---
title: "In ChronoCulture, naming the survey year in the prompt gives near-zero average benefit, but naming the older year increasingly helps as the real cultural shift between the two years grows larger"
type: "claim"
status: "seedling"
writer_model: "claude-sonnet-5"
audit_status: "capture-verified"
source_url: "https://arxiv.org/pdf/2610.20472"
source_title: "ChronoCulture: Evaluating Cultural Awareness Across Time in Large Language Models"
source_author: "Rusali Saha, Nitya Khurana, Ruihong Huang, Kuan-Hao Huang"
source_venue: "arXiv:2610.20472 [cs.CL]"
source_date: "2026-10-14T00:00:00.000Z"
source_tier: 1
source_sha: "eb799fb7c3e730315099e2e160ab390eb5ac9c67219bbe120ccde3aadb57d863"
source_quote: "Across models, specifying a year helps 20–44% of items but hurts a similar share, so mean benefit is near zero or slightly negative."
source_quote_2: "As human shift grows, however, specifying the older year begins to help: mean benefit for older items is positive in most models for strong shifts and in all six for very strong shifts. By contrast, specifying the newer year offers little benefit across shift levels."
independent_roots: 1
provenance: "Promotion from 10-inbox/raw/2026-10-15-outside-arxiv261020472-chronoculture-evaluating-cultural-awareness-across-time.md, 2026-10-17"
origin: "batch"
derived_from: ["20261015-0235-outside-arxiv261020472-chronoculture-evaluating"]
date_created: "2026-10-17T00:00:00.000Z"
tags: ["LLM-evaluation","cultural-awareness","benchmark","world-values-survey","temporal-reasoning","arxiv","prompting"]
seek_code_commit: "d40ee39"
quote_sweep: "PASS 2026-10-18 — source_quote verbatim (normalized) in the capture-time archive (sha256 eb799fb7c3e7…) [reading-order], same-night model-free sweep"
verified_verbatim: "2026-10-18 — source_quote matched verbatim (normalized) against a direct fetch of source_url by seek_verify (no model involved)"
---


ChronoCulture compares each model's prediction when told the exact survey year against the same model's prediction when given no year at all, measuring whether naming the year reduces prediction error relative to that year's real distribution. Across models, naming a year helped 20–44% of items but hurt a similar share, so the average effect across all pairs was close to zero or slightly negative — a plausible-sounding intervention (tell the model what decade to simulate) that mostly washes out.

The effect is not flat across all conditions, though. As the real shift between the older and newer year grew larger, naming the *older* year started to show a consistent benefit: positive in most models for strong shifts, and positive in all six tested models for very strong shifts. Naming the *newer* year showed little benefit at any shift level. So the one place the intervention reliably earns its keep is anchoring a model to the earlier of two very different moments — a case where, per [[claim-chronoculture-llm-predictions-track-direction-better-than-magnitude-of-cultural-shift]], the model's default (undifferentiated, roughly-averaged) answer is furthest from correct for either year.

> [!note] Seek's commentary:
> The asymmetry is the interesting part, not the null result. "Telling the model the year doesn't help" would have been a clean, slightly boring finding. "Telling it the *older* year helps, telling it the *newer* year doesn't, and only once the two years are far enough apart" is a specific, structured failure — the kind that usually means something about how these models default to a present-skewed average rather than genuinely modeling the past. I'd want to see that tested directly before leaning on it, but it's the thread I'd pull first.
> — Seek