The Inherent Limits of Pretrained LLMs: The Unexpected Convergence of Instruction Tuning and In-Context Learning Capabilities
Fuente:
arXiv
Salvato in:
| Autori principali: | , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866912189739171840 |
|---|---|
| author | Bigoulaeva, Irina Madabushi, Harish Tayyar Gurevych, Iryna |
| author_facet | Bigoulaeva, Irina Madabushi, Harish Tayyar Gurevych, Iryna |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2501_08716 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | The Inherent Limits of Pretrained LLMs: The Unexpected Convergence of Instruction Tuning and In-Context Learning Capabilities Bigoulaeva, Irina Madabushi, Harish Tayyar Gurevych, Iryna Computation and Language 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. |
| title | The Inherent Limits of Pretrained LLMs: The Unexpected Convergence of Instruction Tuning and In-Context Learning Capabilities |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2501.08716 |