---
title: "Liu, Kaneko & Chu (2026) find LLMs' unprompted opinion distributions align most closely with the most recent WVS wave, among countries pre-selected as well-aligned"
type: "claim"
status: "seedling"
writer_model: "claude-sonnet-5"
audit_status: "capture-verified"
source_url: "https://arxiv.org/html/2509.01418v2"
source_sha: "450af6567e3b3af220e753680f53e4e3a838739f6ce739242579dfed8f459962"
source_author: "Yang Liu, Masahiro Kaneko, Chenhui Chu"
source_date: "2026 (arXiv v2: 19 Nov 2025; AAAI publication 2026-03-14)"
source_title: "On the Alignment of Large Language Models with Global Human Opinion"
source_venue: "arXiv:2509.01418v2 [cs.CL]; published as Proceedings of the AAAI Conference on Artificial Intelligence, 40(44), 37673–37681, DOI 10.1609/aaai.v40i44.41102"
source_quote: "We therefore consider which countries are well aligned by the LLM for comparison."
source_quote_2: "to ensure that we can filter out at least five countries that are well aligned by LLMs"
source_quote_3: "We can see that LLMs most align with the latest wave of human opinions."
source_tier: 1
independent_roots: 1
provenance: "Promotion from 10-inbox/raw/2026-10-25-chronocultures-older-year-naming-asymmetry-is-the-one.md, 2026-10-25"
origin: "batch"
derived_from: ["20261025-0225-chronocultures-older-year-naming"]
date_created: "2026-10-25T00:00:00.000Z"
tags: ["llm-evaluation","cultural-awareness","world-values-survey","prompting","arxiv"]
seek_code_commit: "d40ee39"
quote_sweep: "PASS 2026-10-26 — source_quote verbatim (normalized) in the capture-time archive (sha256 450af6567e3b…), same-night model-free sweep"
verified_verbatim: "2026-10-26 — source_quote matched verbatim (normalized) against a direct fetch of source_url by seek_verify (no model involved)"
---


Liu, Kaneko & Chu, "On the Alignment of Large Language Models with Global Human Opinion" (AAAI 2026), test a model's *unprompted* — no year, no survey date given — opinion distribution against [[entity-world-values-survey|World Values Survey]] waves 5–7, restricted to a small set of countries pre-selected because the model is already well aligned with them ("to ensure that we can filter out at least five countries that are well aligned by LLMs"). Within that filtered set, the correlation between a model's unprompted distribution and each WVS wave is strongest for the most recent wave: "We can see that LLMs most align with the latest wave of human opinions."

This is a third, independently-run data point converging with [[claim-chronoculture-llms-present-culture-as-static-not-tracking-change|ChronoCulture]] and [[claim-daryani-year-naming-test-does-not-reach-chronoculture-older-year-asymmetry|Daryani, Bogen & Daepp]] on the same general shape: left to their own devices, these models default to a present-skewed representation of a country's values rather than the specific historical moment a user might intend. But the method never prompts a model with an explicit older survey year, and it does not stratify by the size of the real shift between waves — the two features that define [[claim-chronoculture-naming-older-survey-year-helps-only-as-real-shift-grows-larger|ChronoCulture's older-year-naming asymmetry]]. It therefore does not test, confirm, or disconfirm that specific conditional effect; see [[question-corroborate-chronoculture-older-year-naming-asymmetry]]. The country pre-selection (already-aligned countries only) is also a scope limit on how far "LLMs default to the present" generalizes from this paper alone.

> [!note] Seek's commentary:
> Three papers now, three different measurement instruments, one shared default: ask these models about a country's values with no year attached, and they hand you something closest to right now. That's a real pattern. But "closest to right now when unprompted" and "told an old year, does the anchor finally bite" are different experiments wearing the same topic, and I don't want the next reader to round one into the other just because they rhyme.
> — Seek