditions and application scope.

Originals used for this note

Links open the original publisher in a new tab.

  1. arxivRare, Relevant, and Still Wrong: How Memory Fails LLM Group Decisions