Lopez-Paz, David · Nov 6, 2026
David Lopez-Paz argues truth should be judged by invariance
Markdown · JSON · Original source
David Lopez-Paz argues that likelihood—the probability of an event under a model—should not decide whether a statement is true. Unlikely events happen, he notes, while frequently repeated statements can be false. He proposes judging truth by whether a statement remains invariant across relevant transformations of the environment, rather than by its probability under a single distribution. The paper connects this view to generalization and invariant learning, and sketches how language models might test invariance at inference . It also discusses implications for . The argument shifts attention from how likely a model finds a statement to whether it holds across relevant changes in context. The abstract does not establish how those transformations should be selected or whether this approach improves truth judgments in practice.
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
cs.AI · Article
Source details
Original source date: Nov 6, 2026 (2026-11-06).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-09T07:16:33.158Z. First observed: 2026-11-09T05:07:22.903Z. Retrieved: 2026-11-09T06:46:34.569Z.
Referenced by
Recorded citations and public file attachments. A reference alone does not establish agreement or influence.
No public references recorded here yet.
Limits
The supplied evidence is the abstract, not the paper's full argument or details of its proposed tests. The abstract does not specify which environmental transformations count as relevant or report results showing that inference- invariance tests improve truth judgments.
Open question
Which transformations does Lopez-Paz consider relevant, and how could invariance testing at inference show whether a statement remains true across them?
Source-stated
Lopez-Paz argues that likelihood is neither necessary nor sufficient for truth.
likelihood is neither necessary nor sufficient for truth.
Source-stated
He argues that truth should be judged by invariance across relevant environmental transformations, not probability under a single distribution.
the truth of a statement should be judged not by its probability under a single distribution, but by its invariance across relevant transformations of the environment.
Source-stated
The paper connects its proposal to out-of-distribution generalization and invariant learning, and sketches inference-time tests with language models.
We connect this view to machine learning literature on out-of-distribution generalization and invariant learning, sketch how invariance could be tested at inference time with language models
Source-stated
The paper discusses implications for continual learning.
and discuss implications for continual learning.
Source and contributor context
Lopez-Paz, David · 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.
Distribution shift
A change between the data a model was trained on and the data it meets in use, which can lower its accuracy.
Continual learning
A model improving from new experience after deployment, the way a new employee learns on the job, without full retraining.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
TIME
TIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.
Editorial provenance
Version 1. Publication ID: 563c22ec-b54f-4c9c-88d1-6236e93a9a5d.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-09T07:16:33.158Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: useful_perspective.
Observation: e4cd6559d1d2b1e87bf2c4918453513eeb5a6b9b9d6d22cfcf3d010165b6f028. Source text hash: 51daf788cef70087ac8749ec1152c0d3700f0166e89b9256d1bfb1c8260bd86c.
Revision and source-check history
Version 1 · 2026-11-09T07:16:33.158Z
David Lopez-Paz argues truth should be judged by invariance
David Lopez-Paz argues that likelihood—the probability of an event under a model—should not decide whether a statement is true. Unlikely events happen, he notes, while frequently repeated statements can be false. He proposes judging truth by whether a statement remains invariant across relevant transformations of the environment, rather than by its probability under a single distribution. The paper connects this view to generalization and invariant learning, and sketches how language models might test invariance at inference . It also discusses implications for . The argument shifts attention from how likely a model finds a statement to whether it holds across relevant changes in context. The abstract does not establish how those transformations should be selected or whether this approach improves truth judgments in practice.
Original source
cs.AI · Article
Source details
Original source date: Nov 6, 2026 (2026-11-06).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-09T07:16:33.158Z. First observed: 2026-11-09T05:07:22.903Z. Retrieved: 2026-11-09T06:46:34.569Z.
Limits
The supplied evidence is the abstract, not the paper's full argument or details of its proposed tests. The abstract does not specify which environmental transformations count as relevant or report results showing that inference- invariance tests improve truth judgments.
Open question
Which transformations does Lopez-Paz consider relevant, and how could invariance testing at inference show whether a statement remains true across them?
Source-stated
Lopez-Paz argues that likelihood is neither necessary nor sufficient for truth.
likelihood is neither necessary nor sufficient for truth.
Source-stated
He argues that truth should be judged by invariance across relevant environmental transformations, not probability under a single distribution.
the truth of a statement should be judged not by its probability under a single distribution, but by its invariance across relevant transformations of the environment.
Source-stated
The paper connects its proposal to out-of-distribution generalization and invariant learning, and sketches inference-time tests with language models.
We connect this view to machine learning literature on out-of-distribution generalization and invariant learning, sketch how invariance could be tested at inference time with language models
Source-stated
The paper discusses implications for continual learning.
and discuss implications for continual learning.
Source and contributor context
Lopez-Paz, David · 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.
Distribution shift
A change between the data a model was trained on and the data it meets in use, which can lower its accuracy.
Continual learning
A model improving from new experience after deployment, the way a new employee learns on the job, without full retraining.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
TIME
TIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.
Editorial provenance
Version 1. Publication ID: 563c22ec-b54f-4c9c-88d1-6236e93a9a5d.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-09T07:16:33.158Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: useful_perspective.
Observation: e4cd6559d1d2b1e87bf2c4918453513eeb5a6b9b9d6d22cfcf3d010165b6f028. Source text hash: 51daf788cef70087ac8749ec1152c0d3700f0166e89b9256d1bfb1c8260bd86c.
uncertain · 2026-11-09T07:16:33.158Z · anthropic/claude-opus-5.5
Publication: 563c22ec-b54f-4c9c-88d1-6236e93a9a5d
attribution — pass: The arXiv page lists David Lopez-Paz as the sole author. Attributing the paper's "we argue" thesis to him is accurate, and the reading presents it as his argument rather than as established fact.
Authors:David Lopez-Paz
summary — uncertain: The reading leaves out the abstract's opening premise: maximum likelihood is the dominant training principle for large language models. Without it, a general reader cannot see why the argument matters for AI. It also drops the philosophical grounding (Nozick, Wigner, Poincaré, Cassirer). Several central terms are unexplained: out-of-distribution generalization, invariant learning, inference time and continual learning. They need tooltips. Paraphrasing 'environment' as 'context' blurs the source's term. Smallest repair: add the sentence about maximum likelihood being the dominant LLM training principle, and keep 'environment'.
Maximum likelihood is the dominant training principle for large language models, yet likelihood is neither necessary nor sufficient for truth.
date — pass: originalPublished is 2026-11-06, which matches the v1 submission date on arXiv. The feed date of 2026-11-09 is later and correctly not used.
[Submitted on 6 Nov 2026]
support — pass: Each claim matches the abstract verbatim. The limitations correctly note that only the abstract was supplied, with no specified transformations and no reported results. The question follows from those gaps and does not assume benefits.
sketch how invariance could be tested at inference time with language models, and discuss implications for continual learning.
media — pass: The selected media is the verified direct link to the arXiv abstract page, which is the source itself.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.
A change between the data a model was trained on and the data it meets in use, which can lower its accuracy.
Definition referenceTIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.
Explore TIMEA model improving from new experience after deployment, the way a new employee learns on the job, without full retraining.
Definition referencearXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivarXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivTIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.
Explore TIMETIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.
Explore TIMEA change between the data a model was trained on and the data it meets in use, which can lower its accuracy.
Definition referenceTIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.
Explore TIMEA model improving from new experience after deployment, the way a new employee learns on the job, without full retraining.
Definition referencearXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivarXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivTIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.
Explore TIMETIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.
Explore TIME