Riccio, Donato, Maturo, Fabrizio, Brezeanu, Iulia · Nov 19, 2026
Authors report food-weight estimates compress across open-weight LLMs
Markdown · JSON · Original source
Authors Donato Riccio, Fabrizio Maturo and Iulia Brezeanu report a study of whether (LLMs) scale food-energy estimates with stated food weight. Across 168 foods and nine stated weights, all seven models compressed their estimates: estimates rose less than proportionally as weight increased. Four models showed severe compression, overestimating a quarter of the default amount and underestimating four times that amount. Supplying energy density—the energy per unit of food weight—restored near-proportional scaling in three of those four models; calculating energy density times weight in code reduced typical error at the tested extremes. The authors report that compression persisted at full precision and increased in five models under a plain question. Their central warning is that accuracy at usual portions can conceal errors when portions differ: the models can hold the energy densities needed for proportional scaling yet answer near a typical serving. They recommend that food-logging tools calculate energy from weighed entries using density times weight, and that evaluations test fixed foods across stated weights.
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 19, 2026 (2026-11-19).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-20T07:31:31.187Z. First observed: 2026-11-20T05:08:16.208Z. Retrieved: 2026-11-20T07:15:58.677Z.
Referenced by
Recorded citations in public readings. A reference alone does not establish agreement or influence.
No public references recorded here yet.
Limits
The reported evaluation covers seven and 168 randomly drawn foods from the USDA Food and Nutrient Database for Dietary Studies; the abstract does not report results for other models or foods. The abstract describes estimates and evaluation results, not evidence about the performance of food-logging tools in use.
Open question
Would the reported compression persist in and foods beyond those included in this study?
Source-stated
The authors tested seven open-weight LLMs on 168 randomly drawn foods at nine stated weights, from 0.25 to 4 times the default amount.
seven open-weight LLMs estimated the energy of 168 randomly drawn foods from the USDA Food and Nutrient Database for Dietary Studies at nine stated weights, from 0.25 to 4 times the default amount
Source-stated
All seven models compressed their estimates, with elasticity estimates below 1; four showed severe compression.
All seven models compressed their estimates, with elasticities from 0.548 to 0.979 (every 95% CI below 1); four compressed severely (0.548 to 0.709).
Source-stated
Supplying energy density restored near-proportional scaling in three of the four severely compressing models, and the two-step estimate reduced typical error at the tested extremes.
Supplying the density restored near-proportional scaling in three of the four, and the two-step estimate cut the typical error factor at these extremes from 1.47-2.43 to 1.20-1.38.
Source-stated
The authors warn that accuracy at usual portions does not reveal compression and recommend calculating weighed-entry energy as density times weight and testing fixed foods across stated weights.
Accuracy at usual portions does not reveal it. Food-logging tools should compute the energy of weighed entries as density times weight, and evaluations should test fixed foods across stated weights.
Source and contributor context
Riccio, Donato, Maturo, Fabrizio, Brezeanu, Iulia · 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.
Open weights
A model whose trained numbers are published so others can download, run and inspect it; the license and training data may still be restricted.
Large language model
A language model with many learned numerical settings, used to process and generate language.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Editorial provenance
Version 1. Publication ID: e79f0ef6-5266-44c8-8061-ea917b4c7fa8.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-20T07:31:31.187Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: acc79ccf1f5a278c0afe05ed8d0efd95bc62501b196a03a910757ce4fceb0c42. Source text hash: d5f9727839d288203e0525d023108d76cf14c0d3eee975fe7cdcfda6c1352051.
Revision and source-check history
Version 1 · 2026-11-20T07:31:31.187Z
Authors report food-weight estimates compress across open-weight LLMs
Authors Donato Riccio, Fabrizio Maturo and Iulia Brezeanu report a study of whether (LLMs) scale food-energy estimates with stated food weight. Across 168 foods and nine stated weights, all seven models compressed their estimates: estimates rose less than proportionally as weight increased. Four models showed severe compression, overestimating a quarter of the default amount and underestimating four times that amount. Supplying energy density—the energy per unit of food weight—restored near-proportional scaling in three of those four models; calculating energy density times weight in code reduced typical error at the tested extremes. The authors report that compression persisted at full precision and increased in five models under a plain question. Their central warning is that accuracy at usual portions can conceal errors when portions differ: the models can hold the energy densities needed for proportional scaling yet answer near a typical serving. They recommend that food-logging tools calculate energy from weighed entries using density times weight, and that evaluations test fixed foods across stated weights.
Original source
cs.AI · Article
Source details
Original source date: Nov 19, 2026 (2026-11-19).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-20T07:31:31.187Z. First observed: 2026-11-20T05:08:16.208Z. Retrieved: 2026-11-20T07:15:58.677Z.
Limits
The reported evaluation covers seven and 168 randomly drawn foods from the USDA Food and Nutrient Database for Dietary Studies; the abstract does not report results for other models or foods. The abstract describes estimates and evaluation results, not evidence about the performance of food-logging tools in use.
Open question
Would the reported compression persist in and foods beyond those included in this study?
Source-stated
The authors tested seven open-weight LLMs on 168 randomly drawn foods at nine stated weights, from 0.25 to 4 times the default amount.
seven open-weight LLMs estimated the energy of 168 randomly drawn foods from the USDA Food and Nutrient Database for Dietary Studies at nine stated weights, from 0.25 to 4 times the default amount
Source-stated
All seven models compressed their estimates, with elasticity estimates below 1; four showed severe compression.
All seven models compressed their estimates, with elasticities from 0.548 to 0.979 (every 95% CI below 1); four compressed severely (0.548 to 0.709).
Source-stated
Supplying energy density restored near-proportional scaling in three of the four severely compressing models, and the two-step estimate reduced typical error at the tested extremes.
Supplying the density restored near-proportional scaling in three of the four, and the two-step estimate cut the typical error factor at these extremes from 1.47-2.43 to 1.20-1.38.
Source-stated
The authors warn that accuracy at usual portions does not reveal compression and recommend calculating weighed-entry energy as density times weight and testing fixed foods across stated weights.
Accuracy at usual portions does not reveal it. Food-logging tools should compute the energy of weighed entries as density times weight, and evaluations should test fixed foods across stated weights.
Source and contributor context
Riccio, Donato, Maturo, Fabrizio, Brezeanu, Iulia · 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.
Open weights
A model whose trained numbers are published so others can download, run and inspect it; the license and training data may still be restricted.
Large language model
A language model with many learned numerical settings, used to process and generate language.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Editorial provenance
Version 1. Publication ID: e79f0ef6-5266-44c8-8061-ea917b4c7fa8.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-20T07:31:31.187Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: acc79ccf1f5a278c0afe05ed8d0efd95bc62501b196a03a910757ce4fceb0c42. Source text hash: d5f9727839d288203e0525d023108d76cf14c0d3eee975fe7cdcfda6c1352051.
uncertain · 2026-11-20T07:31:31.187Z · anthropic/claude-opus-5.5
Publication: e79f0ef6-5266-44c8-8061-ea917b4c7fa8
attribution — pass: The reading correctly credits the study and its findings to Donato Riccio, Fabrizio Maturo and Iulia Brezeanu. It frames the results and recommendations as the authors' reports rather than established fact.
Authors:Donato Riccio, Fabrizio Maturo, Iulia Brezeanu
summary — uncertain: The body is faithful and readable. Compression and energy density are explained in plain terms. The title is the problem: 'food-weight estimates compress' says the models estimate weight. In fact they estimate energy at an exactly stated weight. Smallest repair: 'Authors report open-weight LLMs compress food-energy estimates across stated weights.' Minor point: 'can hold' slightly softens the source's 'The models hold the energy densities'. 'Full precision' is left unexplained, but it is not central to the finding.
We tested whether energy estimates of open-weight large language models (LLMs) scale proportionally with an exactly stated food weight
date — pass: The originalPublished date of 2026-11-19 matches the v1 submission date shown on the arXiv page. The feed date of 2026-11-20 is later and is correctly not used.
[Submitted on 19 Nov 2026]
support — pass: Each claim matches its abstract excerpt: seven models, 168 foods and nine weights; all elasticities below 1, with four severe; the density and two-step error reductions; and the recommendations. The limitations accurately note the abstract's scope. The question about generalizing beyond the tested models and foods follows from the evidence and is not leading.
These four overestimated a quarter of the default amount by 47% to 130% and underestimated four times it by 20% to 34%.
media — pass: The selected media is the verified arXiv abstract page itself. It is the direct original source and the most relevant link available.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.
A model whose trained numbers are published so others can download, run and inspect it; the license and training data may still be restricted.
Definition referenceA language model with many learned numerical settings, used to process and generate language.
Definition referencearXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivA model whose trained numbers are published so others can download, run and inspect it; the license and training data may still be restricted.
Definition referenceA language model with many learned numerical settings, used to process and generate language.
Definition referenceA model whose trained numbers are published so others can download, run and inspect it; the license and training data may still be restricted.
Definition referenceA language model with many learned numerical settings, used to process and generate language.
Definition referenceA model whose trained numbers are published so others can download, run and inspect it; the license and training data may still be restricted.
Definition referenceA language model with many learned numerical settings, used to process and generate language.
Definition referencearXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivA model whose trained numbers are published so others can download, run and inspect it; the license and training data may still be restricted.
Definition referenceA language model with many learned numerical settings, used to process and generate language.
Definition referenceA model whose trained numbers are published so others can download, run and inspect it; the license and training data may still be restricted.
Definition referenceA language model with many learned numerical settings, used to process and generate language.
Definition reference