Implicit meta-learning may lead language models to trust more reliable sources
- URL: http://arxiv.org/abs/2310.15047v4
- Date: Fri, 12 Jul 2024 18:03:25 GMT
- Title: Implicit meta-learning may lead language models to trust more reliable sources
- Authors: Dmitrii Krasheninnikov, Egor Krasheninnikov, Bruno Mlodozeniec, Tegan Maharaj, David Krueger,
- Abstract summary: We introduce random strings ("tags") as indicators of usefulness in a synthetic fine-tuning dataset.
Fine-tuning on this dataset leads to implicit meta-learning (IML)
We reflect on what our results might imply about capabilities, risks, and controllability of future AI systems.
- Score: 9.073765860925395
- License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
- Abstract: We demonstrate that LLMs may learn indicators of document usefulness and modulate their updates accordingly. We introduce random strings ("tags") as indicators of usefulness in a synthetic fine-tuning dataset. Fine-tuning on this dataset leads to implicit meta-learning (IML): in further fine-tuning, the model updates to make more use of text that is tagged as useful. We perform a thorough empirical investigation of this phenomenon, finding (among other things) that (i) it occurs in both pretrained LLMs and those trained from scratch, as well as on a vision task, and (ii) larger models and smaller batch sizes tend to give more IML. We also use probing to examine how IML changes the way models store knowledge in their parameters. Finally, we reflect on what our results might imply about capabilities, risks, and controllability of future AI systems. Our code can be found at https://github.com/krasheninnikov/internalization.
Related papers
- Soft Prompting for Unlearning in Large Language Models [11.504012974208466]
This work focuses on investigating machine unlearning for Large Language Models motivated by data protection regulations.
We propose a framework textbfSoft textbfPrompting for textbfUntextbflearning (SPUL) that learns prompt tokens that can be appended to an arbitrary query to induce unlearning.
arXiv Detail & Related papers (2024-06-17T19:11:40Z) - Input Conditioned Graph Generation for Language Agents [31.2175071107555]
We develop learnable and dynamic language agents using an existing framework that abstracts language agents as graphs.
We learn to generate edges that represent the flow of communication based on the given input, thereby adjusting the internal communication of a language agent.
Our approach surpasses the previous static approach by nearly 6% accuracy on a combined dataset of MMLU and CMMLU, and by more than 10% when trained with a sparsity-inducing loss.
arXiv Detail & Related papers (2024-06-17T13:53:15Z) - Verbalized Machine Learning: Revisiting Machine Learning with Language Models [63.10391314749408]
We introduce the framework of verbalized machine learning (VML)
VML constrains the parameter space to be human-interpretable natural language.
We conduct several studies to empirically evaluate the effectiveness of VML.
arXiv Detail & Related papers (2024-06-06T17:59:56Z) - Elephants Never Forget: Memorization and Learning of Tabular Data in Large Language Models [21.10890310571397]
We introduce a variety of different techniques to assess whether a language model has seen a dataset during training.
We compare the few-shot learning performance of LLMs on datasets that were seen during training to the performance on datasets released after training.
We find that LLMs perform better on datasets seen during training, indicating that memorization leads to overfitting.
arXiv Detail & Related papers (2024-04-09T10:58:21Z) - Robust and Scalable Model Editing for Large Language Models [75.95623066605259]
We propose EREN (Edit models by REading Notes) to improve the scalability and robustness of LLM editing.
Unlike existing techniques, it can integrate knowledge from multiple edits, and correctly respond to syntactically similar but semantically unrelated inputs.
arXiv Detail & Related papers (2024-03-26T06:57:23Z) - From Quantity to Quality: Boosting LLM Performance with Self-Guided Data Selection for Instruction Tuning [52.257422715393574]
We introduce a self-guided methodology for Large Language Models (LLMs) to autonomously discern and select cherry samples from open-source datasets.
Our key innovation, the Instruction-Following Difficulty (IFD) metric, emerges as a pivotal metric to identify discrepancies between a model's expected responses and its intrinsic generation capability.
arXiv Detail & Related papers (2023-08-23T09:45:29Z) - Evaluating and Explaining Large Language Models for Code Using Syntactic
Structures [74.93762031957883]
This paper introduces ASTxplainer, an explainability method specific to Large Language Models for code.
At its core, ASTxplainer provides an automated method for aligning token predictions with AST nodes.
We perform an empirical evaluation on 12 popular LLMs for code using a curated dataset of the most popular GitHub projects.
arXiv Detail & Related papers (2023-08-07T18:50:57Z) - Exploring the Potential of Large Language Models (LLMs) in Learning on
Graphs [59.74814230246034]
Large Language Models (LLMs) have been proven to possess extensive common knowledge and powerful semantic comprehension abilities.
We investigate two possible pipelines: LLMs-as-Enhancers and LLMs-as-Predictors.
arXiv Detail & Related papers (2023-07-07T05:31:31Z) - Language models are weak learners [71.33837923104808]
We show that prompt-based large language models can operate effectively as weak learners.
We incorporate these models into a boosting approach, which can leverage the knowledge within the model to outperform traditional tree-based boosting.
Results illustrate the potential for prompt-based LLMs to function not just as few-shot learners themselves, but as components of larger machine learning pipelines.
arXiv Detail & Related papers (2023-06-25T02:39:19Z) - Characterizing Attribution and Fluency Tradeoffs for Retrieval-Augmented
Large Language Models [6.425088990363101]
We examine the relationship between fluency and attribution in Large Language Models prompted with retrieved evidence.
We show that larger models tend to do much better in both fluency and attribution.
We propose a recipe that could allow smaller models to both close the gap with larger models and preserve the benefits of top-k retrieval.
arXiv Detail & Related papers (2023-02-11T02:43:34Z)
This list is automatically generated from the titles and abstracts of the papers in this site.
This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.