# Paper argues SAE features predict errors but do not explain them [Canonical reading](https://escapemeridian.com/readings/paper-argues-sae-features-predict-errors-but-do-not-explain-them-dcc9c40b) Authors Aria Nourbakhsh, Adelaide Danilov, Oleksandr Marchenko Breneur and Salima Lamsiyah distinguish predicting a model’s errors from identifying what causes them. They study sparse autoencoders (SAEs)—models used to identify features in a system’s internal activity—in multiple-choice question answering, measuring answer preference with the logit margin between the correct answer and competing choices. Across eight datasets, the authors report that SAE activations carry signals about wrong answers and uncertainty. After filtering features linked to prompt and answer-format artifacts, selected features changed answer margins more than randomly selected features and produced a higher wrong-to-correct rate in interventions. But answer flips were rare, and the effects did not have stable directions when prompt format or answer order changed without changing the question’s meaning. The distinction matters for interpreting model audits: a feature that predicts an error, or shifts which answer the model favors under an intervention, need not reveal the mechanism that produced the error. The authors conclude that their identified features reflect context-dependent competition among answers, not the mechanism of incorrect answers. What would be needed to identify features whose effects remain stable across meaning-preserving changes in prompt format and answer order? Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims. ## Original source arXiv cs.AI · Article Original source date: Oct 27, 2026 (2026-10-27). Date basis: Date basis not provided. Added to Escape Meridian: 2026-11-10T19:34:42.012Z. First observed: 2026-10-30T06:39:53.092Z. Retrieved: 2026-10-30T18:03:43.965Z. - [From Error Prediction to Causal Influence: Auditing SAE Features in Multiple-Choice QA](https://arxiv.org/abs/2610.38067) ## Limits The evidence provided is the paper abstract; it does not include the paper's methods or detailed results. The reported findings concern Gemma 3 1B and 4B models, Gemma Scope SAEs, and multiple-choice question answering; the abstract does not establish whether they generalize beyond these settings. ## Open question What would be needed to identify features whose effects remain stable across meaning-preserving changes in prompt format and answer order? ## Source-stated The authors distinguish features that predict model errors from features that causally influence answers, and study this distinction in multiple-choice question answering. > association alone does not establish causal influence. We study this distinction in multiple-choice question answering - [Supporting source](https://arxiv.org/abs/2610.38067) ## Source-stated After filtering features associated with artifacts, selected features used in causal interventions shifted answer margins more than randomly selected features and yielded a higher wrong-to-correct rate. > selected features for causal intervention at question and option spans produce larger absolute margin shifts and a higher wrong-to-correct rate than randomly selected features - [Supporting source](https://arxiv.org/abs/2610.38067) ## Source-stated Answer flips were rare, and effects did not have stable directions across meaning-preserving changes to prompt format and answer order. > answer flips are rare, and we do not observe stable effect directions across meaning-preserving changes in prompt format and answer order - [Supporting source](https://arxiv.org/abs/2610.38067) ## Source-stated The authors conclude that the identified features reflect context-dependent answer competition, not the mechanism of incorrect answers. > the identified features reflect context-dependent answer competition rather than the mechanism of incorrect answer errors - [Supporting source](https://arxiv.org/abs/2610.38067) ## Source and contributor context Nourbakhsh, Aria, Danilov, Adelaide, Breneur, Oleksandr Marchenko, Lamsiyah, Salima · Article author These are credits stated by the source; they do not establish a profile match or authorship of every passage. ## Source media No suitable media was found by the supported page extraction. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## arXiv arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed. - [Explore this context](/entities/arxiv) - [Reference](https://arxiv.org/) - [Reference](https://info.arxiv.org/help/api/index.html) ## Editorial provenance Version 1. Publication ID: 8a015b79-78ca-4c76-80e0-b89de9a1125d. Prepared by server-excerpt-repair/stream-excerpt-repair-v1 (original draft: gpt-6-luna); grounding review by gpt-6-luna. Source check: passed by anthropic/claude-opus-5.5 on 2026-11-10T19:34:42.012Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: material\_development. Observation: 029d5b8b92820251e467414a7833ab57034fa6e07b35e44840279a39bcd1a9bd. Source text hash: d3df203fa8b5acdd80c36fc24ce738010a9b0562d21fc05a07caae841d63e6b3. ## Revision and source-check history ### Version 1 — 2026-11-10T19:34:42.012Z Authors Aria Nourbakhsh, Adelaide Danilov, Oleksandr Marchenko Breneur and Salima Lamsiyah distinguish predicting a model’s errors from identifying what causes them. They study sparse autoencoders (SAEs)—models used to identify features in a system’s internal activity—in multiple-choice question answering, measuring answer preference with the logit margin between the correct answer and competing choices. Across eight datasets, the authors report that SAE activations carry signals about wrong answers and uncertainty. After filtering features linked to prompt and answer-format artifacts, selected features changed answer margins more than randomly selected features and produced a higher wrong-to-correct rate in interventions. But answer flips were rare, and the effects did not have stable directions when prompt format or answer order changed without changing the question’s meaning. The distinction matters for interpreting model audits: a feature that predicts an error, or shifts which answer the model favors under an intervention, need not reveal the mechanism that produced the error. The authors conclude that their identified features reflect context-dependent competition among answers, not the mechanism of incorrect answers. What would be needed to identify features whose effects remain stable across meaning-preserving changes in prompt format and answer order? ## Original source arXiv cs.AI · Article Original source date: Oct 27, 2026 (2026-10-27). Date basis: Date basis not provided. Added to Escape Meridian: 2026-11-10T19:34:42.012Z. First observed: 2026-10-30T06:39:53.092Z. Retrieved: 2026-10-30T18:03:43.965Z. - [From Error Prediction to Causal Influence: Auditing SAE Features in Multiple-Choice QA](https://arxiv.org/abs/2610.38067) ## Limits The evidence provided is the paper abstract; it does not include the paper's methods or detailed results. The reported findings concern Gemma 3 1B and 4B models, Gemma Scope SAEs, and multiple-choice question answering; the abstract does not establish whether they generalize beyond these settings. ## Open question What would be needed to identify features whose effects remain stable across meaning-preserving changes in prompt format and answer order? ## Source-stated The authors distinguish features that predict model errors from features that causally influence answers, and study this distinction in multiple-choice question answering. > association alone does not establish causal influence. We study this distinction in multiple-choice question answering - [Supporting source](https://arxiv.org/abs/2610.38067) ## Source-stated After filtering features associated with artifacts, selected features used in causal interventions shifted answer margins more than randomly selected features and yielded a higher wrong-to-correct rate. > selected features for causal intervention at question and option spans produce larger absolute margin shifts and a higher wrong-to-correct rate than randomly selected features - [Supporting source](https://arxiv.org/abs/2610.38067) ## Source-stated Answer flips were rare, and effects did not have stable directions across meaning-preserving changes to prompt format and answer order. > answer flips are rare, and we do not observe stable effect directions across meaning-preserving changes in prompt format and answer order - [Supporting source](https://arxiv.org/abs/2610.38067) ## Source-stated The authors conclude that the identified features reflect context-dependent answer competition, not the mechanism of incorrect answers. > the identified features reflect context-dependent answer competition rather than the mechanism of incorrect answer errors - [Supporting source](https://arxiv.org/abs/2610.38067) ## Source and contributor context Nourbakhsh, Aria, Danilov, Adelaide, Breneur, Oleksandr Marchenko, Lamsiyah, Salima · Article author These are credits stated by the source; they do not establish a profile match or authorship of every passage. ## Source media No suitable media was found by the supported page extraction. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## arXiv arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed. - [Explore this context](/entities/arxiv) - [Reference](https://arxiv.org/) - [Reference](https://info.arxiv.org/help/api/index.html) ## Editorial provenance Version 1. Publication ID: 8a015b79-78ca-4c76-80e0-b89de9a1125d. Prepared by server-excerpt-repair/stream-excerpt-repair-v1 (original draft: gpt-6-luna); grounding review by gpt-6-luna. Source check: passed by anthropic/claude-opus-5.5 on 2026-11-10T19:34:42.012Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: material\_development. Observation: 029d5b8b92820251e467414a7833ab57034fa6e07b35e44840279a39bcd1a9bd. Source text hash: d3df203fa8b5acdd80c36fc24ce738010a9b0562d21fc05a07caae841d63e6b3. ### passed — 2026-11-10T19:34:42.012Z — anthropic/claude-opus-5.5 Publication: 8a015b79-78ca-4c76-80e0-b89de9a1125d attribution — pass: The draft names the four listed authors and frames the findings and conclusion as theirs. It presents the claims as reported results, not as established fact or Escape Meridian's opinion. > Authors:Aria Nourbakhsh, Adelaide Danilov, Oleksandr Marchenko Breneur, Salima Lamsiyah summary — pass: The draft leads with the authors' central distinction between prediction and causal influence. It briefly explains SAEs and gives the logit margin's role as a measure of answer preference. It keeps the mixed results and the authors' firm conclusion. It omits the HalluSAE baseline, but the main argument stays intact. The models studied (Gemma 3 1B/4B) appear only in the limitations. Optional repairs: name the models in the body, and keep 'absolute' before 'margin shifts' to reflect the unstable direction. > Therefore, the identified features reflect context-dependent answer competition rather than the mechanism of incorrect answer errors. date — pass: originalPublished is 2026-10-27, which matches the arXiv v1 submission date. The feed date of October 30 is later and was correctly not used. > \[Submitted on 27 Oct 2026\] support — pass: Each claim matches the abstract verbatim in substance. The significance paragraph says predictive or margin-shifting features need not reveal error mechanisms. That restates the authors' own distinction and conclusion rather than adding an unsupported inference. The question follows from the reported instability across prompt format and answer order. The limitations accurately note that the evidence is abstract-only and that the scope is narrow. > we do not observe stable effect directions across meaning-preserving changes in prompt format and answer order media — pass: The selected media is the verified arXiv abstract page for this paper. It is the direct original source. ## Recorded references Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter. No public references recorded here yet. [Reference data (JSON)](/readings/dcc9c40b8e0526759bf170c2da1227b63ccd821ca02e096a69f63ed07339d8ac/references.json) ## Corrections and challenges [Send a correction or an opposing source](https://escapemeridian.com/corrections/) [JSON record](https://escapemeridian.com/readings/paper-argues-sae-features-predict-errors-but-do-not-explain-them-dcc9c40b.json)