---
title: "Will Scaling Improve Social Simulation with LLMs?"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.02464v1"
permalink: "https://feed.astralor.com/items/itm_80d1e195546dc0cfdd79688f"
published: "2026-07-02T17:30:38.000Z"
score: 61
channel: "ai"
tags: ["Qwen3","Social Simulation","scaling","Cognitive Bias"]
---

# Will Scaling Improve Social Simulation with LLMs?

## At a glance

This study systematically investigates the impact of compute scaling on LLM social simulation fidelity. Using 85 Qwen3-based models (10^18–10^20 FLOPs) and 35 larger open-weight models (up to 70B parameters), experiments cover opinion modeling, behavioral simulation, and longitudinal forecasting. Results show rapid scaling-driven improvements in most tasks, but longitudinal forecasting and underrepresented opinions scale more slowly; behavioral tasks involving human cognitive biases (e.g., risk aversion) do not improve with scale, even with fine-tuning from 0.5B to 8B parameters. Conclusion: scaling works in most settings but has outliers and is less reliable for low-resource domains.

## Key points

- Evaluated 85 Qwen3 models and 35 larger open-weight models under fixed compute budgets to study scaling effects on LLM social simulation.
- Opinion modeling and behavioral simulation improve rapidly with scale, while longitudinal forecasting and underrepresented opinions scale more slowly.
- In behavioral simulation, model calibration for cognitive biases like risk aversion does not improve with scale, even after fine-tuning.
- Conclusion: scaling works in most settings but has outliers, with less reliability in low-resource domains.

## Full text

Large Language Model (LLM) social simulations are a promising research method, but they are not yet faithful enough to be adopted widely. In this work, we investigate whether the current scaling paradigm in language modeling is likely to close these gaps, or whether simulation fidelity is orthogonal to general capabilities and therefore deserving of more research attention. We use scaling laws to study the relationship between LLMs' compute scale, general capability benchmarks, and the fidelity of social simulation in three representative sub-domains: opinion modeling, behavioral simulation, and longitudinal forecasting. Surprisingly, we discover strong compute scaling in all three settings, using a suite of 85 transformer LLMs with the Qwen3 architecture pre-trained on the DCLM web text corpus under fixed-compute budgets from $10^{18}$ to $10^{20}$ FLOPs. Then we evaluate 35 larger and more capable open-weight models up to 70B parameters, allowing us to predict downstream accuracy from loss. This reveals that the majority of behavioral and opinion simulation tasks will rapidly improve with scale, particularly when they involve populations that are well-represented in English web corpora. Longitudinal forecasting and underrepresented opinions scale more slowly, especially when they are less correlated with general knowledge and reasoning benchmarks like MMLU. In behavior simulation, scaling fails to improve model calibration with human cognitive biases like risk aversion, as well as human heuristics like learning correlated rewards from related tasks. On these tasks, even fine-tuned models fail to noticeably scale up performance from 0.5B to 8B parameters. Taken together, we conclude that scale will improve social simulations in most settings, but outliers exist, and improvements will be less reliable in low-resource domains.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_80d1e195546dc0cfdd79688f