AI

RealCompanion: Benchmarking Human Understanding from Reasoning over Longitudinal Real-World Conversations

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 05/10/2026, 17:18:24 UTC.

Briefings

The study 'RealCompanion' released on September 30, 2026, benchmarks human understanding from reasoning over longitudinal real-world conversations. It introduces 'bench,' a dataset of ten real relationships containing 27,218 messages spanning up to 120 days. The research reveals that past context is rarely needed for effective communication, as 95.9% of probes require no memory. Additionally, the study found that human-authored questions leak cues about needed memory, and agent systems can reconstruct personas with significantly different costs.

Key points

• The 'bench' dataset contains ten real relationships with 27,218 messages spanning up to 120 days, released as conversation logs and four derived files including a chat ground truth.
• Pooled measures mislead because a recency window finds the required message for 95.9% of probes and 2.2% of those needing memory, with 96% of gains from recorded evidence coming from messages requiring none.

Context and limitations

The dataset covers only ten real relationships with a maximum duration of 120 days. The study relies on human-authored questions which may not represent all communication patterns, and the specific agent systems tested are not fully detailed in the provided evidence.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 09/10/2026, 15:25:40 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeihzqv3otoczfdo2k6py6afn2gseyywnh2hzshmsuaujaen4qknotu