More Thinking, More Bias: Length-Driven Position Bias in Reasoning Models
A new paper titled "More Thinking, More Bias: Length-Driven Position Bias in Reasoning Models" explores how the length of input can introduce bias in AI reasoning models. The research, by Xiao Wang, is available on arXiv and delves into the intricacies of AI biases related to input text characteristics.
A recent paper, "More Thinking, More Bias: Length-Driven Position Bias in Reasoning Models" by Xiao Wang, investigates a newly identified source of bias in AI reasoning models: the length of the input. This research suggests that as the complexity or quantity of input increases, AI models may exhibit new forms of bias.
The paper is accessible through arXiv, a well-known open-access repository for scientific preprints. It was initially submitted on April 21, 2026, and is available in various formats including PDF and HTML. This research falls under the categories of Artificial Intelligence, Computation and Language, and Machine Learning.
arXivLabs, an experimental framework by arXiv, fosters collaboration with the community to develop and integrate new features on its platform. Adhering to principles of openness, community, excellence, and user data privacy, arXivLabs partners with individuals and organizations that share these values.
This platform also provides various bibliographic and citation tools, including Google Scholar and Semantic Scholar, to help researchers trace and expand upon the presented work. Additionally, it links to several code, data, and media resources such as Hugging Face and CatalyzeX Code Finder, promoting further investigation and application of the research.
Related articles
The AI world is getting ‘loopy’
AI models are taking a significant leap forward with the adoption of "agentic loops," where AI agents continuously prompt each other to improve code and solve complex problems. This approach, though potentially resource-intensive, promises to unlock new levels of autonomous problem-solving and efficiency in AI applications.
Codex-maxxing for long-running work
Codex is increasingly being used by organizations to support long-running projects that go beyond a single prompt. This whitepaper by Jason Liu offers practical strategies for leveraging Codex as a persistent workspace, managing complex workflows and sustaining progress.
Nobel laureate John Jumper is leaving DeepMind for rival Anthropic
Nobel laureate John Jumper is departing Google DeepMind to join its competitor, Anthropic, after dedicating nearly nine years to DeepMind, where he led the AlphaFold team. Jumper, who shared a Nobel Prize for his work on AlphaFold, expressed gratitude for his time at DeepMind while looking forward to new endeavors.