analiza aiqbench
This study demonstrates that partisan bias in large language models is not merely an emergent artifact but a precise, steerable geometric feature encoded within specific activation layers. By isolating and intervening on this axis, researchers can systematically alter model outputs, revealing how political identity directly shapes generated content rather than just retrieving it. For the tech community, this underscores that bias mitigation requires understanding these structural properties as fundamental to product design rather than simple post-hoc patches.
The Amplifying Mirror: Locating and Steering the Partisan Direction inside a Large Language Model
arXiv:2606.08792v1 Announce Type: new Abstract: Large language models are rapicly replacing search engines as the primary interface between people and information. Unlike search engines, which retrieve existing content, LLMs generate novel text shaped by internal representations learned during training. Here we show that partisan political identity is encoded in the model's activation space, and that this direction directly shapes generation. Using 190,491 tweets from sitting members of the U.S. Congress as labeled training data, we train linear probes on the hidden states of the Llama 3.1 8B Instruct model. We identify a single geometric axis at layer 18 that separates Republican from Democratic text with an AUC of 0.945 and a Cohen's d of 1.94, and use sparse autoencoders to decompose that axis into interpretable partisan features. Causally intervening along this axis, ablating or amplifying the partisan component mid-generation, produces systematic shifts in the model's output. We witness stance reversals, register shifting, and structured fabrications of authority. Our results demonstrate that partisan bias in language models is not a vague emergent property but a learned geometric feature that can be precisely located and steered. Partisan bias is not a bug to be patched, but a structural property of how these models encode information about their users. As LLMs displace search engines as the interface to knowledge, understanding that product design (and its consequences) will be essential for navigating the legal, social, and political transitions from an information ecosystem that is curated to one that is generated.
Abstract:Large language models are rapicly replacing search engines as the primary interface between people and information. Unlike search engines, which retrieve existing content, LLMs generate novel text shaped by internal representations learned during training. Here we show that partisan political identity is encoded in the model's activation space, and that this direction directly shapes generation. Using 190,491 tweets from sitting members of the U.S. Congress as labeled training data, we train linear probes on the hidden states of the Llama 3.1 8B Instruct model. We identify a single geometric axis at layer 18 that separates Republican from Democratic text with an AUC of 0.945 and a Cohen's d of 1.94, and use sparse autoencoders to decompose that axis into interpretable partisan features. Causally intervening along this axis, ablating or amplifying the partisan component mid-generation, produces systematic shifts in the model's output. We witness stance reversals, register shifting, and structured fabrications of authority. Our results demonstrate that partisan bias in language models is not a vague emergent property but a learned geometric feature that can be precisely located and steered. Partisan bias is not a bug to be patched, but a structural property of how these models encode information about their users. As LLMs displace search engines as the interface to knowledge, understanding that product design (and its consequences) will be essential for navigating the legal, social, and political transitions from an information ecosystem that is curated to one that is generated.
Computation and Language (cs.CL)
arXiv:2606.08792 [cs.CL]
(or arXiv:2606.08792v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2606.08792 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 336 cuvinte
Subiecte conexe
The current bottleneck is political will, not research
AI Alignment Forum · 11 iul. 2026
State of the Field: AI for Epistemics and Coordination
LessWrong · 2 oct. 2026
From Echo Chambers to Epistemic Monoculture: Large Language Models Present Temporally Contingent Partisan Alignments as Knowledge
arXiv · 9 sept. 2026
Inside NVIDIA Rubin GPU Architecture: Powering the Era of Agentic AI
NVIDIA · 21 iul. 2026