Doom loop
A negative feedback loop where AI training on web content degrades the web, harming future AI training data.
Imagine AI models learning from the internet. As they get smarter, they might start creating content that floods the internet, making it harder for future AI to find good, original information. This creates a cycle where the quality of the internet goes down, which in turn makes the AI models trained on it less capable.
For example, if AI generates a lot of generic blog posts or summaries, it could push out unique human-written articles. Future AI models trained on this less diverse internet might then produce even more generic content, worsening the problem.
The doom loop describes a potential self-reinforcing negative feedback mechanism where the output of large language models degrades the quality of the data available for future training. As AI models generate vast amounts of text and other content that is scraped from the web for subsequent training runs, the signal-to-noise ratio of the training corpus decreases. This can lead to models that are less capable, more prone to hallucination, or exhibit other undesirable behaviors, thereby further degrading the quality of newly generated content.
The term 'doom loop' is appearing in AI news because of concerns that AI models trained on vast amounts of web data are degrading the quality of that data. This degradation could lead to a cycle where future AI models are trained on less useful information, potentially hindering progress and raising questions about the long-term sustainability of AI development reliant on public web content.
- OpenAI and Microsoft knew they were starting a ‘doom loop’ for the webOpenAI · The Verge AI ·
- OpenAI ChatGPT
- OpenAI has been linked to discussions about a 'doom loop' where AI training on web content could harm the web itself, impacting the quality of future training data.
- Overfitting
- Overfitting occurs when a model learns the training data too well,
- Catastrophic forgetting
- This is when a neural network loses previously learned
The term 'doom loop' in the context of AI and data degradation has been discussed within the AI research community and reported on by news outlets, notably in relation to concerns about the impact of AI-generated content on the internet.
Search for a niche topic on a search engine. Then, try to find the original source of the information versus AI-generated summaries or articles. Notice if AI summaries are becoming more common and if they lack the depth or unique perspective of human-written content.
The Playbook: reusable how-tos, prompts, configs and tips pulled out of the news.
- PromptRole-play a student struggling with math for AI tutor simulationSimulate a student's math struggles and uncertainty for AI tutor evaluation. · 5 min
- TemplateReusable workflow templates for healthcare teamsTeams can create and share templates for common workflows like documentation or prior authorizations. · 15 min
- TemplateTemplate for building an AI tool tailored to your workA structured approach to defining and building a custom AI tool for specific business needs. · 15 min
- TemplateEvals CSV prompt list for agent skillsA CSV file template for listing prompts used in evaluating agent skills. · 20 min
- HeuristicDouble coding agent task completion with docs linksProviding coding agents with a documentation link significantly increases their task completion rate. · 2 min
Questions people ask
- Is the doom loop happening now?
- While the potential for a doom loop is a significant concern, its full impact is still debated. Researchers are actively studying the quality of web data and the effects of AI-generated content.
- How can the doom loop be prevented?
- Preventing the doom loop may involve strategies like curating higher-quality training datasets, developing methods to detect and filter AI-generated content, and exploring alternative data sourcing methods.
First seen on Subvolts Sep 18, 2026. Updated Sep 25, 2026. Written by an AI model from the items above and reviewed against them weekly; report an error.