Skip to content
AI News HubLIVE
Original source3 min read

Linear Probes to Bayesian Manifolds: Geometry of Arithmetic in Large Language Models

Summary

A new paper proposes a four-stage Bayesian pipeline to separately test decodability, shape, ownership, and causal use, offering a stricter evaluation of whether geometric structures found by linear probes actually participate in model computation. Applied to two-digit addition and multiplication in GPT-J 6B, Llama 3.1 8B, and Pythia 6.9B, results show nearly all concepts are decodable, but strongly supported shapes are ribbons or tori, 81% of shapes are inherited, and intervening on shapes retains about 61% of the effect. The authors recommend reporting probe results as verdicts on all four claims rather than a single accuracy.

SourcearXiv Machine LearningAuthor: Anshul Kumar, Neha Prasad, Manoj Kumar, Barnab\'as P\'oczos, Deeksha Varshney
Linear Probes to Bayesian Manifolds: Geometry of Arithmetic in Large Language Models
Report an error

The correction channel is not available yet. You can copy the article reference below for later.

Correction instructions
Read article

[Submitted on 8 Nov 2026]

Title:Linear Probes to Bayesian Manifolds: Geometry of Arithmetic in Large Language Models

View a PDF of the paper titled Linear Probes to Bayesian Manifolds: Geometry of Arithmetic in Large Language Models, by Anshul Kumar and 4 other authors

View PDF HTML (experimental)

Abstract:Linear probes are a standard method to study geometric structures, such as helices and circles, from the activations of large language models (LLMs), and such structure is often taken as evidence that the model computes with it. Probe accuracy alone cannot establish this, because it conflates four separate claims. These are (1) that the concept is decodable from the activations, (2) that they form a specific shape, (3) that the shape belongs to the probed concept rather than to a related concept the model also represents, and (4) that the output of the model depends on the shape. We call these claims decodability, shape, ownership, and causal use, and we test them separately with a four-stage Bayesian pipeline. Stage 2 compares candidate shapes and selects the best one only when its evidence passes a predefined threshold. Stage 3 removes a set of related concepts, specified in advance, and tests whether the shape is owned or inherited. Stage 4 intervenes on the shape to test whether the output of the model depends on it. We apply the pipeline to two-digit addition and multiplication in GPT-J 6B, Llama 3.1 8B, and Pythia 6.9B. Nearly every concept is decodable, but the decoded directions overlap substantially across concepts. Every strongly supported comparison selects a ribbon, which is a helix whose radius changes with the value, or a torus. Of 1,330 cells with an ownership verdict, 19% of shapes are owned and 81% are inherited. Owned and inherited shapes move the output about equally often. Intervening on the shape retains approximately 61% of the effect of intervening on the full probe while using at most five dimensions, against a median of nine for the probe. Our results indicate that a probe verdict alone overstates what a model computes with, and we propose that probe results be reported as verdicts on all four claims rather than as a single accuracy.

Comments: 26 pages, 4 figures, 20 tables. Under review at ICLR 2027

Subjects:

Machine Learning (cs.LG); Computation and Language (cs.CL)

Cite as: arXiv:2611.12998 [cs.LG]

(or arXiv:2611.12998v1 [cs.LG] for this version)

https://doi.org/10.48550/arXiv.2611.12998

arXiv-issued DOI via DataCite

Submission history

From: Anshul Kumar [view email] [v1] Sun, 8 Nov 2026 14:13:26 UTC (785 KB)

Full-text links:

Access Paper:

View a PDF of the paper titled Linear Probes to Bayesian Manifolds: Geometry of Arithmetic in Large Language Models, by Anshul Kumar and 4 other authors

View PDF

HTML (experimental)

TeX Source

view license

Additional Features

Audio Summary

Current browse context:

cs.LG

new | recent | 2026-11

Change to browse by:

cs cs.CL

References & Citations

SciX

Google Scholar

Semantic Scholar

Loading...

Data provided by:

Bibliographic Tools

Bibliographic and Citation Tools

Bibliographic Explorer Toggle

Bibliographic Explorer (What is the Explorer?)

Connected Papers Toggle

Connected Papers (What is Connected Papers?)

Litmaps Toggle

Litmaps (What is Litmaps?)

scite.ai Toggle

scite Smart Citations (What are Smart Citations?)

Code, Data, Media

Code, Data and Media Associated with this Article

alphaXiv Toggle

alphaXiv (What is alphaXiv?)

Links to Code Toggle

CatalyzeX Code Finder for Papers (What is CatalyzeX?)

DagsHub Toggle

DagsHub (What is DagsHub?)

GotitPub Toggle

Gotit.pub (What is GotitPub?)

Huggingface Toggle

Hugging Face (What is Huggingface?)

ScienceCast Toggle

ScienceCast (What is ScienceCast?)

Demos

Demos

Replicate Toggle

Replicate (What is Replicate?)

Spaces Toggle

Hugging Face Spaces (What is Spaces?)

Spaces Toggle

TXYZ.AI (What is TXYZ.AI?)

Related Papers

Recommenders and Search Tools

Link to Influence Flower

Influence Flower (What are Influence Flowers?)

Core recommender toggle

CORE Recommender (What is CORE?)

IArxiv recommender toggle

IArxiv Recommender (What is IArxiv?)

Author

Venue

Institution

Topic

About arXivLabs

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.

Which authors of this paper are endorsers? | Disable MathJax (What is MathJax?)

Key points and analysis

Article intelligence

InvestorsAdvanced

Key points

  • Linear probe accuracy conflates four separate claims: decodability, shape, ownership, and causal use.
  • A four-stage Bayesian pipeline tests these claims separately and selects a shape only when evidence passes a threshold.
  • All strongly supported comparisons select a ribbon (a helix with radius varying by value) or a torus, and 81% of shapes are inherited.
  • Intervening on the shape retains ~61% of the full probe effect using at most five dimensions.

Highlights and analysis are generated automatically and may contain errors. Check the original source.