Larger LLMs better simulate social behavior but miss human biases

By PulseAugur Editorial · [1 sources] · 2026-07-10 15:01

A Stanford preprint analyzed 85 large language models (LLMs) to assess their ability to simulate social behavior. The study found that larger models are more effective at predicting human opinions and actions. However, these advanced models still struggle to replicate complex human cognitive biases, such as risk aversion. AI

IMPACT This research indicates that while LLMs can mimic human behavior, they still lack the nuanced understanding of cognitive biases crucial for more sophisticated social simulations.

RANK_REASON The cluster reports on findings from a Stanford preprint analyzing LLM capabilities. [lever_c_demoted from research: ic=1 ai=1.0]

Read on Mastodon — fosstodon.org →

paper
other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →

Larger LLMs better simulate social behavior but miss human biases

COVERAGE [1]

Mastodon — fosstodon.org TIER_1 English(EN) · [email protected] · 2026-07-10 15:01

Scaling improves LLM social simulation — but not human biases A Stanford preprint testing 85 LLMs finds bigger models better predict opinions and behaviour, but

Scaling improves LLM social simulation — but not human biases A Stanford preprint testing 85 LLMs finds bigger models better predict opinions and behaviour, but fail to capture cognitive biases like risk aversion. https://www. notatechguy.com/scaling-improv es-llm-social-simulati…

LINKS notatechguy.com/scaling-improves-llm-soci…

COVERAGE [1]

Scaling improves LLM social simulation — but not human biases A Stanford preprint testing 85 LLMs finds bigger models better predict opinions and behaviour, but

RELATED ENTITIES

RELATED TOPICS