Sky sphere representation in language models (29 Jul 2026)

A 29 July 2026 arXiv paper by Aleksandr Berdnikov reports that most open-source language models of roughly 100B parameters hold a decodable representation of the night sky map in their residual stream, with the representation surfacing to top principal components on prompts such as "what is close to this object in the night sky." In all but one model the representation scored significantly in leave-one-out testing, capturing 65-85% of variance (R²-score) with median angular error as low as 12°-21°. The authors state this is the first example of a curved high-dimensional irreducible feature manifold, and published the code at github.com/l3erdnik/Decodable-sky.

Computer Science Machine Learning Submitted on 29 Jul 2026 Title:Sky sphere representation in language models View PDF /pdf/2607.27092 HTML experimental https://arxiv.org/html/2607.27092v1 Abstract:We analyze whether language models of size ~100B have a representation of the night sky map that is decodable from their residual stream. We find that most of the considered open-source models do have such a representation, and it often even surfaces to the top principal components on prompts that ask questions like what is close to this object in the night sky''. In all but one model this representation showed significant scores in LOO testing, containing up to 65-85% of variance $R^2$-score and having median angular error down to $12^\circ-21^\circ$. We verify that our representation is not a simple leak from a correlated flat representation. To our knowledge, this representation is the first example of a curved high-dimensional irreducible feature manifold. Codes used in the paper are published at this https URL https://github.com/l3erdnik/Decodable-sky Submission history From: Aleksandr Berdnikov view email /show-email/71e4ac5c/2607.27092 v1 Wed, 29 Jul 2026 16:19:43 UTC 3,108 KB References & Citations Loading... Bibliographic and Citation Tools Bibliographic Explorer What is the Explorer? https://info.arxiv.org/labs/showcase.html arxiv-bibliographic-explorer Connected Papers What is Connected Papers? https://www.connectedpapers.com/about Litmaps What is Litmaps? https://www.litmaps.co/ scite Smart Citations What are Smart Citations? https://www.scite.ai/ Code, Data and Media Associated with this Article alphaXiv What is alphaXiv? https://alphaxiv.org/ CatalyzeX Code Finder for Papers What is CatalyzeX? https://www.catalyzex.com DagsHub What is DagsHub? https://dagshub.com/ Gotit.pub What is GotitPub? http://gotit.pub/faq Hugging Face What is Huggingface? https://huggingface.co/huggingface ScienceCast What is ScienceCast? https://sciencecast.org/welcome Demos Recommenders and Search Tools Influence Flower What are Influence Flowers? https://influencemap.cmlab.dev/ CORE Recommender What is CORE? https://core.ac.uk/services/recommender IArxiv Recommender What is IArxiv? https://iarxiv.org/about arXivLabs: experimental projects with community collaborators 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 https://info.arxiv.org/labs/index.html .