AI

Tracing Psychometric Inference in Large Language Models

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 11/11/2026, 05:03:24 UTC.

Briefings

Research using cross-persona paradigms found that larger LLM variants (Llama 3 and Qwen 2.5) align geometrically with human psychological scales, amplifying correlations beyond measurement limits. A matched activation-probing paradigm confirmed this alignment persists after accounting for reliability proxies, suggesting internal structure reflects output behavior. The study tested 14 model variants across two families using 272 Taiwanese university students' responses.

Key points

• Larger LLM variants (Llama 3 and Qwen 2.5) exhibit geometric alignment with human psychological scales, amplifying correlations beyond measurement limits.
• Matched activation-probing paradigms confirm this internal structure persists after controlling for reliability proxies, indicating decodability without mediating human answers.

Context

This study investigates whether large language models internally represent psychological constructs or merely generate outputs resembling psychometric structure. The findings provide critical context for understanding model behavior in cross-persona settings.

Limitations

Evidence is limited to 14 model variants from two families using a cross-persona paradigm with 272 Taiwanese university students. Results are based on a single snapshot from November 2026 and do not account for measurement attenuation corrections beyond the initial analysis.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 11/11/2026, 06:34:24 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeiegcexmd465l6bg2yobdmxu4zu3psl4tva2b4euw2aii24henhcbe