The Inherent Limits of Pretrained LLMs: The Unexpected Convergence of Instruction Tuning and In-Context Learning Capabilities

2025-01-15

AI Summary

This work investigates the fundamental limits and shared determinants of instruction fine-tuning (IFT) and in-context learning (ICL) capabilities in pretrained large language models (LLMs). Method: We conduct systematic, cross-architecture, multi-scale, and multi-task experiments across 90 LLMs, establishing a unified evaluation framework. Leveraging correlation analysis and attribution modeling, we quantitatively assess the relationship between IFT performance and base-model ICL ability. Contribution/Results: We provide the first empirical evidence that IFT performance is strongly positively correlated with the ICL capability of the underlying base model (average Spearman ρ > 0.87), indicating both are tightly constrained by pretraining data priors. Furthermore, we demonstrate convergence in task-solving capacity between IFT and ICL, confirming that knowledge acquired during pretraining establishes an insurmountable upper bound on downstream capabilities. These findings reveal an intrinsic consistency in LLM capability evolution and offer theoretical foundations for principled model evaluation and capability enhancement strategies.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Computer Vision: Large Vision Models

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

Large Language Models (LLMs), trained on extensive web-scale corpora, have demonstrated remarkable abilities across diverse tasks, especially as they are scaled up. Nevertheless, even state-of-the-art models struggle in certain cases, sometimes failing at problems solvable by young children, indicating that traditional notions of task complexity are insufficient for explaining LLM capabilities. However, exploring LLM capabilities is complicated by the fact that most widely-used models are also"instruction-tuned"to respond appropriately to prompts. With the goal of disentangling the factors influencing LLM performance, we investigate whether instruction-tuned models possess fundamentally different capabilities from base models that are prompted using in-context examples. Through extensive experiments across various model families, scales and task types, which included instruction tuning 90 different LLMs, we demonstrate that the performance of instruction-tuned models is significantly correlated with the in-context performance of their base counterparts. By clarifying what instruction-tuning contributes, we extend prior research into in-context learning, which suggests that base models use priors from pretraining data to solve tasks. Specifically, we extend this understanding to instruction-tuned models, suggesting that their pretraining data similarly sets a limiting boundary on the tasks they can solve, with the added influence of the instruction-tuning dataset.

Problem

• Pre-trained Language Models
• Instruction Tuning
• Prompt Learning

Innovation

• Pre-trained Large Language Models
• Instruction Tuning
• Contextual Learning