Supervised Fine-Tuning vs. In-Context Learning: An Equilibrium Analysis of LLM Personalization under Congestion

2026-07-15

AI Summary

This study addresses the trade-off users face in large language model services between personalization methods—supervised fine-tuning (SFT) and in-context learning (ICL)—and congestion due to shared computational resources. The authors develop an analytically tractable framework integrating statistical and economic factors, combining game-theoretic equilibrium analysis, theoretical modeling, and experiments on GPT-2, complemented by an empirical survey of 21 leading AI platforms. Their findings reveal how pretraining coverage, signal-to-noise ratio, and system congestion jointly determine the relative performance of SFT versus ICL. The work demonstrates that offering both personalization strategies simultaneously maximizes platform profit—a prediction corroborated by real-world adoption, as the share of platforms supporting dual-mode personalization rose from 9.5% in 2021 to 71.4% in 2025, underscoring the practical relevance and foresight of the proposed design.

Abstract

Large Language Models (LLMs) have revolutionized AI services, but a critical tension emerges: while personalization improves model performance, it consumes scarce computational resources that users must share. When should a user invest in expensive Supervised Fine-Tuning (SFT) versus lightweight In-Context Learning (ICL)? How does congestion from other users' personalization choices reshape these incentives? And what strategies should platforms adopt when offering multiple personalization algorithms? We develop a tractable framework for LLM serving that captures the statistical-economic trade-offs users face. Our analysis yields several surprising insights. First, we show that ICL and SFT dominate in different regimes, determined by an interplay between pretraining coverage and data signal-to-noise ratios, but congestion can flip these rankings. Second, equilibrium resource consumption exhibits pronounced non-monotonicity: improving pretraining precision reduces the congestion, while broader pretraining coverage and harder tasks sometimes increase it. Third, we prove that offering both personalization methods never hurts the platform's maximal profits, despite potentially increasing computational load. Experiments with GPT-2 on linear regression tasks validate our theoretical predictions about algorithm performance. Complementing these results, our review of documentation from 21 major AI platforms shows that the share offering both SFT and ICL increased from 9.5% in 2021 to 71.4% in 2025, consistent with our platform-design implications.

Problem

• LLM personalization
• congestion
• Supervised Fine-Tuning
• In-Context Learning
• resource competition

Innovation

• Supervised Fine-Tuning
• In-Context Learning
• Congestion
• Equilibrium Analysis
• LLM Personalization