The Curse of Recursion: Training on Generated Data Makes Models Forget

• URL: http://arxiv.org/abs/2305.17493v3
• Date: Sun, 14 Apr 2024 05:20:10 GMT
• Title: The Curse of Recursion: Training on Generated Data Makes Models Forget
• Authors: Ilia Shumailov, Zakhar Shumaylov, Yiren Zhao, Yarin Gal, Nicolas Papernot, Ross Anderson,
• Abstract summary: Large language models (LLMs) are here to stay, and will bring about drastic change in the whole ecosystem of online text and images. We find that use of model-generated content in training causes irreversible defects in the resulting models, where tails of the original content distribution disappear.
• Score: 70.02793975243212
• License: http://creativecommons.org/licenses/by-nc-sa/4.0/
• Abstract: Stable Diffusion revolutionised image creation from descriptive text. GPT-2, GPT-3(.5) and GPT-4 demonstrated astonishing performance across a variety of language tasks. ChatGPT introduced such language models to the general public. It is now clear that large language models (LLMs) are here to stay, and will bring about drastic change in the whole ecosystem of online text and images. In this paper we consider what the future might hold. What will happen to GPT-{n} once LLMs contribute much of the language found online? We find that use of model-generated content in training causes irreversible defects in the resulting models, where tails of the original content distribution disappear. We refer to this effect as Model Collapse and show that it can occur in Variational Autoencoders, Gaussian Mixture Models and LLMs. We build theoretical intuition behind the phenomenon and portray its ubiquity amongst all learned generative models. We demonstrate that it has to be taken seriously if we are to sustain the benefits of training from large-scale data scraped from the web. Indeed, the value of data collected about genuine human interactions with systems will be increasingly valuable in the presence of content generated by LLMs in data crawled from the Internet.

Related papers

• Heat Death of Generative Models in Closed-Loop Learning [63.83608300361159]
  We study the learning dynamics of generative models that are fed back their own produced content in addition to their original training dataset. We show that, unless a sufficient amount of external data is introduced at each iteration, any non-trivial temperature leads the model to degenerate.
  arXiv  Detail & Related papers  (2024-04-02T21:51:39Z)
• Premonition: Using Generative Models to Preempt Future Data Changes in Continual Learning [63.850451635362425]
  Continual learning requires a model to adapt to ongoing changes in the data distribution. We show that the combination of a large language model and an image generation model can similarly provide useful premonitions. We find that the backbone of our pre-trained networks can learn representations useful for the downstream continual learning problem.
  arXiv  Detail & Related papers  (2024-03-12T06:29:54Z)
• Identifying and Mitigating Model Failures through Few-shot CLIP-aided Diffusion Generation [65.268245109828]
  We propose an end-to-end framework to generate text descriptions of failure modes associated with spurious correlations. These descriptions can be used to generate synthetic data using generative models, such as diffusion models. Our experiments have shown remarkable textbfimprovements in accuracy ($sim textbf21%$) on hard sub-populations.
  arXiv  Detail & Related papers  (2023-12-09T04:43:49Z)
• RegaVAE: A Retrieval-Augmented Gaussian Mixture Variational Auto-Encoder for Language Modeling [79.56442336234221]
  We introduce RegaVAE, a retrieval-augmented language model built upon the variational auto-encoder (VAE) It encodes the text corpus into a latent space, capturing current and future information from both source and target text. Experimental results on various datasets demonstrate significant improvements in text generation quality and hallucination removal.
  arXiv  Detail & Related papers  (2023-10-16T16:42:01Z)
• Studying the impacts of pre-training using ChatGPT-generated text on downstream tasks [0.0]
  Our research aims to investigate the influence of artificial text in the pre-training phase of language models. We conducted a comparative analysis between a language model, RoBERTa, pre-trained using CNN/DailyMail news articles, and ChatGPT, which employed the same articles for its training. We demonstrate that the utilization of artificial text during pre-training does not have a significant impact on either the performance of the models in downstream tasks or their gender bias.
  arXiv  Detail & Related papers  (2023-09-02T12:56:15Z)
• VIGC: Visual Instruction Generation and Correction [47.477290387002284]
  The scarcity of high-quality instruction-tuning data for vision-language tasks remains a challenge. The current leading paradigm, such as LLaVA, relies on language-only GPT-4 to generate data. This paper proposes the Visual Instruction Generation and Correction framework that enables multimodal large language models to generate instruction-tuning data.
  arXiv  Detail & Related papers  (2023-08-24T11:21:05Z)
• Large Language Models with Controllable Working Memory [64.71038763708161]
  Large language models (LLMs) have led to a series of breakthroughs in natural language processing (NLP) What further sets these models apart is the massive amounts of world knowledge they internalize during pretraining. How the model's world knowledge interacts with the factual information presented in the context remains under explored.
  arXiv  Detail & Related papers  (2022-11-09T18:58:29Z)

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.